mdbx: исправление опечатки MDBX_ENOMEM.

Вместо `MDBX_ENOMEM` был использован идентификатор `ENOMEM`,
что могло ломать сборку на не-POSIX/Windows платформах,
в зависимости от конфигурации и/или версии SDK.

.circleci/config.yml

@@ -1,20 +0,0 @@
-version: 2
-jobs:
-  build:
-    docker:
-      - image: circleci/buildpack-deps:artful
-    environment:
-      - TESTDB: /tmp/test.db
-      - TESTLOG: /tmp/test.log
-    steps:
-      - checkout
-      - run: make all
-      - run: ulimit -c unlimited && make check
-      - run:
-          command: |
-            mkdir -p /tmp/artifacts
-            mv -t /tmp/artifacts $TESTLOG $TESTDB core.*
-          when: on_fail
-      - store_artifacts:
-          path: /tmp/artifacts
-          destination: test-artifacts

.clang-format

@@ -1,3 +1,3 @@
 BasedOnStyle: LLVM
-Standard: Cpp11
-ReflowComments: true
+Standard: c++20
+ColumnLimit: 120

.cmake-format.yaml

@@ -0,0 +1,3 @@
+format:
+  line_width: 120
+  tab_size: 2

.gitignore

@@ -1,38 +1,69 @@
+*[~#]
+@*
 *.[ao]
+*.autosave
 *.bak
+build.ninja
+cmake-build-*
+CMakeCache.txt
+CMakeFiles/
+cmake_install.cmake
+CMakeLists.txt.user
+core
+CTestTestfile.cmake
+DartConfiguration.tcl
+dist/
+*.dll
+docs/Doxyfile
+docs/html/
+*.dSYM
+*.dylib
+*.err
 *.exe
 *.gcda
 *.gcno
 *.gcov
-*.lo
-*.orig
-*.rej
-*.so
-*[~#]
 .idea
-.le.ini
-.vs/
-Win32/
-build-*
-cmake-build-*
-core
-example
 libmdbx.creator.user
-mdbx-dll.VC.VC.opendb
-mdbx-dll.VC.db
-mdbx-dll.vcxproj.filters
+*.lo
 mdbx_chk
 mdbx_copy
+mdbx_drop
 mdbx_dump
+mdbx_example
 mdbx_load
 mdbx_stat
 mdbx_test
+.ninja_deps
+.ninja_log
+*.orig
+*.rej
+*.so
+src/config.h
+src/version.c
+*.tar*
+test/cmake_install.cmake
+test/CTestTestfile.cmake
+test_extra_crunched_delete
+test_extra_cursor_closing
+test_extra_dbi
+test_extra_doubtless_positioning
+test_extra_dupfix_addodd
+test_extra_dupfix_multiple
+test_extra_early_close_dbi
+test_extra_hex_base64_base58
+test_extra_maindb_ordinal
+test_extra_open
+test_extra_pcrf
+test_extra_upsert_alldups
+Testing/
 test.log
-test/test.vcxproj.user
 test/tmp.db
 test/tmp.db-lck
 tmp.db
 tmp.db-lck
 valgrind.*
-x64/
-x86/
+version.c
+.vs/
+.vscode/
+*.zip

.le.ini

@@ -0,0 +1,40 @@
+tabsize=8
+indentsize=8
+autoindent=1
+bsunindents=1
+insert=1
+inputmode=0
+editmode=1
+makebak=0
+bakpath=
+make=exec make
+shell=exec $SHELL
+run=exec make run
+compile=exec make "$FNAME.o"
+scroll=1
+hscroll=32
+rblock=0
+savepos=1
+savehst=1
+noreg=1
+match_case=1
+linelen=72
+leftmrg=0
+flnmarg=0
+leftadj=1
+rightadj=0
+helpcmd=exec /usr/share/le/help
+usecolor=1
+usetabs=1
+scrollbar=0
+statusline=0
+backupext=.~%d~
+backupnum=9
+preferpagetop=1
+wordwrap=0
+syntaxhl=1
+undo_enable=1
+undo_min_levels=4
+undo_max_memory=128
+undo_glue=1
+usemouse=0

.travis.yml

@@ -1,30 +0,0 @@
-language: c
-sudo: required
-dist: trusty
-
-compiler:
-- gcc
-- clang
-
-os:
-- linux
-
-script: if [ "${COVERITY_SCAN_BRANCH}" != 1 ]; then make all check; fi
-
-env:
-  global:
-   - secure: "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"
-
-before_install:
-  - echo -n | openssl s_client -connect scan.coverity.com:443 | sed -ne '/-BEGIN CERTIFICATE-/,/-END CERTIFICATE-/p' | sudo tee -a /etc/ssl/certs/ca-
-
-addons:
-  coverity_scan:
-    project:
-      name: "ReOpen/libmdbx"
-      version: 0.1
-      description: "Build submitted via Travis CI"
-    notification_email: leo@yuriev.ru
-    build_command_prepend: "make clean"
-    build_command:   "make all -j 4"
-    branch_pattern: coverity_scan

AUTHORS

@@ -1,31 +0,0 @@
-Contributors
-============
-
-Alexey Naumov <alexey.naumov@gmail.com>
-Chris Mikkelson <cmikk@qwest.net>
-Claude Brisson <claude.brisson@gmail.com>
-David Barbour <dmbarbour@gmail.com>
-David Wilson <dw@botanicus.net>
-dreamsxin <dreamsxin@126.com>
-Hallvard Furuseth <hallvard@openldap.org>, <h.b.furuseth@usit.uio.no>
-Heiko Becker <heirecka@exherbo.org>
-Howard Chu <hyc@openldap.org>, <hyc@symas.com>
-Ignacio Casal Quinteiro <ignacio.casal@nice-software.com>
-James Rouzier <rouzier@gmail.com>
-Jean-Christophe DUBOIS <jcd@tribudubois.net>
-John Hewson <john@jahewson.com>
-Klaus Malorny <klaus.malorny@knipp.de>
-Kurt Zeilenga <kurt.zeilenga@isode.com>
-Leonid Yuriev <leo@yuriev.ru>, <lyuryev@ptsecurity.com>
-Lorenz Bauer <lmb@cloudflare.com>
-Luke Yeager <lyeager@nvidia.com>
-Martin Hedenfalk <martin@bzero.se>
-Ondrej Kuznik <ondrej.kuznik@acision.com>
-Orivej Desh <orivej@gmx.fr>
-Oskari Timperi <oskari.timperi@iki.fi>
-Pavel Medvedev <pmedvedev@gmail.com>
-Philipp Storz <philipp.storz@bareos.com>
-Quanah Gibson-Mount <quanah@openldap.org>
-Salvador Ortiz <sog@msg.com.mx>
-Sebastien Launay <sebastien@slaunay.fr>
-Vladimir Romanov <vromanov@gmail.com>

COPYRIGHT

@@ -1,7 +1,139 @@
-Copyright 2015-2018 Leonid Yuriev <leo@yuriev.ru>.
-Copyright 2011-2015 Howard Chu, Symas Corp.
-Copyright 2015,2016 Peter-Service R&D LLC.
-All rights reserved.
+Copyright (c) 2015-2025 Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru>
+
+Licensed under the Apache License, Version 2.0 (the "License");
+you may not use this file except in compliance with the License.
+You may obtain a copy of the License at
+
+    http://www.apache.org/licenses/LICENSE-2.0
+
+Unless required by applicable law or agreed to in writing, software
+distributed under the License is distributed on an "AS IS" BASIS,
+WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+See the License for the specific language governing permissions and
+limitations under the License.
+
+-----------------------------------------------------------------------
+
+СМЕНА ЛИЦЕНЗИИ (THE LICENSE CHANGE)
+
+OpenLDAP Public License → Apache 2.0
+
+Briefly:
+  Historically, in 2015 an early MDBX source code was derived from the
+  "LMDB engine" created by Howard Chu <hyc@symas.com> in 2011-2015,
+  which based on btree.c written by Martin Hedenfalk <martin@bzero.se>.
+
+  By 2024, MDBX source code has actually been rewritten and has so
+  little in common with the original LMDB that I thought it admissible
+  to change the license. Below are more detailed explanations.
+
+Кратко:
+
+  Исторически в 2015 году ранний исходный код MDBX был заимствован из
+  «LMDB engine», созданной Howard Chu <hyc@symas.com> в 2011-2015,
+  на основе btree.c, ранее созданного Martin Hedenfalk <martin@bzero.se>.
+
+  К 2024 году исходный код MDBX фактически переписан и имеет настолько
+  мало общего с первоначальным заимствованием из LMDB, что я счел
+  уместным сменить лицензию. Ниже более подробные пояснения.
+
+---
+
+Первоисточник текста формулирован на Русском языке, который является
+родным для автора. Предполагается что все заинтересованные могут легко
+воспользоваться машинным переводом, который при всех недостатках сможет
+донести суть, намерения и местами даже передать тональность.
+
+The original source of this text is in Russian, which is the author's
+native language. It is assumed that all concerned can easily use machine
+translation, which, with all the disadvantages, will be able to convey
+the essence, intentions and, in some places, even convey the tonality of
+a wording.
+
+1. Причины
+
+1.1. Лицензия Apache-2.0 является одной из самых популярных, так как
+содержит ряд уточнений, проясняющих и упрощающих использование исходного
+кода в производных работах и больших проектах. Эти особенности лицензии
+Apache-2.0 я нахожу достаточно ценными и удобными. Соответственно,
+переход на лицензию Apache-2.0 полезным в целом.
+
+1.2. Проект OpenLDAP имеет определенную известность, в том числе, к
+сожалению, среди специалистов славится кране плохим качеством кода и
+сбоями при отходе от простых/базовых сценариев использования. Поэтому
+использование лицензии OpenLDAP, в глазах части аудитории, бросает тень
+на качества кода libmdbx, несмотря на то, что исходный код библиотеки
+переписан, в том числе, с целью повышения качества, надежности,
+стабильности и пригодности к тестированию.
+
+Отмечу, что здесь не место для обсуждения объективности подобных мнений
+и причин, равно как и не место для оценки компетентности специалистов
+высказывающих такие суждения. Однако, здесь необходимо озвучить сам факт
+наличия такой негативной коннотации качества кода при упоминании
+OpenLDAP, совершенно без намерения как-либо задеть или обидеть
+контрибьюторов OpenLDAP.
+
+1.3. С точки зрения исходного кода, к настоящему времени libmdbx стала
+совсем другим продуктом, о котором сейчас правильнее сказать что
+разработка вдохновлена LMDB, нежели основывается на заимствовании кода.
+Смена лицензии на переписанный код подчеркивает, что это действительно
+новый исходный код.
+
+2. Легитимность
+
+2.1. Исходная лицензия OpenLDAP 2.8 и актуальная лицензия Apache 2.0
+совпадают по базовым условиям. При этом лицензия Apache 2.0 уточняет,
+определяет и проясняет многие аспекты. Поэтому смену лицензии я склонен
+трактовать как уточнение, но НЕ как принципиальное изменение, которое
+могло-бы нарушить чьи-либо права.
+
+2.2. С процедурной точки зрения, у меня есть право сменить лицензию на
+новый, написанный мной, исходный код. При этом объективно существует как
+техническая, так и юридическая проблемы отделения «нового кода» от
+«заимствованного», а также выделение/классификация кода, который
+является общественным достоянием и/или общеупотребительным воплощением
+«математических моделей и других публичных знаний».
+
+Основываясь на собственной субъективной оценке кодовой базы, включая
+соотношения «нового», «заимствованного» и «общеупотребительного»
+исходного кода, я считаю что смена лицензии допустима. Одновременно с
+этим, я понимаю и признаю, что можно найти повод, чтобы трактовать
+ситуацию как «стакан наполовину полон/пуст». Поэтому декларирую
+готовность принимать претензии и устранять их путем полного
+переписывания оставшегося исходного кода, который попадает под критерии
+«заимствованного» и кто-то из контрибьюторов которого будет против
+изменения лицензии.
+
+2.3. Вне зависимости от истории происхождения каждой строки исходного
+кода и её буквального авторства, прошу не считать производимую смену
+лицензии, и связанных с этим технических действий, как попытку плагиата,
+присвоения чужого труда, присвоения авторства или принижения вклада
+других авторов/контрибьторов. Безусловно проект MDBX/libmdbx не появился
+бы без LMDB и всех участников проекта LMDB, в особенности Говарда Чу
+(Howard Chu), Холлварда Фурусет (Hallvard Furuseth) и Мартина Хеденфок
+(Martin Hedenfalk). Как-бы исходный код не переписывался он всё равно
+будет основываться на базовых идеях и включать основные концепции LMDB.
+
+3. Последствия и актуальные требования
+
+Всё очень просто. Потребуется обеспечить требования новой лицензии в
+соответствии с 4-м пунктом лицензции Apache 2.0.
+
+В частности, при использовании/распространении libmdbx потребуется
+обеспечить наличие файлов с текстом лицензии и файла NOTICE, а также
+обеспечить пользователям возможность ознакомиться с их содержимым в
+работах/продуктах использующих libmdbx.
+
+-----------------------------------------------------------------------
+
+Далее в справочных целях приведены уведомления об авторских правах из
+первоначально заимствованного кода.
+
+---
+
+Original source code was derived from LMDB in 2015,
+and later evolutionarily rewritten in 2015-2024:
+Copyright (c) 2011-2015 Howard Chu, Symas Corp. All rights reserved.
 
 Redistribution and use in source and binary forms, with or without
 modification, are permitted only as authorized by the OpenLDAP
@@ -11,12 +143,17 @@ A copy of this license is available in the file LICENSE in the
 top-level directory of the distribution or, alternatively, at
 <http://www.OpenLDAP.org/license.html>.
 
-OpenLDAP is a registered trademark of the OpenLDAP Foundation.
+LMDB itself devived code from btree.c written by Martin Hedenfalk:
+Copyright (c) 2009, 2010 Martin Hedenfalk <martin@bzero.se>
 
-Individual files and/or contributed packages may be copyright by
-other parties and/or subject to additional restrictions.
+Permission to use, copy, modify, and distribute this software for any
+purpose with or without fee is hereby granted, provided that the above
+ copyright notice and this permission notice appear in all copies.
 
-This work also contains materials derived from public sources.
-
-Additional information about OpenLDAP can be obtained at
-<http://www.openldap.org/>.
+THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
+WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
+MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
+ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
+WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
+ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
+OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.

GNUmakefile

@@ -0,0 +1,947 @@
+# This makefile is for GNU Make 3.81 or above, and nowadays provided
+# just for compatibility and preservation of traditions.
+#
+# Please use CMake in case of any difficulties or
+# problems with this old-school's magic.
+#
+################################################################################
+#
+# Basic internal definitions. For a customizable variables and options see below.
+#
+$(info // The GNU Make $(MAKE_VERSION))
+SHELL         := $(shell env bash -c 'echo $$BASH')
+MAKE_VERx3    := $(shell printf "%3s%3s%3s" $(subst ., ,$(MAKE_VERSION)))
+make_lt_3_81  := $(shell expr "$(MAKE_VERx3)" "<" "  3 81")
+ifneq ($(make_lt_3_81),0)
+$(error Please use GNU Make 3.81 or above)
+endif
+make_ge_4_1   := $(shell expr "$(MAKE_VERx3)" ">=" "  4  1")
+make_ge_4_4   := $(shell expr "$(MAKE_VERx3)" ">=" "  4  4")
+SRC_PROBE_C   := $(shell [ -f mdbx.c ] && echo mdbx.c || echo src/osal.c)
+SRC_PROBE_CXX := $(shell [ -f mdbx.c++ ] && echo mdbx.c++ || echo src/mdbx.c++)
+UNAME         := $(shell uname -s 2>/dev/null || echo Unknown)
+
+define cxx_filesystem_probe
+  int main(int argc, const char*argv[]) {
+    mdbx::filesystem::path probe(argv[0]);
+    if (argc != 1) throw mdbx::filesystem::filesystem_error(std::string("fake"), std::error_code());
+    return mdbx::filesystem::is_directory(probe.relative_path());
+  }
+endef
+#
+################################################################################
+#
+# Use `make options` to list the available libmdbx build options.
+#
+# Note that the defaults should already be correct for most platforms;
+# you should not need to change any of these. Read their descriptions
+# in README and source code (see src/options.h) if you do.
+#
+
+# install sandbox
+DESTDIR ?=
+INSTALL ?= install
+# install prefixes (inside sandbox)
+prefix  ?= /usr/local
+mandir  ?= $(prefix)/man
+# lib/bin suffix for multiarch/biarch, e.g. '.x86_64'
+suffix  ?=
+
+# toolchain
+CC      ?= gcc
+CXX     ?= g++
+CFLAGS_EXTRA ?=
+LD      ?= ld
+CMAKE	?= cmake
+CMAKE_OPT ?=
+CTEST	?= ctest
+CTEST_OPT ?=
+# target directory for `make dist`
+DIST_DIR ?= dist
+
+# build options
+MDBX_BUILD_OPTIONS   ?=-DNDEBUG=1
+MDBX_BUILD_TIMESTAMP ?=$(if $(SOURCE_DATE_EPOCH),$(SOURCE_DATE_EPOCH),$(shell date +%Y-%m-%dT%H:%M:%S%z))
+MDBX_BUILD_CXX       ?=YES
+MDBX_BUILD_METADATA  ?=
+
+# probe and compose common compiler flags with variable expansion trick (seems this work two times per session for GNU Make 3.81)
+CFLAGS       ?= $(strip $(eval CFLAGS := -std=gnu11 -O2 -g -Wall -Werror -Wextra -Wpedantic -ffunction-sections -fPIC -fvisibility=hidden -pthread -Wno-error=attributes $$(shell for opt in -fno-semantic-interposition -Wno-unused-command-line-argument -Wno-tautological-compare; do [ -z "$$$$($(CC) '-DMDBX_BUILD_FLAGS="probe"' $$$${opt} -c $(SRC_PROBE_C) -o /dev/null >/dev/null 2>&1 || echo failed)" ] && echo "$$$${opt} "; done)$(CFLAGS_EXTRA))$(CFLAGS))
+
+# choosing C++ standard with variable expansion trick (seems this work two times per session for GNU Make 3.81)
+CXXSTD       ?= $(eval CXXSTD := $$(shell for std in gnu++23 c++23 gnu++2b c++2b gnu++20 c++20 gnu++2a c++2a gnu++17 c++17 gnu++1z c++1z gnu++14 c++14 gnu++1y c++1y gnu+11 c++11 gnu++0x c++0x; do $(CXX) -std=$$$${std} -DMDBX_BUILD_CXX=1 -c $(SRC_PROBE_CXX) -o /dev/null 2>probe4std-$$$${std}.err >/dev/null && echo "-std=$$$${std}" && exit; done))$(CXXSTD)
+CXXFLAGS     ?= $(strip $(CXXSTD) $(filter-out -std=gnu11,$(CFLAGS)))
+
+# libraries and options for linking
+EXE_LDFLAGS  ?= -pthread
+ifneq ($(make_ge_4_1),1)
+# don't use variable expansion trick as workaround for bugs of GNU Make before 4.1
+LIBS         ?= $(shell $(uname2libs))
+LDFLAGS      ?= $(shell $(uname2ldflags))
+LIB_STDCXXFS ?= $(shell echo '$(cxx_filesystem_probe)' | cat mdbx.h++ - | sed $$'1s/\xef\xbb\xbf//' | $(CXX) -x c++ $(CXXFLAGS) -Wno-error - -Wl,--allow-multiple-definition -lstdc++fs $(LIBS) $(LDFLAGS) $(EXE_LDFLAGS) -o /dev/null 2>probe4lstdfs.err >/dev/null && echo '-Wl,--allow-multiple-definition -lstdc++fs')
+else
+# using variable expansion trick to avoid repeaded probes
+LIBS         ?= $(eval LIBS := $$(shell $$(uname2libs)))$(LIBS)
+LDFLAGS      ?= $(eval LDFLAGS := $$(shell $$(uname2ldflags)))$(LDFLAGS)
+LIB_STDCXXFS ?= $(eval LIB_STDCXXFS := $$(shell echo '$$(cxx_filesystem_probe)' | cat mdbx.h++ - | sed $$$$'1s/\xef\xbb\xbf//' | $(CXX) -x c++ $(CXXFLAGS) -Wno-error - -Wl,--allow-multiple-definition -lstdc++fs $(LIBS) $(LDFLAGS) $(EXE_LDFLAGS) -o /dev/null 2>probe4lstdfs.err >/dev/null && echo '-Wl,--allow-multiple-definition -lstdc++fs'))$(LIB_STDCXXFS)
+endif
+
+ifneq ($(make_ge_4_4),1)
+.NOTPARALLEL:
+WAIT         =
+else
+WAIT         = .WAIT
+endif
+
+################################################################################
+
+define uname2sosuffix
+  case "$(UNAME)" in
+    Darwin*|Mach*) echo dylib;;
+    CYGWIN*|MINGW*|MSYS*|Windows*) echo dll;;
+    *) echo so;;
+  esac
+endef
+
+define uname2ldflags
+  case "$(UNAME)" in
+    CYGWIN*|MINGW*|MSYS*|Windows*)
+      echo '-Wl,--gc-sections,-O1';
+      ;;
+    *)
+      $(LD) --help 2>/dev/null | grep -q -- --gc-sections && echo '-Wl,--gc-sections,-z,relro,-O1';
+      $(LD) --help 2>/dev/null | grep -q -- -dead_strip && echo '-Wl,-dead_strip';
+      ;;
+  esac
+endef
+
+# TIP: try add the'-Wl, --no-as-needed,-lrt' for ability to built with modern glibc, but then use with the old.
+define uname2libs
+  case "$(UNAME)" in
+    CYGWIN*|MINGW*|MSYS*|Windows*)
+      echo '-lm -lntdll -lwinmm';
+      ;;
+    *SunOS*|*Solaris*)
+      echo '-lm -lkstat -lrt';
+      ;;
+    *Darwin*|OpenBSD*)
+      echo '-lm';
+      ;;
+    *)
+      echo '-lm -lrt';
+      ;;
+  esac
+endef
+
+SO_SUFFIX  := $(shell $(uname2sosuffix))
+HEADERS    := mdbx.h mdbx.h++
+LIBRARIES  := libmdbx.a libmdbx.$(SO_SUFFIX)
+TOOLS      := chk copy drop dump load stat
+MDBX_TOOLS := $(addprefix mdbx_,$(TOOLS))
+MANPAGES   := mdbx_stat.1 mdbx_copy.1 mdbx_dump.1 mdbx_load.1 mdbx_chk.1 mdbx_drop.1
+TIP        := // TIP:
+
+.PHONY: all help options lib libs tools clean install uninstall check_buildflags_tag tools-static
+.PHONY: install-strip install-no-strip strip libmdbx mdbx show-options lib-static lib-shared cmake-build ninja
+
+boolean = $(if $(findstring $(strip $($1)),YES Yes yes y ON On on 1 true True TRUE),1,$(if $(findstring $(strip $($1)),NO No no n OFF Off off 0 false False FALSE),,$(error Wrong value `$($1)` of $1 for YES/NO option)))
+select_by = $(if $(call boolean,$(1)),$(2),$(3))
+
+ifeq ("$(origin V)", "command line")
+  MDBX_BUILD_VERBOSE := $(V)
+endif
+ifndef MDBX_BUILD_VERBOSE
+  MDBX_BUILD_VERBOSE := 0
+endif
+
+ifeq ($(call boolean,MDBX_BUILD_VERBOSE),1)
+  QUIET :=
+  HUSH :=
+  $(info $(TIP) Use `make V=0` for quiet.)
+else
+  QUIET := @
+  HUSH := >/dev/null
+  $(info $(TIP) Use `make V=1` for verbose.)
+endif
+
+ifeq ($(UNAME),Darwin)
+  $(info $(TIP) Use `brew install gnu-sed gnu-tar` and add ones to the beginning of the PATH.)
+endif
+
+all: show-options $(LIBRARIES) $(MDBX_TOOLS)
+
+help:
+	@echo "  make all                 - build libraries and tools"
+	@echo "  make help                - print this help"
+	@echo "  make options             - list build options"
+	@echo "  make lib                 - build libraries, also lib-static and lib-shared"
+	@echo "  make tools               - build the tools"
+	@echo "  make tools-static        - build the tools with statically linking with system libraries and compiler runtime"
+	@echo "  make clean               "
+	@echo "  make install             "
+	@echo "  make uninstall           "
+	@echo "  make cmake-build | ninja - build by CMake & Ninja"
+	@echo ""
+	@echo "  make strip               - strip debug symbols from binaries"
+	@echo "  make install-no-strip    - install explicitly without strip"
+	@echo "  make install-strip       - install explicitly with strip"
+	@echo ""
+	@echo "  make bench               - run ioarena-benchmark"
+	@echo "  make bench-couple        - run ioarena-benchmark for mdbx and lmdb"
+	@echo "  make bench-triplet       - run ioarena-benchmark for mdbx, lmdb, sqlite3"
+	@echo "  make bench-quartet       - run ioarena-benchmark for mdbx, lmdb, rocksdb, wiredtiger"
+	@echo "  make bench-clean         - remove temp database(s) after benchmark"
+#> dist-cutoff-begin
+	@echo ""
+	@echo "  make check               - smoke test with amalgamation and installation checking"
+	@echo "  make smoke               - fast smoke test"
+	@echo "  make smoke-memcheck      - build with Valgrind support and run smoke test under memcheck tool"
+	@echo "  make smoke-fault         - execute transaction owner failure smoke testcase"
+	@echo "  make smoke-singleprocess - execute single-process smoke test"
+	@echo "  make test                - basic test"
+	@echo "  make test-memcheck       - build with Valgrind support and run basic test under memcheck tool"
+	@echo "  make test-long           - execute long test which runs for several weeks, or until interruption"
+	@echo "  make test-asan           - build with AddressSanitizer and run basic test"
+	@echo "  make test-leak           - build with LeakSanitizer and run basic test"
+	@echo "  make test-ubsan          - build with UndefinedBehaviourSanitizer and run basic test"
+	@echo "  make test-singleprocess  - execute single-process basic test (also used by make cross-qemu)"
+	@echo "  make cross-gcc           - check cross-compilation without test execution"
+	@echo "  make cross-qemu          - run cross-compilation and execution basic test with QEMU"
+	@echo "  make gcc-analyzer        - run gcc-analyzer (mostly useless for now)"
+	@echo "  make build-test          - build test executable(s)"
+	@echo ""
+	@echo "  make dist                - build amalgamated source code"
+	@echo "  make doxygen             - build HTML documentation"
+	@echo "  make release-assets      - build release assets"
+	@echo "  make reformat            - reformat source code with clang-format"
+#< dist-cutoff-end
+
+show-options:
+	@echo "  MDBX_BUILD_OPTIONS   = $(MDBX_BUILD_OPTIONS)"
+	@echo "  MDBX_BUILD_CXX       = $(MDBX_BUILD_CXX)"
+	@echo "  MDBX_BUILD_TIMESTAMP = $(MDBX_BUILD_TIMESTAMP)"
+	@echo "  MDBX_BUILD_METADATA  = $(MDBX_BUILD_METADATA)"
+	@echo '$(TIP) Use `make options` to listing available build options.'
+	@echo $(call select_by,MDBX_BUILD_CXX,"  CXX      =`which $(CXX)` | `$(CXX) --version | head -1`","  CC       =`which $(CC)` | `$(CC) --version | head -1`")
+	@echo $(call select_by,MDBX_BUILD_CXX,"  CXXFLAGS =$(CXXFLAGS)","  CFLAGS   =$(CFLAGS)")
+	@echo $(call select_by,MDBX_BUILD_CXX,"  LDFLAGS  =$(LDFLAGS) $(LIB_STDCXXFS) $(LIBS) $(EXE_LDFLAGS)","  LDFLAGS  =$(LDFLAGS) $(LIBS) $(EXE_LDFLAGS)")
+	@echo '$(TIP) Use `make help` to listing available targets.'
+
+options:
+	@echo "  INSTALL      =$(INSTALL)"
+	@echo "  DESTDIR      =$(DESTDIR)"
+	@echo "  prefix       =$(prefix)"
+	@echo "  mandir       =$(mandir)"
+	@echo "  suffix       =$(suffix)"
+	@echo ""
+	@echo "  CC           =$(CC)"
+	@echo "  CFLAGS_EXTRA =$(CFLAGS_EXTRA)"
+	@echo "  CFLAGS       =$(CFLAGS)"
+	@echo "  CXX          =$(CXX)"
+	@echo "  CXXSTD       =$(CXXSTD)"
+	@echo "  CXXFLAGS     =$(CXXFLAGS)"
+	@echo ""
+	@echo "  LD           =$(LD)"
+	@echo "  LDFLAGS      =$(LDFLAGS)"
+	@echo "  EXE_LDFLAGS  =$(EXE_LDFLAGS)"
+	@echo "  LIBS         =$(LIBS)"
+	@echo ""
+	@echo "  MDBX_BUILD_OPTIONS   = $(MDBX_BUILD_OPTIONS)"
+	@echo "  MDBX_BUILD_TIMESTAMP = $(MDBX_BUILD_TIMESTAMP)"
+	@echo "  MDBX_BUILD_METADATA  = $(MDBX_BUILD_METADATA)"
+	@echo ""
+	@echo "## Assortment items for MDBX_BUILD_OPTIONS:"
+	@echo "##   Note that the defaults should already be correct for most platforms;"
+	@echo "##   you should not need to change any of these. Read their descriptions"
+#> dist-cutoff-begin
+ifeq ($(wildcard mdbx.c),mdbx.c)
+#< dist-cutoff-end
+	@echo "##   in README and source code (see mdbx.c) if you do."
+	@grep -h '#ifndef MDBX_' mdbx.c | grep -v BUILD | sort -u | sed 's/#ifndef /  /'
+#> dist-cutoff-begin
+else
+	@echo "##   in README and source code (see src/options.h) if you do."
+	@grep -h '#ifndef MDBX_' src/*.h | grep -v BUILD | sort -u | sed 's/#ifndef /  /'
+endif
+#< dist-cutoff-end
+
+lib libs libmdbx mdbx: libmdbx.a libmdbx.$(SO_SUFFIX)
+
+tools: $(MDBX_TOOLS)
+tools-static: $(addsuffix .static,$(MDBX_TOOLS)) $(addsuffix .static-lto,$(MDBX_TOOLS))
+
+strip: all
+	@echo '  STRIP libmdbx.$(SO_SUFFIX) $(MDBX_TOOLS)'
+	$(TRACE )strip libmdbx.$(SO_SUFFIX) $(MDBX_TOOLS)
+
+clean:
+	@echo '  REMOVE ...'
+	$(QUIET)rm -rf $(MDBX_TOOLS) mdbx_test @* *.[ao] *.[ls]o *.$(SO_SUFFIX) *.dSYM *~ tmp.db/* \
+		*.gcov *.log *.err src/*.o test/*.o mdbx_example dist @dist-check \
+		config.h src/config.h src/version.c *.tar* @buildflags.tag @dist-checked.tag \
+		mdbx_*.static mdbx_*.static-lto CMakeFiles
+
+MDBX_BUILD_FLAGS =$(strip MDBX_BUILD_CXX=$(MDBX_BUILD_CXX) $(MDBX_BUILD_OPTIONS) $(call select_by,MDBX_BUILD_CXX,$(CXXFLAGS) $(LDFLAGS) $(LIB_STDCXXFS) $(LIBS),$(CFLAGS) $(LDFLAGS) $(LIBS)))
+check_buildflags_tag:
+	$(QUIET)if [ "$(MDBX_BUILD_FLAGS)" != "$$(cat @buildflags.tag 2>&1)" ]; then \
+		echo -n "  CLEAN for build with specified flags..." && \
+		$(MAKE) IOARENA=false CXXSTD= -s clean >/dev/null && echo " Ok" && \
+		echo '$(MDBX_BUILD_FLAGS)' > @buildflags.tag; \
+	fi
+
+@buildflags.tag: check_buildflags_tag
+
+lib-static libmdbx.a: mdbx-static.o $(call select_by,MDBX_BUILD_CXX,mdbx++-static.o)
+	@echo '  AR $@'
+	$(QUIET)$(AR) rcs $@ $? $(HUSH)
+
+lib-shared libmdbx.$(SO_SUFFIX): mdbx-dylib.o $(call select_by,MDBX_BUILD_CXX,mdbx++-dylib.o)
+	@echo '  LD $@'
+	$(QUIET)$(call select_by,MDBX_BUILD_CXX,$(CXX) $(CXXFLAGS),$(CC) $(CFLAGS)) $^ -pthread -shared $(LDFLAGS) $(call select_by,MDBX_BUILD_CXX,$(LIB_STDCXXFS)) $(LIBS) -o $@
+
+ninja-assertions: CMAKE_OPT += -DMDBX_FORCE_ASSERTIONS=ON
+ninja-assertions: cmake-build
+ninja-debug: CMAKE_OPT += -DCMAKE_BUILD_TYPE=Debug
+ninja-debug: cmake-build
+ninja: cmake-build
+cmake-build:
+	@echo "  RUN: cmake -G Ninja && cmake --build"
+	$(QUIET)mkdir -p @cmake-ninja-build && $(CMAKE) $(CMAKE_OPT) -G Ninja -S . -B @cmake-ninja-build && $(CMAKE) --build @cmake-ninja-build
+
+ctest: cmake-build
+	@echo "  RUN: ctest .."
+	$(QUIET)$(CTEST) --test-dir @cmake-ninja-build --parallel `(nproc | sysctl -n hw.ncpu | echo 2) 2>/dev/null` --schedule-random $(CTEST_OPT)
+
+#> dist-cutoff-begin
+ifeq ($(wildcard mdbx.c),mdbx.c)
+#< dist-cutoff-end
+
+################################################################################
+# Amalgamated source code, i.e. distributed after `make dist`
+MAN_SRCDIR := man1/
+
+config.h: @buildflags.tag $(WAIT) mdbx.c $(lastword $(MAKEFILE_LIST)) LICENSE NOTICE
+	@echo '  MAKE $@'
+	$(QUIET)(echo '#define MDBX_BUILD_TIMESTAMP "$(MDBX_BUILD_TIMESTAMP)"' \
+	&& echo "#define MDBX_BUILD_FLAGS \"$$(cat @buildflags.tag)\"" \
+	&& echo '#define MDBX_BUILD_COMPILER "$(shell (LC_ALL=C $(CC) --version || echo 'Please use GCC or CLANG compatible compiler') | head -1)"' \
+	&& echo '#define MDBX_BUILD_TARGET "$(shell set -o pipefail; (LC_ALL=C $(CC) -v 2>&1 | grep -i '^Target:' | cut -d ' ' -f 2- || (LC_ALL=C $(CC) --version | grep -qi e2k && echo E2K) || echo 'Please use GCC or CLANG compatible compiler') | head -1)"' \
+	&& echo '#define MDBX_BUILD_CXX $(call select_by,MDBX_BUILD_CXX,1,0)' \
+	&& echo '#define MDBX_BUILD_METADATA "$(MDBX_BUILD_METADATA)"' \
+	) >$@
+
+mdbx-dylib.o: config.h mdbx.c mdbx.h $(lastword $(MAKEFILE_LIST)) LICENSE NOTICE
+	@echo '  CC $@'
+	$(QUIET)$(CC) $(CFLAGS) $(MDBX_BUILD_OPTIONS) '-DMDBX_CONFIG_H="config.h"' -DLIBMDBX_EXPORTS=1 -c mdbx.c -o $@
+
+mdbx-static.o: config.h mdbx.c mdbx.h $(lastword $(MAKEFILE_LIST)) LICENSE NOTICE
+	@echo '  CC $@'
+	$(QUIET)$(CC) $(CFLAGS) $(MDBX_BUILD_OPTIONS) '-DMDBX_CONFIG_H="config.h"' -ULIBMDBX_EXPORTS -c mdbx.c -o $@
+
+mdbx++-dylib.o: config.h mdbx.c++ mdbx.h mdbx.h++ $(lastword $(MAKEFILE_LIST)) LICENSE NOTICE
+	@echo '  CC $@'
+	$(QUIET)$(CXX) $(CXXFLAGS) $(MDBX_BUILD_OPTIONS) '-DMDBX_CONFIG_H="config.h"' -DLIBMDBX_EXPORTS=1 -c mdbx.c++ -o $@
+
+mdbx++-static.o: config.h mdbx.c++ mdbx.h mdbx.h++ $(lastword $(MAKEFILE_LIST)) LICENSE NOTICE
+	@echo '  CC $@'
+	$(QUIET)$(CXX) $(CXXFLAGS) $(MDBX_BUILD_OPTIONS) '-DMDBX_CONFIG_H="config.h"' -ULIBMDBX_EXPORTS -c mdbx.c++ -o $@
+
+mdbx_%:	mdbx_%.c mdbx-static.o
+	@echo '  CC+LD $@'
+	$(QUIET)$(CC) $(CFLAGS) $(MDBX_BUILD_OPTIONS) '-DMDBX_CONFIG_H="config.h"' $^ $(EXE_LDFLAGS) $(LIBS) -o $@
+
+mdbx_%.static: mdbx_%.c mdbx-static.o
+	@echo '  CC+LD $@'
+	$(QUIET)$(CC) $(CFLAGS) $(MDBX_BUILD_OPTIONS) '-DMDBX_CONFIG_H="config.h"' $^ $(EXE_LDFLAGS) -static -Wl,--strip-all -o $@
+
+mdbx_%.static-lto: mdbx_%.c config.h mdbx.c mdbx.h
+	@echo '  CC+LD $@'
+	$(QUIET)$(CC) $(CFLAGS) -Os -flto $(MDBX_BUILD_OPTIONS) '-DLIBMDBX_API=' '-DMDBX_CONFIG_H="config.h"' \
+		$< mdbx.c $(EXE_LDFLAGS) $(LIBS) -static -Wl,--strip-all -o $@
+
+#> dist-cutoff-begin
+else
+################################################################################
+# Plain (non-amalgamated) sources with test
+
+.PHONY: build-test build-test-with-valgrind check cross-gcc cross-qemu dist doxygen gcc-analyzer long-test
+.PHONY: reformat release-assets tags smoke test test-asan smoke-fault test-leak
+.PHONY: smoke-singleprocess test-singleprocess test-ubsan test-valgrind test-memcheck memcheck smoke-memcheck
+.PHONY: smoke-assertion test-assertion long-test-assertion test-ci test-ci-extra
+
+test-ci-extra: test-ci cross-gcc cross-qemu
+
+test-ci: check \
+	smoke-singleprocess smoke-fault smoke-memcheck smoke \
+	test-leak test-asan test-ubsan test-singleprocess test test-memcheck
+
+define uname2osal
+  case "$(UNAME)" in
+    CYGWIN*|MINGW*|MSYS*|Windows*) echo windows;;
+    *) echo unix;;
+  esac
+endef
+
+define uname2titer
+  case "$(UNAME)" in
+    Darwin*|Mach*) echo 2;;
+    *) echo 12;;
+  esac
+endef
+
+DIST_EXTRA := LICENSE NOTICE README.md CMakeLists.txt GNUmakefile Makefile ChangeLog.md VERSION.json config.h.in ntdll.def \
+	$(addprefix man1/, $(MANPAGES)) cmake/compiler.cmake cmake/profile.cmake cmake/utils.cmake .clang-format-ignore
+DIST_SRC   := mdbx.h mdbx.h++ mdbx.c mdbx.c++ $(addsuffix .c, $(MDBX_TOOLS))
+
+TEST_DB    ?= $(shell [ -d /dev/shm ] && echo /dev/shm || echo /tmp)/mdbx-test.db
+TEST_LOG   ?= $(shell [ -d /dev/shm ] && echo /dev/shm || echo /tmp)/mdbx-test.log
+TEST_OSAL  := $(shell $(uname2osal))
+TEST_ITER  := $(shell $(uname2titer))
+TEST_SRC   := test/osal-$(TEST_OSAL).c++ $(filter-out $(wildcard test/osal-*.c++),$(wildcard test/*.c++)) $(call select_by,MDBX_BUILD_CXX,,src/mdbx.c++)
+TEST_INC   := $(wildcard test/*.h++)
+TEST_OBJ   := $(patsubst %.c++,%.o,$(TEST_SRC))
+ifndef SED
+SED        := $(shell which gnu-sed 2>&- || echo sed)
+endif
+TAR        ?= $(shell which gnu-tar 2>&- || echo tar)
+ZIP        ?= $(shell which zip || echo "echo 'Please install zip'")
+CLANG_FORMAT ?= $(shell (which clang-format-19 || which clang-format) 2>/dev/null)
+
+reformat:
+	@echo '  RUNNING clang-format...'
+	$(QUIET)if [ -n "$(CLANG_FORMAT)" ]; then \
+		git ls-files | grep -E '\.(c|c++|h|h++)(\.in)?$$' | xargs -r $(CLANG_FORMAT) -i --style=file; \
+	else \
+		echo "clang-format version 19 not found for 'reformat'"; \
+	fi
+
+MAN_SRCDIR := src/man1/
+ALLOY_DEPS := $(shell git ls-files src/ | grep -e /tools -e /man -v)
+MDBX_GIT_DIR := $(shell if [ -d .git ]; then echo .git; elif [ -s .git -a -f .git ]; then grep '^gitdir: ' .git | cut -d ':' -f 2; else echo git_directory_is_absent; fi)
+MDBX_GIT_LASTVTAG := $(shell git describe --tags --dirty=-DIRTY --abbrev=0 '--match=v[0-9]*' 2>&- || echo 'Please fetch tags and/or install non-obsolete git version')
+MDBX_GIT_3DOT := $(shell set -o pipefail; echo "$(MDBX_GIT_LASTVTAG)" | $(SED) -n 's|^v*\([0-9]\{1,\}\.[0-9]\{1,\}\.[0-9]\{1,\}\)\(.*\)|\1|p' || echo 'Please fetch tags and/or use non-obsolete git version')
+MDBX_GIT_TWEAK := $(shell set -o pipefail; git rev-list $(shell git describe --tags --abbrev=0 '--match=v[0-9]*')..HEAD --count 2>&- || echo 'Please fetch tags and/or use non-obsolete git version')
+MDBX_GIT_TIMESTAMP := $(shell git show --no-patch --format=%cI HEAD 2>&- || echo 'Please install latest get version')
+MDBX_GIT_DESCRIBE := $(shell git describe --tags --long --dirty '--match=v[0-9]*' 2>&- || echo 'Please fetch tags and/or install non-obsolete git version')
+MDBX_GIT_PRERELEASE := $(shell echo "$(MDBX_GIT_LASTVTAG)" | $(SED) -n 's|^v*\([0-9]\{1,\}\.[0-9]\{1,\}\.[0-9]\{1,\}\)\(.*\)-\([-.0-1a-zA-Z]\+\)|\3|p')
+MDBX_VERSION_PURE = $(MDBX_GIT_3DOT)$(if $(filter-out 0,$(MDBX_GIT_TWEAK)),.$(MDBX_GIT_TWEAK),)$(if $(MDBX_GIT_PRERELEASE),-$(MDBX_GIT_PRERELEASE),)
+MDBX_VERSION_IDENT = $(shell set -o pipefail; echo -n '$(MDBX_GIT_DESCRIBE)' | tr -c -s '[a-zA-Z0-9.]' _)
+MDBX_VERSION_NODOT = $(subst .,_,$(MDBX_VERSION_IDENT))
+MDBX_BUILD_SOURCERY = $(shell set -o pipefail; $(MAKE) IOARENA=false CXXSTD= -s src/version.c >/dev/null && (openssl dgst -r -sha256 src/version.c || sha256sum src/version.c || shasum -a 256 src/version.c) 2>/dev/null | cut -d ' ' -f 1 || (echo 'Please install openssl or sha256sum or shasum' >&2 && echo sha256sum_is_no_available))_$(MDBX_VERSION_NODOT)
+MDBX_DIST_DIR = libmdbx-$(MDBX_VERSION_NODOT)
+
+# Extra options mdbx_test utility
+MDBX_SMOKE_EXTRA ?=
+
+check: DESTDIR = $(shell pwd)/@check-install
+check: CMAKE_OPT = -Werror=dev
+check: smoke-assertion ninja-assertions dist install test ctest
+
+smoke-assertion: MDBX_BUILD_OPTIONS:=$(strip $(MDBX_BUILD_OPTIONS) -DMDBX_FORCE_ASSERTIONS=1 -UNDEBUG -DMDBX_DEBUG=0)
+smoke-assertion: smoke
+test-assertion: MDBX_BUILD_OPTIONS:=$(strip $(MDBX_BUILD_OPTIONS) -DMDBX_FORCE_ASSERTIONS=1 -UNDEBUG -DMDBX_DEBUG=0)
+test-assertion: smoke
+long-test-assertion: MDBX_BUILD_OPTIONS:=$(strip $(MDBX_BUILD_OPTIONS) -DMDBX_FORCE_ASSERTIONS=1 -UNDEBUG -DMDBX_DEBUG=0)
+long-test-assertion: smoke
+
+smoke: build-test
+	@echo '  SMOKE `mdbx_test basic`...'
+	$(QUIET)rm -f $(TEST_DB) $(TEST_LOG).gz && (set -o pipefail; \
+		(./mdbx_test --table=+data.integer --keygen.split=29 --datalen.min=min --datalen.max=max --progress --console=no --repeat=$(TEST_ITER) --pathname=$(TEST_DB) --dont-cleanup-after $(MDBX_SMOKE_EXTRA) basic && \
+		./mdbx_test --mode=-writemap,-nosync-safe,-lifo --progress --console=no --repeat=$(TEST_ITER) --pathname=$(TEST_DB) --dont-cleanup-after $(MDBX_SMOKE_EXTRA) basic) \
+		| tee >(gzip --stdout >$(TEST_LOG).gz) | tail -n 42) \
+	&& ./mdbx_chk -vvn $(TEST_DB) && ./mdbx_chk -vvn $(TEST_DB)-copy
+
+smoke-singleprocess: build-test
+	@echo '  SMOKE `mdbx_test --nested`...'
+	$(QUIET)rm -f $(TEST_DB) $(TEST_LOG).gz && (set -o pipefail; \
+		(./mdbx_test --table=+data.integer --keygen.split=29 --datalen.min=min --datalen.max=max --progress --console=no --repeat=42 --pathname=$(TEST_DB) --dont-cleanup-after $(MDBX_SMOKE_EXTRA) --hill && \
+		./mdbx_test --progress --console=no --repeat=2 --pathname=$(TEST_DB) --dont-cleanup-before --dont-cleanup-after --copy && \
+		./mdbx_test --mode=-writemap,-nosync-safe,-lifo --progress --console=no --repeat=42 --pathname=$(TEST_DB) --dont-cleanup-after $(MDBX_SMOKE_EXTRA) --nested) \
+		| tee >(gzip --stdout >$(TEST_LOG).gz) | tail -n 42) \
+	&& ./mdbx_chk -vvn $(TEST_DB) && ./mdbx_chk -vvn $(TEST_DB)-copy
+
+smoke-fault: build-test
+	@echo '  SMOKE `mdbx_test --inject-writefault=42 basic`...'
+	$(QUIET)rm -f $(TEST_DB) $(TEST_LOG).gz && (set -o pipefail; ./mdbx_test --progress --console=no --pathname=$(TEST_DB) --inject-writefault=42 --dump-config --dont-cleanup-after $(MDBX_SMOKE_EXTRA) basic \
+		| tee >(gzip --stdout >$(TEST_LOG).gz) | tail -n 42) \
+	; ./mdbx_chk -vvnw $(TEST_DB) && ([ ! -e $(TEST_DB)-copy ] || ./mdbx_chk -vvn $(TEST_DB)-copy)
+
+test: build-test
+	@echo '  RUNNING `test/stochastic.sh --loops 2`...'
+	$(QUIET)test/stochastic.sh --dont-check-ram-size --loops 2 --db-upto-mb 256 --skip-make --taillog >$(TEST_LOG) || (cat $(TEST_LOG) && false)
+
+long-test: test-long
+test-long: build-test
+	@echo '  RUNNING `test/stochastic.sh --loops 42`...'
+	$(QUIET)test/stochastic.sh --loops 42 --db-upto-mb 1024 --extra --skip-make --taillog
+
+test-singleprocess: build-test
+	@echo '  RUNNING `test/stochastic.sh --single --loops 2`...'
+	$(QUIET)test/stochastic.sh --dont-check-ram-size --single --loops 2 --db-upto-mb 256 --skip-make --taillog >$(TEST_LOG) || (cat $(TEST_LOG) && false)
+
+test-valgrind: test-memcheck
+test-memcheck: CFLAGS_EXTRA=-Ofast -DENABLE_MEMCHECK
+test-memcheck: build-test
+	@echo '  RUNNING `test/stochastic.sh --with-valgrind --loops 2`...'
+	$(QUIET)test/stochastic.sh --with-valgrind --loops 2 --db-upto-mb 256 --skip-make >$(TEST_LOG) || (cat $(TEST_LOG) && false)
+
+memcheck: smoke-memcheck
+smoke-memcheck: VALGRIND=valgrind --trace-children=yes --log-file=valgrind-%p.log --leak-check=full --track-origins=yes --read-var-info=yes --error-exitcode=42 --suppressions=test/valgrind_suppress.txt
+smoke-memcheck: CFLAGS_EXTRA=-Ofast -DENABLE_MEMCHECK
+smoke-memcheck: build-test
+	@echo "  SMOKE \`mdbx_test basic\` under Valgrind's memcheck..."
+	$(QUIET)rm -f valgrind-*.log $(TEST_DB) $(TEST_LOG).gz && (set -o pipefail; ( \
+		$(VALGRIND) ./mdbx_test --table=+data.fixed --keygen.split=29 --datalen=35 --progress --console=no --repeat=2 --pathname=$(TEST_DB) --dont-cleanup-after $(MDBX_SMOKE_EXTRA) basic && \
+		$(VALGRIND) ./mdbx_test --progress --console=no --pathname=$(TEST_DB) --dont-cleanup-before --dont-cleanup-after --copy && \
+		$(VALGRIND) ./mdbx_test --mode=-writemap,-nosync-safe,-lifo --progress --console=no --repeat=4 --pathname=$(TEST_DB) --dont-cleanup-after $(MDBX_SMOKE_EXTRA) basic && \
+		$(VALGRIND) ./mdbx_chk -vvn $(TEST_DB) && \
+		$(VALGRIND) ./mdbx_chk -vvn $(TEST_DB)-copy \
+	) | tee >(gzip --stdout >$(TEST_LOG).gz) | tail -n 42)
+
+gcc-analyzer:
+	@echo '  RE-BUILD with `-fanalyzer` option...'
+	@echo "NOTE: There a lot of false-positive warnings at 2020-05-01 by pre-release GCC-10 (20200328, Red Hat 10.0.1-0.11)"
+	$(QUIET)$(MAKE) IOARENA=false CXXSTD=$(CXXSTD) CFLAGS_EXTRA="-Og -fanalyzer -Wno-error" build-test
+
+test-ubsan:
+	@echo '  RE-TEST with `-fsanitize=undefined` option...'
+	$(QUIET)$(MAKE) IOARENA=false CXXSTD=$(CXXSTD) CFLAGS_EXTRA="-DENABLE_UBSAN -Ofast -fsanitize=undefined -fsanitize-undefined-trap-on-error" test
+
+test-asan:
+	@echo '  RE-TEST with `-fsanitize=address` option...'
+	$(QUIET)$(MAKE) IOARENA=false CXXSTD=$(CXXSTD) CFLAGS_EXTRA="-Os -fsanitize=address" test
+
+test-leak:
+	@echo '  RE-TEST with `-fsanitize=leak` option...'
+	$(QUIET)$(MAKE) IOARENA=false CXXSTD=$(CXXSTD) CFLAGS_EXTRA="-fsanitize=leak" test
+
+mdbx_example: mdbx.h example/example-mdbx.c libmdbx.$(SO_SUFFIX)
+	@echo '  CC+LD $@'
+	$(QUIET)$(CC) $(CFLAGS) -I. example/example-mdbx.c ./libmdbx.$(SO_SUFFIX) -o $@
+
+build-test: all mdbx_example mdbx_test
+
+define test-rule
+$(patsubst %.c++,%.o,$(1)): $(1) $(TEST_INC) $(HEADERS) $(lastword $(MAKEFILE_LIST))
+	@echo '  CC $$@'
+	$(QUIET)$$(CXX) $$(CXXFLAGS) $$(MDBX_BUILD_OPTIONS) -DMDBX_BUILD_CXX=1 -DMDBX_WITHOUT_MSVC_CRT=0 -c $(1) -o $$@
+
+endef
+$(foreach file,$(TEST_SRC),$(eval $(call test-rule,$(file))))
+
+define tool-rule
+mdbx_$(1):	src/tools/$(1).c libmdbx.a
+	@echo '  CC+LD $$@'
+	$(QUIET)$$(CC) $$(CFLAGS) $$(MDBX_BUILD_OPTIONS) -Isrc '-DMDBX_CONFIG_H="config.h"' $$^ $$(EXE_LDFLAGS) $$(LIBS) -o $$@
+
+mdbx_$(1).static:	src/tools/$(1).c mdbx-static.o
+	@echo '  CC+LD $$@'
+	$(QUIET)$$(CC) $$(CFLAGS) $$(MDBX_BUILD_OPTIONS) -Isrc '-DMDBX_CONFIG_H="config.h"' $$^ $$(EXE_LDFLAGS) $$(LIBS) -static -Wl,--strip-all -o $$@
+
+mdbx_$(1).static-lto: src/tools/$(1).c src/config.h src/version.c src/alloy.c $(ALLOY_DEPS)
+	@echo '  CC+LD $$@'
+	$(QUIET)$$(CC) $$(CFLAGS) -Os -flto $$(MDBX_BUILD_OPTIONS) -Isrc '-DLIBMDBX_API=' '-DMDBX_CONFIG_H="config.h"' \
+		$$< src/alloy.c $$(EXE_LDFLAGS) $$(LIBS) -static -Wl,--strip-all -o $$@
+
+endef
+$(foreach file,$(TOOLS),$(eval $(call tool-rule,$(file))))
+
+mdbx_test: $(TEST_OBJ) libmdbx.$(SO_SUFFIX)
+	@echo '  LD $@'
+	$(QUIET)$(CXX) $(CXXFLAGS) $(TEST_OBJ) -Wl,-rpath . -L . -l mdbx $(EXE_LDFLAGS) $(LIBS) -o $@
+
+$(MDBX_GIT_DIR)/HEAD $(MDBX_GIT_DIR)/index $(MDBX_GIT_DIR)/refs/tags:
+	@echo '*** ' >&2
+	@echo '*** Please don''t use tarballs nor zips which are automatically provided by Github !' >&2
+	@echo '*** These archives do not contain version information and thus are unfit to build libmdbx.' >&2
+	@echo '*** ' >&2
+	@echo '*** Instead just follow the https://libmdbx.dqdkfa.ru/usage.html' >&2
+	@echo '*** PLEASE, AVOID USING ANY OTHER TECHNIQUES.' >&2
+	@echo '*** ' >&2
+	@false
+
+src/version.c: src/version.c.in $(lastword $(MAKEFILE_LIST)) $(MDBX_GIT_DIR)/HEAD $(MDBX_GIT_DIR)/index $(MDBX_GIT_DIR)/refs/tags LICENSE NOTICE
+	@echo '  MAKE $@'
+	$(QUIET)$(SED) \
+		-e "s|@MDBX_GIT_TIMESTAMP@|$(MDBX_GIT_TIMESTAMP)|" \
+		-e "s|@MDBX_GIT_TREE@|$(shell git show --no-patch --format=%T HEAD || echo 'Please install latest get version')|" \
+		-e "s|@MDBX_GIT_COMMIT@|$(shell git show --no-patch --format=%H HEAD || echo 'Please install latest get version')|" \
+		-e "s|@MDBX_GIT_DESCRIBE@|$(MDBX_GIT_DESCRIBE)|" \
+		-e "s|\$${MDBX_VERSION_MAJOR}|$(shell echo '$(MDBX_GIT_3DOT)' | cut -d . -f 1)|" \
+		-e "s|\$${MDBX_VERSION_MINOR}|$(shell echo '$(MDBX_GIT_3DOT)' | cut -d . -f 2)|" \
+		-e "s|\$${MDBX_VERSION_PATCH}|$(shell echo '$(MDBX_GIT_3DOT)' | cut -d . -f 3)|" \
+		-e "s|\$${MDBX_VERSION_TWEAK}|$(MDBX_GIT_TWEAK)|" \
+		-e "s|@MDBX_VERSION_PRERELEASE@|$(MDBX_GIT_PRERELEASE)|" \
+		-e "s|@MDBX_VERSION_PURE@|$(MDBX_VERSION_PURE)|" \
+	src/version.c.in >$@
+
+src/config.h: @buildflags.tag $(WAIT) src/version.c $(lastword $(MAKEFILE_LIST)) LICENSE NOTICE
+	@echo '  MAKE $@'
+	$(QUIET)(echo '#define MDBX_BUILD_TIMESTAMP "$(MDBX_BUILD_TIMESTAMP)"' \
+	&& echo "#define MDBX_BUILD_FLAGS \"$$(cat @buildflags.tag)\"" \
+	&& echo '#define MDBX_BUILD_COMPILER "$(shell (LC_ALL=C $(CC) --version || echo 'Please use GCC or CLANG compatible compiler') | head -1)"' \
+	&& echo '#define MDBX_BUILD_TARGET "$(shell set -o pipefail; (LC_ALL=C $(CC) -v 2>&1 | grep -i '^Target:' | cut -d ' ' -f 2- || (LC_ALL=C $(CC) --version | grep -qi e2k && echo E2K) || echo 'Please use GCC or CLANG compatible compiler') | head -1)"' \
+	&& echo '#define MDBX_BUILD_SOURCERY $(MDBX_BUILD_SOURCERY)' \
+	&& echo '#define MDBX_BUILD_CXX $(call select_by,MDBX_BUILD_CXX,1,0)' \
+	&& echo '#define MDBX_BUILD_METADATA "$(MDBX_BUILD_METADATA)"' \
+	) >$@
+
+mdbx-dylib.o: src/config.h src/version.c src/alloy.c $(ALLOY_DEPS) $(lastword $(MAKEFILE_LIST)) LICENSE NOTICE
+	@echo '  CC $@'
+	$(QUIET)$(CC) $(CFLAGS) $(MDBX_BUILD_OPTIONS) '-DMDBX_CONFIG_H="config.h"' -DLIBMDBX_EXPORTS=1 -c src/alloy.c -o $@
+
+mdbx-static.o: src/config.h src/version.c src/alloy.c $(ALLOY_DEPS) $(lastword $(MAKEFILE_LIST)) LICENSE NOTICE
+	@echo '  CC $@'
+	$(QUIET)$(CC) $(CFLAGS) $(MDBX_BUILD_OPTIONS) '-DMDBX_CONFIG_H="config.h"' -ULIBMDBX_EXPORTS -c src/alloy.c -o $@
+
+docs/Doxyfile: docs/Doxyfile.in src/version.c $(lastword $(MAKEFILE_LIST))
+	@echo '  MAKE $@'
+	$(QUIET)$(SED) \
+		-e "s|@MDBX_GIT_TIMESTAMP@|$(MDBX_GIT_TIMESTAMP)|" \
+		-e "s|@MDBX_GIT_TREE@|$(shell git show --no-patch --format=%T HEAD || echo 'Please install latest get version')|" \
+		-e "s|@MDBX_GIT_COMMIT@|$(shell git show --no-patch --format=%H HEAD || echo 'Please install latest get version')|" \
+		-e "s|@MDBX_GIT_DESCRIBE@|$(MDBX_GIT_DESCRIBE)|" \
+		-e "s|\$${MDBX_VERSION_MAJOR}|$(shell echo '$(MDBX_GIT_3DOT)' | cut -d . -f 1)|" \
+		-e "s|\$${MDBX_VERSION_MINOR}|$(shell echo '$(MDBX_GIT_3DOT)' | cut -d . -f 2)|" \
+		-e "s|\$${MDBX_VERSION_PATCH}|$(shell echo '$(MDBX_GIT_3DOT)' | cut -d . -f 3)|" \
+		-e "s|\$${MDBX_VERSION_TWEAK}|$(MDBX_GIT_TWEAK)|" \
+		-e "s|@MDBX_VERSION_PRERELEASE@|$(MDBX_GIT_PRERELEASE)|" \
+		-e "s|@MDBX_VERSION_PURE@|$(MDBX_VERSION_PURE)|" \
+	docs/Doxyfile.in >$@
+
+define md-extract-section
+docs/__$(1).md: $(2) $(lastword $(MAKEFILE_LIST))
+	@echo '  EXTRACT $1'
+	$(QUIET)$(SED) -n '/<!-- section-begin $(1) -->/,/<!-- section-end -->/p' $(2) >$$@ && test -s $$@
+
+endef
+$(foreach section,overview mithril characteristics improvements history usage performance bindings,$(eval $(call md-extract-section,$(section),README.md)))
+
+docs/contrib.fame: src/version.c $(lastword $(MAKEFILE_LIST))
+	@echo '  MAKE $@'
+	$(QUIET)echo "" > $@ && git fame --show-email --format=md --silent-progress -w -M -C | grep '^|' >> $@
+
+docs/overall.md: docs/__overview.md docs/_toc.md docs/__mithril.md docs/__history.md COPYRIGHT LICENSE NOTICE $(lastword $(MAKEFILE_LIST))
+	@echo '  MAKE $@'
+	$(QUIET)echo -e "\\mainpage Overall\n\\section brief Brief" | cat - $(filter %.md, $^) >$@ && echo -e "\n\n\nLicense\n=======\n" | cat - LICENSE >>$@
+
+docs/intro.md: docs/_preface.md docs/__characteristics.md docs/__improvements.md docs/_restrictions.md docs/__performance.md
+	@echo '  MAKE $@'
+	$(QUIET)cat $^ | $(SED) 's/^Performance comparison$$/Performance comparison {#performance}/;s/^Improvements beyond LMDB$$/Improvements beyond LMDB {#improvements}/' >$@
+
+docs/usage.md: docs/__usage.md docs/_starting.md docs/__bindings.md
+	@echo '  MAKE $@'
+	$(QUIET)echo -e "\\page usage Usage\n\\section getting Building & Embedding" | cat - $^ | $(SED) 's/^Bindings$$/Bindings {#bindings}/' >$@
+
+doxygen: docs/Doxyfile docs/overall.md docs/intro.md docs/usage.md mdbx.h mdbx.h++ src/options.h ChangeLog.md COPYRIGHT LICENSE NOTICE docs/favicon.ico docs/manifest.webmanifest docs/ld+json $(lastword $(MAKEFILE_LIST))
+	@echo '  RUNNING doxygen...'
+	$(QUIET)rm -rf docs/html && \
+	cat mdbx.h | tr '\n' '\r' | $(SED) -e 's/LIBMDBX_INLINE_API\s*(\s*\([^,]\+\),\s*\([^,]\+\),\s*(\s*\([^)]\+\)\s*)\s*)\s*{/inline \1 \2(\3) {/g' | tr '\r' '\n' >docs/mdbx.h && \
+	cp mdbx.h++ src/options.h ChangeLog.md docs/ && (cd docs && doxygen Doxyfile $(HUSH)) && cp COPYRIGHT LICENSE NOTICE docs/favicon.ico docs/manifest.webmanifest docs/html/ && \
+	$(SED) -i docs/html/index.html -e '/\/MathJax.js"><\/script>/r docs/ld+json' -e 's/<title>libmdbx: Overall<\/title>//;T;r docs/title'
+
+mdbx++-dylib.o: src/config.h src/mdbx.c++ mdbx.h mdbx.h++ $(lastword $(MAKEFILE_LIST))
+	@echo '  CC $@'
+	$(QUIET)$(CXX) $(CXXFLAGS) $(MDBX_BUILD_OPTIONS) '-DMDBX_CONFIG_H="config.h"' -DLIBMDBX_EXPORTS=1 -c src/mdbx.c++ -o $@
+
+mdbx++-static.o: src/config.h src/mdbx.c++ mdbx.h mdbx.h++ $(lastword $(MAKEFILE_LIST))
+	@echo '  CC $@'
+	$(QUIET)$(CXX) $(CXXFLAGS) $(MDBX_BUILD_OPTIONS) '-DMDBX_CONFIG_H="config.h"' -ULIBMDBX_EXPORTS -c src/mdbx.c++ -o $@
+
+dist: tags $(WAIT) @dist-checked.tag libmdbx-sources-$(MDBX_VERSION_IDENT).tar.gz $(lastword $(MAKEFILE_LIST))
+	@echo '  AMALGAMATION is done'
+
+tags:
+	@echo '  FETCH git tags...'
+	$(QUIET)git fetch --tags --force
+
+release-assets: libmdbx-amalgamated-$(MDBX_GIT_3DOT).zpaq \
+	libmdbx-amalgamated-$(MDBX_GIT_3DOT).tar.xz \
+	libmdbx-amalgamated-$(MDBX_GIT_3DOT).tar.bz2 \
+	libmdbx-amalgamated-$(MDBX_GIT_3DOT).tar.gz \
+	libmdbx-amalgamated-$(subst .,_,$(MDBX_GIT_3DOT)).zip
+	$(QUIET)([ \
+		"$$(set -o pipefail; git describe | $(SED) -n '/^v[0-9]\{1,\}\.[0-9]\{1,\}\.[0-9]\{1,\}$$/p' || echo fail-left)" \
+	== \
+		"$$(git describe --tags --dirty=-dirty || echo fail-right)" ] \
+		|| (echo 'ERROR: Is not a valid release because not in the clean state with a suitable annotated tag!!!' >&2 && false)) \
+	&& echo '  RELEASE ASSETS are done'
+
+@dist-checked.tag: $(addprefix $(DIST_DIR)/, $(DIST_SRC) $(DIST_EXTRA))
+	@echo -n '  VERIFY amalgamated sources...'
+	$(QUIET)rm -rf $@ $(DIST_DIR)/@tmp-essentials.inc $(DIST_DIR)/@tmp-internals.inc \
+	&& if grep -R "define xMDBX_ALLOY" dist | grep -q MDBX_BUILD_SOURCERY; then echo "sed output is WRONG!" >&2; exit 2; fi \
+	&& rm -rf @dist-check && cp -r -p $(DIST_DIR) @dist-check && ($(MAKE) -j IOARENA=false CXXSTD=$(CXXSTD) -C @dist-check all ninja-assertions >@dist-check.log 2>@dist-check.err || (cat @dist-check.err && exit 1)) \
+	&& touch $@ || (echo " FAILED! See @dist-check.log and @dist-check.err" >&2; exit 2) && echo " Ok"
+
+%.tar.gz: @dist-checked.tag
+	@echo '  CREATE $@'
+	$(QUIET)$(TAR) -c $(shell LC_ALL=C $(TAR) --help | grep -q -- '--owner' && echo '--owner=0 --group=0') -f - -C dist $(DIST_SRC) $(DIST_EXTRA) | gzip -c -9 >$@
+
+%.tar.xz: @dist-checked.tag
+	@echo '  CREATE $@'
+	$(QUIET)$(TAR) -c $(shell LC_ALL=C $(TAR) --help | grep -q -- '--owner' && echo '--owner=0 --group=0') -f - -C dist $(DIST_SRC) $(DIST_EXTRA) | xz -9 -z >$@
+
+%.tar.bz2: @dist-checked.tag
+	@echo '  CREATE $@'
+	$(QUIET)$(TAR) -c $(shell LC_ALL=C $(TAR) --help | grep -q -- '--owner' && echo '--owner=0 --group=0') -f - -C dist $(DIST_SRC) $(DIST_EXTRA) | bzip2 -9 -z >$@
+
+%.zip: @dist-checked.tag
+	@echo '  CREATE $@'
+	$(QUIET)rm -rf $@ && (cd dist && $(ZIP) -9 ../$@ $(DIST_SRC) $(DIST_EXTRA)) &>@zip.log
+
+%.zpaq: @dist-checked.tag
+	@echo '  CREATE $@'
+	$(QUIET)rm -rf $@ && (cd dist && zpaq a ../$@ $(DIST_SRC) $(DIST_EXTRA) -m59) &>@zpaq.log
+
+$(DIST_DIR)/@tmp-essentials.inc: src/version.c $(ALLOY_DEPS) $(lastword $(MAKEFILE_LIST))
+	@echo '  ALLOYING...'
+	$(QUIET)mkdir -p dist \
+	&& (grep -v '#include ' src/alloy.c && echo '#define MDBX_BUILD_SOURCERY $(MDBX_BUILD_SOURCERY)' \
+	&& $(SED) \
+		-e 's|#include "../mdbx.h"|@INCLUDE "mdbx.h"|' \
+		-e '/#include "preface.h"/r src/preface.h' \
+		-e '/#include "osal.h"/r src/osal.h' \
+		-e '/#include "options.h"/r src/options.h' \
+		-e '/#include "atomics-types.h"/r src/atomics-types.h' \
+		-e '/#include "layout-dxb.h"/r src/layout-dxb.h' \
+		-e '/#include "layout-lck.h"/r src/layout-lck.h' \
+		-e '/#include "logging_and_debug.h"/r src/logging_and_debug.h' \
+		-e '/#include "utils.h"/r src/utils.h' \
+		-e '/#include "pnl.h"/r src/pnl.h' \
+		src/essentials.h \
+	| $(SED) \
+		-e '/#pragma once/d' -e '/#include "/d' \
+		-e '/ clang-format o/d' -e '/ \*INDENT-O/d' \
+		| grep -v '^/// ') >$@
+
+$(DIST_DIR)/@tmp-internals.inc: $(DIST_DIR)/@tmp-essentials.inc src/version.c $(ALLOY_DEPS) $(lastword $(MAKEFILE_LIST))
+	$(QUIET)(cat $(DIST_DIR)/@tmp-essentials.inc \
+	&& $(SED) \
+		-e '/#include "essentials.h"/d' \
+		-e '/#include "atomics-ops.h"/r src/atomics-ops.h' \
+		-e '/#include "proto.h"/r src/proto.h' \
+		-e '/#include "rkl.h"/r src/rkl.h' \
+		-e '/#include "txl.h"/r src/txl.h' \
+		-e '/#include "unaligned.h"/r src/unaligned.h' \
+		-e '/#include "cogs.h"/r src/cogs.h' \
+		-e '/#include "cursor.h"/r src/cursor.h' \
+		-e '/#include "dbi.h"/r src/dbi.h' \
+		-e '/#include "dpl.h"/r src/dpl.h' \
+		-e '/#include "gc.h"/r src/gc.h' \
+		-e '/#include "lck.h"/r src/lck.h' \
+		-e '/#include "meta.h"/r src/meta.h' \
+		-e '/#include "node.h"/r src/node.h' \
+		-e '/#include "page-iov.h"/r src/page-iov.h' \
+		-e '/#include "page-ops.h"/r src/page-ops.h' \
+		-e '/#include "spill.h"/r src/spill.h' \
+		-e '/#include "sort.h"/r src/sort.h' \
+		-e '/#include "tls.h"/r src/tls.h' \
+		-e '/#include "walk.h"/r src/walk.h' \
+		-e '/#include "windows-import.h"/r src/windows-import.h' \
+		src/internals.h \
+	| $(SED) \
+		-e '/#pragma once/d' -e '/#include "/d' \
+		-e '/ clang-format o/d' -e '/ \*INDENT-O/d' \
+		| grep -v '^/// ') >$@
+
+$(DIST_DIR)/mdbx.c: $(DIST_DIR)/@tmp-internals.inc $(lastword $(MAKEFILE_LIST))
+	@echo '  MAKE $@'
+	$(QUIET)(cat $(DIST_DIR)/@tmp-internals.inc $(shell git ls-files src/*.c | grep -v alloy) src/version.c | $(SED) \
+		-e '/#include "debug_begin.h"/r src/debug_begin.h' \
+		-e '/#include "debug_end.h"/r src/debug_end.h' \
+	) | $(SED) -e '/#include "/d;/#pragma once/d' -e 's|@INCLUDE|#include|' \
+		-e '/ clang-format o/d;/ \*INDENT-O/d' -e '3i /* clang-format off */' | cat -s >$@
+
+$(DIST_DIR)/mdbx.c++: $(DIST_DIR)/@tmp-essentials.inc src/mdbx.c++ $(lastword $(MAKEFILE_LIST))
+	@echo '  MAKE $@'
+	$(QUIET)cat $(DIST_DIR)/@tmp-essentials.inc src/mdbx.c++ | $(SED) \
+		-e '/#define xMDBX_ALLOY/d' \
+		-e '/#include "/d;/#pragma once/d' \
+		-e 's|@INCLUDE|#include|;s|"mdbx.h"|"mdbx.h++"|' \
+		-e '/ clang-format o/d;/ \*INDENT-O/d' -e '3i /* clang-format off */' | cat -s >$@
+
+define dist-tool-rule
+$(DIST_DIR)/mdbx_$(1).c: src/tools/$(1).c src/tools/wingetopt.h src/tools/wingetopt.c \
+		$(DIST_DIR)/@tmp-internals.inc $(lastword $(MAKEFILE_LIST))
+	@echo '  MAKE $$@'
+	$(QUIET)mkdir -p dist && $(SED) \
+		-e '/#include "essentials.h"/r $(DIST_DIR)/@tmp-essentials.inc' \
+		-e '/#include "wingetopt.h"/r src/tools/wingetopt.c' \
+		-e '/ clang-format o/d' -e '/ \*INDENT-O/d' \
+		src/tools/$(1).c \
+	| $(SED) -e '/#include "/d;/#pragma once/d;/#define xMDBX_ALLOY/d' -e 's|@INCLUDE|#include|' \
+		-e '/ clang-format o/d;/ \*INDENT-O/d' -e '9i /* clang-format off */' | cat -s >$$@
+
+endef
+$(foreach file,$(TOOLS),$(eval $(call dist-tool-rule,$(file))))
+
+define dist-extra-rule
+$(DIST_DIR)/$(1): $(1) src/version.c $(lastword $(MAKEFILE_LIST))
+	@echo '  REFINE $$@'
+	$(QUIET)mkdir -p $$(dir $$@) && $(SED) -e '/^#> dist-cutoff-begin/,/^#< dist-cutoff-end/d' $$< | cat -s >$$@
+
+endef
+$(foreach file,mdbx.h mdbx.h++ $(filter-out man1/% VERSION.json .clang-format-ignore %.in ntdll.def,$(DIST_EXTRA)),$(eval $(call dist-extra-rule,$(file))))
+
+$(DIST_DIR)/VERSION.json: src/version.c
+	@echo '  MAKE $@'
+	$(QUIET)mkdir -p $(DIST_DIR)/ && echo "{ \"git_describe\": \"$(MDBX_GIT_DESCRIBE)\", \"git_timestamp\": \"$(MDBX_GIT_TIMESTAMP)\", \"git_tree\": \"$(shell git show --no-patch --format=%T HEAD 2>&1)\", \"git_commit\": \"$(shell git show --no-patch --format=%H HEAD 2>&1)\", \"semver\": \"$(MDBX_VERSION_PURE)\" }" >$@
+
+$(DIST_DIR)/.clang-format-ignore: $(lastword $(MAKEFILE_LIST))
+	@echo '  MAKE $@'
+	$(QUIET)echo "$(filter-out %.h %h++,$(DIST_SRC))" | tr ' ' \\n > $@
+
+$(DIST_DIR)/ntdll.def: src/ntdll.def
+	@echo '  COPY $@'
+	$(QUIET)mkdir -p $(DIST_DIR)/ && cp $< $@
+
+$(DIST_DIR)/config.h.in: src/config.h.in
+	@echo '  COPY $@'
+	$(QUIET)mkdir -p $(DIST_DIR)/ && cp $< $@
+
+$(DIST_DIR)/man1/mdbx_%.1: src/man1/mdbx_%.1
+	@echo '  COPY $@'
+	$(QUIET)mkdir -p $(DIST_DIR)/man1/ && cp $< $@
+
+endif
+
+################################################################################
+# Cross-compilation simple test
+
+CROSS_LIST = \
+	mips64-linux-gnuabi64-gcc mips-linux-gnu-gcc \
+	hppa-linux-gnu-gcc s390x-linux-gnu-gcc \
+	powerpc64-linux-gnu-gcc powerpc-linux-gnu-gcc \
+	arm-linux-gnueabihf-gcc aarch64-linux-gnu-gcc
+
+## On Ubuntu Focal (22.04) with QEMU 6.2 (1:6.2+dfsg-2ubuntu6.6) & GCC 11.3 (11.3.0-1ubuntu1~22.04)
+# sh4-linux-gnu-gcc           - coredump (qemu mmap-troubles)
+# sparc64-linux-gnu-gcc       - coredump (qemu mmap-troubles, previously: qemu fails fcntl for F_SETLK/F_GETLK)
+# alpha-linux-gnu-gcc         - coredump (qemu mmap-troubles)
+# risc64-linux-gnu-gcc        - coredump (qemu qemu fails fcntl for F_SETLK/F_GETLK)
+CROSS_LIST_NOQEMU = sh4-linux-gnu-gcc sparc64-linux-gnu-gcc alpha-linux-gnu-gcc riscv64-linux-gnu-gcc
+
+cross-gcc:
+	@echo '  Re-building by cross-compiler for: $(CROSS_LIST_NOQEMU) $(CROSS_LIST)'
+	@echo "CORRESPONDING CROSS-COMPILERs ARE REQUIRED."
+	@echo "FOR INSTANCE: sudo apt install \$$(apt list 'g++-*' | grep 'g++-[a-z0-9]\+-linux-gnu/' | cut -f 1 -d / | sort -u)"
+	$(QUIET)for CC in $(CROSS_LIST_NOQEMU) $(CROSS_LIST); do \
+		echo "===================== $$CC"; \
+		$(MAKE) IOARENA=false CXXSTD= clean && CC=$$CC CXX=$$(echo $$CC | $(SED) 's/-gcc/-g++/') EXE_LDFLAGS=-static $(MAKE) IOARENA=false all || exit $$?; \
+	done
+
+# Unfortunately qemu don't provide robust support for futexes.
+# Therefore it is impossible to run full multi-process tests.
+cross-qemu:
+	@echo '  Re-building by cross-compiler and re-check by QEMU for: $(CROSS_LIST)'
+	@echo "CORRESPONDING CROSS-COMPILERs AND QEMUs ARE REQUIRED."
+	@echo "FOR INSTANCE: "
+	@echo "	1) sudo apt install \$$(apt list 'g++-*' | grep 'g++-[a-z0-9]\+-linux-gnu/' | cut -f 1 -d / | sort -u)"
+	@echo "	2) sudo apt install binfmt-support qemu-user-static qemu-user \$$(apt list 'qemu-system-*' | grep 'qemu-system-[a-z0-9]\+/' | cut -f 1 -d / | sort -u)"
+	$(QUIET)for CC in $(CROSS_LIST); do \
+		echo "===================== $$CC + qemu"; \
+		$(MAKE) IOARENA=false CXXSTD= clean && \
+			CC=$$CC CXX=$$(echo $$CC | $(SED) 's/-gcc/-g++/') EXE_LDFLAGS=-static MDBX_BUILD_OPTIONS="-DMDBX_SAFE4QEMU $(MDBX_BUILD_OPTIONS)" \
+			$(MAKE) IOARENA=false smoke-singleprocess test-singleprocess || exit $$?; \
+	done
+
+#< dist-cutoff-end
+
+install: $(LIBRARIES) $(MDBX_TOOLS) $(HEADERS)
+	@echo '  INSTALLING...'
+	$(QUIET)mkdir -p $(DESTDIR)$(prefix)/bin$(suffix) && \
+		$(INSTALL) -p $(EXE_INSTALL_FLAGS) $(MDBX_TOOLS) $(DESTDIR)$(prefix)/bin$(suffix)/ && \
+	mkdir -p $(DESTDIR)$(prefix)/lib$(suffix)/ && \
+		$(INSTALL) -p $(EXE_INSTALL_FLAGS) $(filter-out libmdbx.a,$(LIBRARIES)) $(DESTDIR)$(prefix)/lib$(suffix)/ && \
+	mkdir -p $(DESTDIR)$(prefix)/lib$(suffix)/ && \
+		$(INSTALL) -p libmdbx.a $(DESTDIR)$(prefix)/lib$(suffix)/ && \
+	mkdir -p $(DESTDIR)$(prefix)/include/ && \
+		$(INSTALL) -p -m 444 $(HEADERS) $(DESTDIR)$(prefix)/include/ && \
+	mkdir -p $(DESTDIR)$(mandir)/man1/ && \
+		$(INSTALL) -p -m 444 $(addprefix $(MAN_SRCDIR), $(MANPAGES)) $(DESTDIR)$(mandir)/man1/
+
+install-strip: EXE_INSTALL_FLAGS = -s
+install-strip: install
+
+install-no-strip: EXE_INSTALL_FLAGS =
+install-no-strip: install
+
+uninstall:
+	@echo '  UNINSTALLING/REMOVE...'
+	$(QUIET)rm -f $(addprefix $(DESTDIR)$(prefix)/bin$(suffix)/,$(MDBX_TOOLS)) \
+		$(addprefix $(DESTDIR)$(prefix)/lib$(suffix)/,$(LIBRARIES)) \
+		$(addprefix $(DESTDIR)$(prefix)/include/,$(HEADERS)) \
+		$(addprefix $(DESTDIR)$(mandir)/man1/,$(MANPAGES))
+
+################################################################################
+# Benchmarking by ioarena
+
+ifeq ($(origin IOARENA),undefined)
+IOARENA := $(shell \
+  (test -x ../ioarena/@BUILD/src/ioarena && echo ../ioarena/@BUILD/src/ioarena) || \
+  (test -x ../../@BUILD/src/ioarena && echo ../../@BUILD/src/ioarena) || \
+  (test -x ../../src/ioarena && echo ../../src/ioarena) || which ioarena 2>&- || \
+  (echo false && echo '$(TIP) Clone and build the https://abf.io/erthink/ioarena.git within a neighbouring directory for availability of benchmarking.' >&2))
+endif
+NN	?= 25000000
+BENCH_CRUD_MODE ?= nosync
+
+bench-clean:
+	@echo '  REMOVE bench-*.txt _ioarena/*'
+	$(QUIET)rm -rf bench-*.txt _ioarena/*
+
+re-bench: bench-clean bench
+
+ifeq ($(or $(IOARENA),false),false)
+bench bench-quartet bench-triplet bench-couple:
+	$(QUIET)echo 'The `ioarena` benchmark is required.' >&2 && \
+	echo 'Please clone and build the https://abf.io/erthink/ioarena.git within a neighbouring `ioarena` directory.' >&2 && \
+	false
+
+else
+
+.PHONY: bench bench-clean bench-couple re-bench bench-quartet bench-triplet
+
+define bench-rule
+bench-$(1)_$(2).txt: $(3) $(IOARENA) $(lastword $(MAKEFILE_LIST))
+	@echo '  RUNNING ioarena for $1/$2...'
+	$(QUIET)(export LD_LIBRARY_PATH="./:$$$${LD_LIBRARY_PATH}"; \
+		ldd $(IOARENA) | grep -i $(1) && \
+		$(IOARENA) -D $(1) -B batch -m $(BENCH_CRUD_MODE) -n $(2) \
+			| tee $$@ | grep throughput | $(SED) 's/throughput/batch×N/' && \
+		$(IOARENA) -D $(1) -B crud -m $(BENCH_CRUD_MODE) -n $(2) \
+			| tee -a $$@ | grep throughput | $(SED) 's/throughput/   crud/' && \
+		$(IOARENA) -D $(1) -B iterate,get,iterate,get,iterate -m $(BENCH_CRUD_MODE) -r 4 -n $(2) \
+			| tee -a $$@ | grep throughput | $(SED) '0,/throughput/{s/throughput/iterate/};s/throughput/    get/' && \
+		$(IOARENA) -D $(1) -B delete -m $(BENCH_CRUD_MODE) -n $(2) \
+			| tee -a $$@ | grep throughput | $(SED) 's/throughput/ delete/' && \
+	true) || mv -f $$@ $$@.error
+
+endef
+
+
+$(eval $(call bench-rule,mdbx,$(NN),libmdbx.$(SO_SUFFIX)))
+
+$(eval $(call bench-rule,sophia,$(NN)))
+$(eval $(call bench-rule,leveldb,$(NN)))
+$(eval $(call bench-rule,rocksdb,$(NN)))
+$(eval $(call bench-rule,wiredtiger,$(NN)))
+$(eval $(call bench-rule,forestdb,$(NN)))
+$(eval $(call bench-rule,lmdb,$(NN)))
+$(eval $(call bench-rule,nessdb,$(NN)))
+$(eval $(call bench-rule,sqlite3,$(NN)))
+$(eval $(call bench-rule,ejdb,$(NN)))
+$(eval $(call bench-rule,vedisdb,$(NN)))
+$(eval $(call bench-rule,dummy,$(NN)))
+bench: bench-mdbx_$(NN).txt
+bench-quartet: bench-mdbx_$(NN).txt bench-lmdb_$(NN).txt bench-rocksdb_$(NN).txt bench-wiredtiger_$(NN).txt
+bench-triplet: bench-mdbx_$(NN).txt bench-lmdb_$(NN).txt bench-sqlite3_$(NN).txt
+bench-couple: bench-mdbx_$(NN).txt bench-lmdb_$(NN).txt
+
+# $(eval $(call bench-rule,debug,10))
+# .PHONY: bench-debug
+# bench-debug: bench-debug_10.txt
+
+endif

LICENSE

@@ -1,47 +1,177 @@
-The OpenLDAP Public License
-  Version 2.8, 17 August 2003
 
-Redistribution and use of this software and associated documentation
-("Software"), with or without modification, are permitted provided
-that the following conditions are met:
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
 
-1. Redistributions in source form must retain copyright statements
-   and notices,
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
 
-2. Redistributions in binary form must reproduce applicable copyright
-   statements and notices, this list of conditions, and the following
-   disclaimer in the documentation and/or other materials provided
-   with the distribution, and
+   1. Definitions.
 
-3. Redistributions must contain a verbatim copy of this document.
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
 
-The OpenLDAP Foundation may revise this license from time to time.
-Each revision is distinguished by a version number.  You may use
-this Software under terms of this license revision or under the
-terms of any subsequent revision of the license.
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
 
-THIS SOFTWARE IS PROVIDED BY THE OPENLDAP FOUNDATION AND ITS
-CONTRIBUTORS ``AS IS'' AND ANY EXPRESSED OR IMPLIED WARRANTIES,
-INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY
-AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT
-SHALL THE OPENLDAP FOUNDATION, ITS CONTRIBUTORS, OR THE AUTHOR(S)
-OR OWNER(S) OF THE SOFTWARE BE LIABLE FOR ANY DIRECT, INDIRECT,
-INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING,
-BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
-LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER
-CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
-LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN
-ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
-POSSIBILITY OF SUCH DAMAGE.
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
 
-The names of the authors and copyright holders must not be used in
-advertising or otherwise to promote the sale, use or other dealing
-in this Software without specific, written prior permission.  Title
-to copyright in this Software shall at all times remain with copyright
-holders.
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
 
-OpenLDAP is a registered trademark of the OpenLDAP Foundation.
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
 
-Copyright 1999-2003 The OpenLDAP Foundation, Redwood City,
-California, USA.  All Rights Reserved.  Permission to copy and
-distribute verbatim copies of this document is granted.
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS

Makefile

@@ -1,217 +1,16 @@
-# GNU Makefile for libmdbx, https://github.com/leo-yuriev/libmdbx
+# This is thunk-Makefile for calling GNU Make 3.80 or above
 
-########################################################################
-# Configuration. The compiler options must enable threaded compilation.
-#
-# Preprocessor macros (for XCFLAGS) of interest...
-# Note that the defaults should already be correct for most
-# platforms; you should not need to change any of these.
-# Read their descriptions in mdb.c if you do. There may be
-# other macros of interest. You should read mdb.c
-# before changing any of them.
-#
-
-# install sandbox
-SANDBOX	?=
-
-# install prefixes (inside sandbox)
-prefix	?= /usr/local
-mandir	?= $(prefix)/man
-
-# lib/bin suffix for multiarch/biarch, e.g. '.x86_64'
-suffix	?=
-
-CC	?= gcc
-CXX	?= g++
-CFLAGS	?= -O2 -g3 -Wall -Werror -Wextra -ffunction-sections -fPIC -fvisibility=hidden
-
-XCFLAGS	?= -DNDEBUG=1 -DMDBX_DEBUG=0 -DLIBMDBX_EXPORTS=1
-CFLAGS	+= -D_GNU_SOURCE=1 -std=gnu11 -pthread $(XCFLAGS)
-CXXFLAGS = -std=c++11 $(filter-out -std=gnu11,$(CFLAGS))
-TESTDB	?= $(shell [ -d /dev/shm ] && echo /dev/shm || echo /tmp)/mdbx-test.db
-TESTLOG ?= $(shell [ -d /dev/shm ] && echo /dev/shm || echo /tmp)/mdbx-test.log
-
-# LY: '--no-as-needed,-lrt' for ability to built with modern glibc, but then run with the old
-LDFLAGS	?= -Wl,--gc-sections,-z,relro,-O,--no-as-needed,-lrt
-EXE_LDFLAGS ?= -pthread -lrt
-
-# LY: just for benchmarking
-IOARENA ?= $(shell \
-  (test -x ../ioarena/@BUILD/src/ioarena && echo ../ioarena/@BUILD/src/ioarena) || \
-  (test -x ../../@BUILD/src/ioarena && echo ../../@BUILD/src/ioarena) || \
-  (test -x ../../src/ioarena && echo ../../src/ioarena) || which ioarena)
-NN	?= 25000000
-
-########################################################################
-
-HEADERS		:= mdbx.h
-LIBRARIES	:= libmdbx.a libmdbx.so
-TOOLS		:= mdbx_stat mdbx_copy mdbx_dump mdbx_load mdbx_chk
-MANPAGES	:= mdbx_stat.1 mdbx_copy.1 mdbx_dump.1 mdbx_load.1
-SHELL		:= /bin/bash
-
-CORE_SRC	:= $(filter-out src/lck-windows.c, $(wildcard src/*.c))
-CORE_INC	:= $(wildcard src/*.h)
-CORE_OBJ	:= $(patsubst %.c,%.o,$(CORE_SRC))
-TEST_SRC	:= $(filter-out test/osal-windows.cc, $(wildcard test/*.cc))
-TEST_INC	:= $(wildcard test/*.h)
-TEST_OBJ	:= $(patsubst %.cc,%.o,$(TEST_SRC))
-
-.PHONY: mdbx all install clean check coverage
-
-all: $(LIBRARIES) $(TOOLS) mdbx_test example
-
-mdbx: libmdbx.a libmdbx.so
-
-example: mdbx.h tutorial/sample-mdbx.c libmdbx.so
-	$(CC) $(CFLAGS) -I. tutorial/sample-mdbx.c ./libmdbx.so -o example
-
-tools: $(TOOLS)
-
-install: $(LIBRARIES) $(TOOLS) $(HEADERS)
-	mkdir -p $(SANDBOX)$(prefix)/bin$(suffix) \
-		&& cp -t $(SANDBOX)$(prefix)/bin$(suffix) $(TOOLS) && \
-	mkdir -p $(SANDBOX)$(prefix)/lib$(suffix) \
-		&& cp -t $(SANDBOX)$(prefix)/lib$(suffix) $(LIBRARIES) && \
-	mkdir -p $(SANDBOX)$(prefix)/include \
-		&& cp -t $(SANDBOX)$(prefix)/include $(HEADERS) && \
-	mkdir -p $(SANDBOX)$(mandir)/man1 \
-		&& cp -t $(SANDBOX)$(mandir)/man1 $(MANPAGES)
-
-clean:
-	rm -rf $(TOOLS) mdbx_test @* *.[ao] *.[ls]o *~ tmp.db/* *.gcov *.log *.err src/*.o test/*.o
-
-check:	all
-	rm -f $(TESTDB) $(TESTLOG) && (set -o pipefail; ./mdbx_test --pathname=$(TESTDB) --dont-cleanup-after basic | tee -a $(TESTLOG) | tail -n 42) && ./mdbx_chk -vvn $(TESTDB)
-
-check-singleprocess:	all
-	rm -f $(TESTDB) $(TESTLOG) && (set -o pipefail; ./mdbx_test --pathname=$(TESTDB) --dont-cleanup-after --hill | tee -a $(TESTLOG) | tail -n 42) && ./mdbx_chk -vvn $(TESTDB)
-
-check-fault:	all
-	rm -f $(TESTDB) $(TESTLOG) && (set -o pipefail; ./mdbx_test --pathname=$(TESTDB) --inject-writefault=42 --dump-config --dont-cleanup-after basic | tee -a $(TESTLOG) | tail -n 42) && ./mdbx_chk -vvn $(TESTDB)
-
-define core-rule
-$(patsubst %.c,%.o,$(1)): $(1) $(CORE_INC) mdbx.h Makefile
-	$(CC) $(CFLAGS) -c $(1) -o $$@
-
-endef
-$(foreach file,$(CORE_SRC),$(eval $(call core-rule,$(file))))
-
-define test-rule
-$(patsubst %.cc,%.o,$(1)): $(1) $(TEST_INC) mdbx.h Makefile
-	$(CXX) $(CXXFLAGS) -c $(1) -o $$@
-
-endef
-$(foreach file,$(TEST_SRC),$(eval $(call test-rule,$(file))))
-
-libmdbx.a: $(CORE_OBJ)
-	$(AR) rs $@ $?
-
-libmdbx.so: $(CORE_OBJ)
-	$(CC) $(CFLAGS) -save-temps $^ -pthread -shared $(LDFLAGS) -o $@
-
-mdbx_%:	src/tools/mdbx_%.c libmdbx.a
-	$(CC) $(CFLAGS) $^ $(EXE_LDFLAGS) -o $@
-
-mdbx_test: $(TEST_OBJ) libmdbx.so
-	$(CXX) $(CXXFLAGS) $(TEST_OBJ) -Wl,-rpath . -L . -l mdbx $(EXE_LDFLAGS) -o $@
-
-###############################################################################
-
-ifneq ($(wildcard $(IOARENA)),)
-
-.PHONY: bench clean-bench re-bench
-
-clean-bench:
-	rm -rf bench-*.txt _ioarena/*
-
-re-bench: clean-bench bench
-
-define bench-rule
-bench-$(1)_$(2).txt: $(3) $(IOARENA) Makefile
-	LD_LIBRARY_PATH="./:$$$${LD_LIBRARY_PATH}" \
-		$(IOARENA) -D $(1) -B crud -m nosync -n $(2) \
-		| tee $$@ | grep throughput && \
-	LD_LIBRARY_PATH="./:$$$${LD_LIBRARY_PATH}" \
-		$(IOARENA) -D $(1) -B get,iterate -m sync -r 4 -n $(2) \
-		| tee -a $$@ | grep throughput \
-	|| mv -f $$@ $$@.error
-
-endef
-
-$(eval $(call bench-rule,mdbx,$(NN),libmdbx.so))
-
-$(eval $(call bench-rule,sophia,$(NN)))
-$(eval $(call bench-rule,leveldb,$(NN)))
-$(eval $(call bench-rule,rocksdb,$(NN)))
-$(eval $(call bench-rule,wiredtiger,$(NN)))
-$(eval $(call bench-rule,forestdb,$(NN)))
-$(eval $(call bench-rule,lmdb,$(NN)))
-$(eval $(call bench-rule,nessdb,$(NN)))
-$(eval $(call bench-rule,sqlite3,$(NN)))
-$(eval $(call bench-rule,ejdb,$(NN)))
-$(eval $(call bench-rule,vedisdb,$(NN)))
-$(eval $(call bench-rule,dummy,$(NN)))
-
-$(eval $(call bench-rule,debug,10))
-
-bench: bench-mdbx_$(NN).txt
-
-.PHONY: bench-debug
-
-bench-debug: bench-debug_10.txt
-
-bench-quartet: bench-mdbx_$(NN).txt bench-lmdb_$(NN).txt bench-rocksdb_$(NN).txt bench-wiredtiger_$(NN).txt
-
-endif
-
-###############################################################################
-
-ci-rule = ( CC=$$(which $1); if [ -n "$$CC" ]; then \
-		echo -n "probe by $2 ($$(readlink -f $$(which $$CC))): " && \
-		$(MAKE) clean >$1.log 2>$1.err && \
-		$(MAKE) CC=$$(readlink -f $$CC) XCFLAGS="-UNDEBUG -DMDBX_DEBUG=2 -DLIBMDBX_EXPORTS=1" check 1>$1.log 2>$1.err && echo "OK" \
-			|| ( echo "FAILED"; cat $1.err >&2; exit 1 ); \
-	else echo "no $2 ($1) for probe"; fi; )
-ci:
-	@if [ "$$(readlink -f $$(which $(CC)))" != "$$(readlink -f $$(which gcc || echo /bin/false))" -a \
-		"$$(readlink -f $$(which $(CC)))" != "$$(readlink -f $$(which clang || echo /bin/false))" -a \
-		"$$(readlink -f $$(which $(CC)))" != "$$(readlink -f $$(which icc || echo /bin/false))" ]; then \
-		$(call ci-rule,$(CC),default C compiler); \
-	fi
-	@$(call ci-rule,gcc,GCC)
-	@$(call ci-rule,clang,clang LLVM)
-	@$(call ci-rule,icc,Intel C)
-
-###############################################################################
-
-CROSS_LIST = alpha-linux-gnu-gcc mips-linux-gnu-gcc \
-	powerpc64-linux-gnu-gcc powerpc-linux-gnu-gcc \
-	arm-linux-gnueabihf-gcc aarch64-linux-gnu-gcc
-
-# hppa-linux-gnu-gcc		- don't supported by current qemu release
-# s390x-linux-gnu-gcc		- qemu troubles (hang/abort)
-# sh4-linux-gnu-gcc		- qemu troubles (pread syscall, etc)
-# mips64-linux-gnuabi64-gcc	- qemu troubles (pread syscall, etc)
-# sparc64-linux-gnu-gcc		- qemu troubles (fcntl for F_SETLK/F_GETLK)
-CROSS_LIST_NOQEMU = hppa-linux-gnu-gcc s390x-linux-gnu-gcc \
-	sh4-linux-gnu-gcc mips64-linux-gnuabi64-gcc sparc64-linux-gnu-gcc
-
-cross-gcc:
-	@echo "CORRESPONDING CROSS-COMPILERs ARE REQUIRED."
-	@echo "FOR INSTANCE: apt install g++-aarch64-linux-gnu g++-alpha-linux-gnu g++-arm-linux-gnueabihf g++-hppa-linux-gnu g++-mips-linux-gnu g++-mips64-linux-gnuabi64 g++-powerpc-linux-gnu g++-powerpc64-linux-gnu g++-s390x-linux-gnu g++-sh4-linux-gnu"
-	@for CC in $(CROSS_LIST_NOQEMU) $(CROSS_LIST); do \
-		echo "===================== $$CC"; \
-		$(MAKE) clean && CC=$$CC CXX=$$(echo $$CC | sed 's/-gcc/-g++/') EXE_LDFLAGS=-static $(MAKE) all || exit $$?; \
-	done
-
-#
-# Unfortunately qemu don't provide robust support for futexes.
-# Therefore it is impossible to run full multi-process tests.
-cross-qemu:
-	@echo "CORRESPONDING CROSS-COMPILERs AND QEMUs ARE REQUIRED."
-	@echo "FOR INSTANCE: apt install binfmt-support qemu-user-static qemu-user qemu-system-arm qemu-system-mips qemu-system-misc qemu-system-ppc qemu-system-sparc g++-aarch64-linux-gnu g++-alpha-linux-gnu g++-arm-linux-gnueabihf g++-hppa-linux-gnu g++-mips-linux-gnu g++-mips64-linux-gnuabi64 g++-powerpc-linux-gnu g++-powerpc64-linux-gnu g++-s390x-linux-gnu g++-sh4-linux-gnu"
-	@for CC in $(CROSS_LIST); do \
-		echo "===================== $$CC + qemu"; \
-		$(MAKE) clean && CC=$$CC CXX=$$(echo $$CC | sed 's/-gcc/-g++/') EXE_LDFLAGS=-static $(MAKE) check-singleprocess || exit $$?; \
-	done
+all help options cmake-build ninja \
+clean install install-no-strip install-strip strip tools uninstall \
+bench bench-clean bench-couple bench-quartet bench-triplet re-bench \
+lib libs lib-static lib-shared tools-static \
+libmdbx mdbx mdbx_chk mdbx_copy mdbx_drop mdbx_dump mdbx_load mdbx_stat \
+check dist memcheck cross-gcc cross-qemu doxygen gcc-analyzer reformat \
+release-assets tags build-test mdbx_test \
+smoke smoke-fault smoke-singleprocess smoke-assertion smoke-memcheck \
+test test-assertion test-long test-long-assertion test-ci test-ci-extra \
+test-asan test-leak test-singleprocess test-ubsan test-memcheck:
+	@CC=$(CC) \
+	CXX=`if test -n "$(CXX)" && which "$(CXX)" > /dev/null; then echo "$(CXX)"; elif test -n "$(CCC)" && which "$(CCC)" > /dev/null; then echo "$(CCC)"; else echo "c++"; fi` \
+	`which gmake || which gnumake || echo 'echo "GNU Make 3.80 or above is required"; exit 2;'` \
+		$(MAKEFLAGS) -f GNUmakefile $@

NOTICE

@@ -0,0 +1,39 @@
+libmdbx (aka MDBX) is an extremely fast, compact, powerful, embeddedable,
+transactional key-value storage engine with open-source code. MDBX has a
+specific set of properties and capabilities, focused on creating unique
+lightweight solutions.
+
+Please visit https://libmdbx.dqdkfa.ru for more information, changelog,
+documentation, C++ API description and links to the original git repo
+with the source code. Questions, feedback and suggestions are welcome
+to the Telegram' group https://t.me/libmdbx.
+
+Donations are welcome to the Ethereum/ERC-20 `0xD104d8f8B2dC312aaD74899F83EBf3EEBDC1EA3A`.
+Всё будет хорошо!
+
+Copyright 2015-2025 Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru>
+SPDX-License-Identifier: Apache-2.0
+For notes about the license change, credits and acknowledgments,
+please refer to the COPYRIGHT file within libmdbx source.
+
+---
+
+On 2022-04-15, without any warnings or following explanations, the
+Github administration deleted _libmdbx_, my account and all other
+projects (status 404). A few months later, without any involvement or
+notification from/to me, the projects were restored/opened in the "public
+read-only archive" status from some kind of incomplete backup. I regard
+these actions of Github as malicious sabotage, and I consider the Github
+service itself to have lost trust forever.
+
+As a result of what has happened, I will never, under any circumstances,
+post the primary sources (aka origins) of my projects on Github, or rely
+in any way on the Github infrastructure.
+
+Nevertheless, realizing that it is more convenient for users of
+_libmdbx_ and other my projects to access ones on Github, I do not want
+to restrict their freedom or create inconvenience, and therefore I place
+mirrors (aka mirrors) of such repositories on Github since 2025. At the
+same time, I would like to emphasize once again that these are only
+mirrors that can be frozen, blocked or deleted at any time, as was the
+case in 2022.

README-RU.md

@@ -1,692 +0,0 @@
-libmdbx
-======================================
-**The revised and extended descendant of [Symas LMDB](https://symas.com/lmdb/).**
-
-*The Future will Positive. Всё будет хорошо.*
-[![Build Status](https://travis-ci.org/leo-yuriev/libmdbx.svg?branch=master)](https://travis-ci.org/leo-yuriev/libmdbx)
-[![Build status](https://ci.appveyor.com/api/projects/status/ue94mlopn50dqiqg/branch/master?svg=true)](https://ci.appveyor.com/project/leo-yuriev/libmdbx/branch/master)
-[![Coverity Scan Status](https://scan.coverity.com/projects/12915/badge.svg)](https://scan.coverity.com/projects/reopen-libmdbx)
-
-English version [by Google](https://translate.googleusercontent.com/translate_c?act=url&ie=UTF8&sl=ru&tl=en&u=https://github.com/leo-yuriev/libmdbx/tree/master)
-and [by Yandex](https://translate.yandex.ru/translate?url=https%3A%2F%2Fgithub.com%2FReOpen%2Flibmdbx%2Ftree%2Fmaster&lang=ru-en).
-
-### Project Status
-
-**Сейчас MDBX _активно перерабатывается_** и к середине 2018 ожидается
-большое изменение как API, так и формата базы данных. К сожалению,
-обновление приведет к потере совместимости с предыдущими версиями.
-
-Цель этой революции - обеспечение более четкого надежного API и
-добавление новых функции, а также наделение базы данных новыми
-свойствами.
-
-В настоящее время MDBX предназначена для Linux, а также поддерживает
-Windows (начиная с Windows Server 2008) в качестве дополнительной
-платформы. Поддержка других ОС может быть обеспечена на коммерческой
-основе. Однако такие усовершенствования (т. е. pull-requests) могут быть
-приняты в мейнстрим только в том случае, если будет доступен
-соответствующий публичный и бесплатный сервис непрерывной интеграции
-(aka Continuous Integration).
-
-## Содержание
-- [Обзор](#Обзор)
-  - [Сравнение с другими СУБД](#Сравнение-с-другими-СУБД)
-  - [История & Acknowledgments](#История)
-- [Основные свойства](#Основные-свойства)
-- [Доработки и усовершенствования относительно LMDB](#Доработки-и-усовершенствования-относительно-lmdb)
-- [Недостатки и Компромиссы](#Недостатки-и-Компромиссы)
-  - [Проблема долгих чтений](#Проблема-долгих-чтений)
-  - [Сохранность данных в режиме асинхронной фиксации](#Сохранность-данных-в-режиме-асинхронной-фиксации)
-- [Сравнение производительности](#Сравнение-производительности)
-  - [Интегральная производительность](#Интегральная-производительность)
-  - [Масштабируемость чтения](#Масштабируемость-чтения)
-  - [Синхронная фиксация](#Синхронная-фиксация)
-  - [Отложенная фиксация](#Отложенная-фиксация)
-  - [Асинхронная фиксация](#Асинхронная-фиксация)
-  - [Потребление ресурсов](#Потребление-ресурсов)
-
-
-## Обзор
-_libmdbx_ - это встраиваемый key-value движок хранения со специфическим
-набором свойств и возможностей, ориентированный на создание уникальных
-легковесных решений с предельной производительностью под Linux и
-Windows.
-
-_libmdbx_ позволяет множеству процессов совместно читать и обновлять
-несколько key-value таблиц с соблюдением
-[ACID](https://ru.wikipedia.org/wiki/ACID), при минимальных накладных
-расходах и амортизационной стоимости любых операций Olog(N).
-
-_libmdbx_ обеспечивает
-[serializability](https://en.wikipedia.org/wiki/Serializability)
-изменений и согласованность данных после аварий. При этом транзакции,
-изменяющие данные, никак не мешают операциям чтения и выполняются строго
-последовательно с использованием единственного
-[мьютекса](https://en.wikipedia.org/wiki/Mutual_exclusion).
-
-_libmdbx_ позволяет выполнять операции чтения с гарантиями
-[wait-free](https://en.wikipedia.org/wiki/Non-blocking_algorithm#Wait-freedom),
-параллельно на каждом ядре CPU, без использования атомарных операций
-и/или примитивов синхронизации.
-
-_libmdbx_ не использует
-[LSM](https://en.wikipedia.org/wiki/Log-structured_merge-tree), а
-основан на [B+Tree](https://en.wikipedia.org/wiki/B%2B_tree) с
-[отображением](https://en.wikipedia.org/wiki/Memory-mapped_file) всех
-данных в память, при этом текущая версия не использует
-[WAL](https://en.wikipedia.org/wiki/Write-ahead_logging). Это
-предопределяет многие свойства, в том числе удачные и противопоказанные
-сценарии использования.
-
-
-### Сравнение с другими СУБД
-Ввиду того, что в _libmdbx_ сейчас происходит революция, я посчитал
-лучшим решением ограничится здесь ссылкой на [главу Comparison with
-other databases](https://github.com/coreos/bbolt#comparison-with-other-databases)
-в описании _BoltDB_.
-
-
-### История
-_libmdbx_ является результатом переработки и развития "Lightning
-Memory-Mapped Database", известной под аббревиатурой
-[LMDB](https://en.wikipedia.org/wiki/Lightning_Memory-Mapped_Database).
-Изначально доработка производилась в составе проекта
-[ReOpenLDAP](https://github.com/leo-yuriev/ReOpenLDAP). Примерно за год
-работы внесенные изменения приобрели самостоятельную ценность. Осенью
-2015 доработанный движок был выделен в отдельный проект, который был
-[представлен на конференции Highload++
-2015](http://www.highload.ru/2015/abstracts/1831.html).
-
-В начале 2017 года движок _libmdbx_ получил новый импульс развития,
-благодаря использованию в [Fast Positive
-Tables](https://github.com/leo-yuriev/libfpta), aka ["Позитивные
-Таблицы"](https://github.com/leo-yuriev/libfpta) by [Positive
-Technologies](https://www.ptsecurity.ru).
-
-
-#### Acknowledgments
-Howard Chu (Symas Corporation) - the author of LMDB, from which
-originated the MDBX in 2015.
-
-Martin Hedenfalk <martin@bzero.se> - the author of `btree.c` code, which
-was used for begin development of LMDB.
-
-
-Основные свойства
-=================
-
-_libmdbx_ наследует все ключевые возможности и особенности своего
-прародителя
-[LMDB](https://en.wikipedia.org/wiki/Lightning_Memory-Mapped_Database),
-но с устранением ряда описываемых далее проблем и архитектурных
-недочетов.
-
-1. Данные хранятся в упорядоченном отображении (ordered map), ключи
-всегда отсортированы, поддерживается выборка диапазонов (range lookups).
-
-2. Данные отображается в память каждого работающего с БД процесса. К
-данным и ключам обеспечивается прямой доступ в памяти без необходимости
-их копирования.
-
-3. Транзакции согласно [ACID](https://ru.wikipedia.org/wiki/ACID),
-посредством [MVCC](https://ru.wikipedia.org/wiki/MVCC) и
-[COW](https://ru.wikipedia.org/wiki/%D0%9A%D0%BE%D0%BF%D0%B8%D1%80%D0%BE%D0%B2%D0%B0%D0%BD%D0%B8%D0%B5_%D0%BF%D1%80%D0%B8_%D0%B7%D0%B0%D0%BF%D0%B8%D1%81%D0%B8).
-Изменения строго последовательны и не блокируются чтением, конфликты
-между транзакциями невозможны. При этом гарантируется чтение только
-зафиксированных данных, см [relaxing
-serializability](https://en.wikipedia.org/wiki/Serializability).
-
-4. Чтение и поиск [без
-блокировок](https://ru.wikipedia.org/wiki/%D0%9D%D0%B5%D0%B1%D0%BB%D0%BE%D0%BA%D0%B8%D1%80%D1%83%D1%8E%D1%89%D0%B0%D1%8F_%D1%81%D0%B8%D0%BD%D1%85%D1%80%D0%BE%D0%BD%D0%B8%D0%B7%D0%B0%D1%86%D0%B8%D1%8F),
-без [атомарных
-операций](https://ru.wikipedia.org/wiki/%D0%90%D1%82%D0%BE%D0%BC%D0%B0%D1%80%D0%BD%D0%B0%D1%8F_%D0%BE%D0%BF%D0%B5%D1%80%D0%B0%D1%86%D0%B8%D1%8F).
-Читатели не блокируются операциями записи и не конкурируют между собой,
-чтение масштабируется линейно по ядрам CPU.
-   > Для точности следует отметить, что "подключение к БД" (старт первой
-   > читающей транзакции в потоке) и "отключение от БД" (закрытие БД или
-   > завершение потока) требуют краткосрочного захвата блокировки для
-   > регистрации/дерегистрации текущего потока в "таблице читателей".
-
-5. Эффективное хранение дубликатов (ключей с несколькими значениями),
-без дублирования ключей, с сортировкой значений, в том числе
-целочисленных (для вторичных индексов).
-
-6. Эффективная поддержка коротких ключей фиксированной длины, в том
-числе целочисленных.
-
-7. Амортизационная стоимость любой операции Olog(N),
-[WAF](https://en.wikipedia.org/wiki/Write_amplification) (Write
-Amplification Factor) и RAF (Read Amplification Factor) также Olog(N).
-
-8. Нет [WAL](https://en.wikipedia.org/wiki/Write-ahead_logging) и
-журнала транзакций, после сбоев не требуется восстановление. Не
-требуется компактификация или какое-либо периодическое обслуживание.
-Поддерживается резервное копирование "по горячему", на работающей БД без
-приостановки изменения данных.
-
-9. Отсутствует какое-либо внутреннее управление памятью или
-кэшированием. Всё необходимое штатно выполняет ядро ОС.
-
-
-Доработки и усовершенствования относительно LMDB
-================================================
-
-1. Утилита `mdbx_chk` для проверки целостности структуры БД.
-
-2. Автоматическое динамическое управление размером БД согласно
-параметрам задаваемым функцией `mdbx_env_set_geometry()`, включая шаг
-приращения и порог уменьшения размера БД, а также выбор размера
-страницы. Соответственно, это позволяет снизить фрагментированность
-файла БД на диске и освободить место, в том числе в **Windows**.
-
-3. Автоматическая без-затратная компактификация БД путем возврата
-освобождающихся страниц в область нераспределенного резерва в конце
-файла данных. При этом уменьшается количество страниц находящихся в
-памяти и участвующих в в обмене с диском.
-
-4. Поддержка ключей и значений нулевой длины, включая сортированные
-дубликаты.
-
-5. Возможность связать с каждой завершаемой транзакцией до 3
-дополнительных маркеров посредством `mdbx_canary_put()`, и прочитать их
-в транзакции чтения посредством `mdbx_canary_get()`.
-
-6. Возможность посредством `mdbx_replace()` обновить или удалить запись
-с получением предыдущего значения данных, а также адресно изменить
-конкретное multi-значение.
-
-7. Режим `LIFO RECLAIM`.
-
-    Для повторного использования выбираются не самые старые, а
-    самые новые страницы из доступных. За счет этого цикл
-    использования страниц всегда имеет минимальную длину и не
-    зависит от общего числа выделенных страниц.
-
-    В результате механизмы кэширования и обратной записи работают с
-    максимально возможной эффективностью. В случае использования
-    контроллера дисков или системы хранения с
-    [BBWC](https://en.wikipedia.org/wiki/BBWC) возможно
-    многократное увеличение производительности по записи
-    (обновлению данных).
-
-8. Генерация последовательностей посредством `mdbx_dbi_sequence()`.
-
-9. Обработчик `OOM-KICK`.
-
-    Посредством `mdbx_env_set_oomfunc()` может быть установлен
-    внешний обработчик (callback), который будет вызван при
-    исчерпании свободных страниц по причине долгой операцией чтения
-    на фоне интенсивного изменения данных.
-    Обработчику будет передан PID и pthread_id виновника.
-    В свою очередь обработчик может предпринять одно из действий:
-
-    * нейтрализовать виновника (отправить сигнал kill #9), если
-      долгое чтение выполняется сторонним процессом;
-
-    * отменить или перезапустить проблемную операцию чтения, если
-      операция выполняется одним из потоков текущего процесса;
-
-    * подождать некоторое время, в расчете на то, что проблемная операция
-      чтения будет штатно завершена;
-
-    * прервать текущую операцию изменения данных с возвратом кода
-      ошибки.
-
-10. Возможность открыть БД в эксклюзивном режиме посредством флага
-`MDBX_EXCLUSIVE`.
-
-11. Возможность получить отставание текущей транзакции чтения от
-последней версии данных в БД посредством `mdbx_txn_straggler()`.
-
-12. Возможность явно запросить обновление существующей записи, без
-создания новой посредством флажка `MDBX_CURRENT` для `mdbx_put()`.
-
-13. Исправленный вариант `mdbx_cursor_count()`, возвращающий корректное
-количество дубликатов для всех типов таблиц и любого положения курсора.
-
-14. Возможность получить посредством `mdbx_env_info()` дополнительную
-информацию, включая номер самой старой версии БД (снимка данных),
-который используется одним из читателей.
-
-15. Функция `mdbx_del()` не игнорирует дополнительный (уточняющий)
-аргумент `data` для таблиц без дубликатов (без флажка `MDBX_DUPSORT`), а
-при его ненулевом значении всегда использует его для сверки с удаляемой
-записью.
-
-16. Возможность открыть dbi-таблицу, одновременно с установкой
-компараторов для ключей и данных, посредством `mdbx_dbi_open_ex()`.
-
-17. Возможность посредством `mdbx_is_dirty()` определить находятся ли
-некоторый ключ или данные в "грязной" странице БД. Таким образом,
-избегая лишнего копирования данных перед выполнением модифицирующих
-операций (значения, размещенные в "грязных" страницах, могут быть
-перезаписаны при изменениях, иначе они будут неизменны).
-
-18. Корректное обновление текущей записи, в том числе сортированного
-дубликата, при использовании режима `MDBX_CURRENT` в
-`mdbx_cursor_put()`.
-
-19. Возможность узнать есть ли за текущей позицией курсора строка данных
-посредством `mdbx_cursor_eof()`.
-
-20. Дополнительный код ошибки `MDBX_EMULTIVAL`, который возвращается из
-`mdbx_put()` и `mdbx_replace()` при попытке выполнить неоднозначное
-обновление или удаления одного из нескольких значений с одним ключом.
-
-21. Возможность посредством `mdbx_get_ex()` получить значение по
-заданному ключу, одновременно с количеством дубликатов.
-
-22. Наличие функций `mdbx_cursor_on_first()` и `mdbx_cursor_on_last()`,
-которые позволяют быстро выяснить стоит ли курсор на первой/последней
-позиции.
-
-23. Возможность автоматического формирования контрольных точек (сброса
-данных на диск) при накоплении заданного объёма изменений,
-устанавливаемого функцией `mdbx_env_set_syncbytes()`.
-
-24. Управление отладкой и получение отладочных сообщений посредством
-`mdbx_setup_debug()`.
-
-25. Функция `mdbx_env_pgwalk()` для обхода всех страниц БД.
-
-26. Три мета-страницы вместо двух, что позволяет гарантированно
-консистентно обновлять слабые контрольные точки фиксации без риска
-повредить крайнюю сильную точку фиксации.
-
-27. Гарантия сохранности БД в режиме `WRITEMAP+MAPSYNC`.
-  > В текущей версии _libmdbx_ вам предоставляется выбор между безопасным
-  > режимом (по умолчанию) асинхронной фиксации, и режимом `UTTERLY_NOSYNC`
-  > когда при системной аварии есть шанс полного разрушения БД как в LMDB.
-  > Для подробностей смотрите раздел
-  > [Сохранность данных в режиме асинхронной фиксации](#Сохранность-данных-в-режиме-асинхронной-фиксации).
-
-28. Возможность закрыть БД в "грязном" состоянии (без сброса данных и
-формирования сильной точки фиксации) посредством `mdbx_env_close_ex()`.
-
-29. При завершении читающих транзакций, открытые в них DBI-хендлы не
-закрываются и не теряются при завершении таких транзакций посредством
-`mdbx_txn_abort()` или `mdbx_txn_reset()`. Что позволяет избавится от ряда
-сложно обнаруживаемых ошибок.
-
-30. Все курсоры, как в транзакциях только для чтения, так и в пишущих,
-могут быть переиспользованы посредством `mdbx_cursor_renew()` и ДОЛЖНЫ
-ОСВОБОЖДАТЬСЯ ЯВНО.
-  >
-  > ## _ВАЖНО_, Обратите внимание!
-  >
-  > Это единственное изменение в API, которое значимо меняет
-  > семантику управления курсорами и может приводить к утечкам
-  > памяти. Следует отметить, что это изменение вынужденно.
-  > Так устраняется неоднозначность с массой тяжких последствий:
-  >
-  >  - обращение к уже освобожденной памяти;
-  >  - попытки повторного освобождения памяти;
-  >  - повреждение памяти и ошибки сегментации.
-
---------------------------------------------------------------------------------
-
-## Недостатки и Компромиссы
-
-1. Единовременно может выполняться не более одной транзакция изменения данных
-   (один писатель). Зато все изменения всегда последовательны, не может быть
-   конфликтов или логических ошибок при откате транзакций.
-
-2. Отсутствие [WAL](https://en.wikipedia.org/wiki/Write-ahead_logging)
-   обуславливает относительно большой
-   [WAF](https://en.wikipedia.org/wiki/Write_amplification) (Write
-   Amplification Factor). Поэтому фиксация изменений на диске может быть
-   достаточно дорогой и являться главным ограничением производительности
-   при интенсивном изменении данных.
-   > В качестве компромисса _libmdbx_ предлагает несколько режимов ленивой
-   > и/или периодической фиксации. В том числе режим `MAPASYNC`, при котором
-   > изменения происходят только в памяти и асинхронно фиксируются на диске
-   > ядром ОС.
-   >
-   > Однако, следует воспринимать это свойство аккуратно и взвешенно.
-   > Например, полная фиксация транзакции в БД с журналом потребует минимум 2
-   > IOPS (скорее всего 3-4) из-за накладных расходов в файловой системе. В
-   > _libmdbx_ фиксация транзакции также требует от 2 IOPS. Однако, в БД с
-   > журналом кол-во IOPS будет меняться в зависимости от файловой системы,
-   > но не от кол-ва записей или их объема. Тогда как в _libmdbx_ кол-во
-   > будет расти логарифмически от кол-ва записей/строк в БД (по высоте
-   > b+tree).
-
-3. [COW](https://ru.wikipedia.org/wiki/%D0%9A%D0%BE%D0%BF%D0%B8%D1%80%D0%BE%D0%B2%D0%B0%D0%BD%D0%B8%D0%B5_%D0%BF%D1%80%D0%B8_%D0%B7%D0%B0%D0%BF%D0%B8%D1%81%D0%B8)
-   для реализации [MVCC](https://ru.wikipedia.org/wiki/MVCC) выполняется на
-   уровне страниц в [B+
-   дереве](https://ru.wikipedia.org/wiki/B-%D0%B4%D0%B5%D1%80%D0%B5%D0%B2%D0%BE).
-   Поэтому изменение данных амортизационно требует копирования Olog(N)
-   страниц, что расходует [пропускную способность оперативной
-   памяти](https://en.wikipedia.org/wiki/Memory_bandwidth) и является
-   основным ограничителем производительности в режиме `MAPASYNC`.
-   > Этот недостаток неустраним, тем не менее следует дать некоторые пояснения.
-   > Дело в том, что фиксация изменений на диске потребует гораздо более
-   > значительного копирования данных в памяти и массы других затратных операций.
-   > Поэтому обусловленное этим недостатком падение производительности становится
-   > заметным только при отказе от фиксации изменений на диске.
-   > Соответственно, корректнее сказать, что _libmdbx_ позволяет
-   > получить персистентность ценой минимального падения производительности.
-   > Если же нет необходимости оперативно сохранять данные, то логичнее
-   > использовать `std::map`.
-
-4. В _LMDB_ существует проблема долгих чтений (приостановленных читателей),
-   которая приводит к деградации производительности и переполнению БД.
-   > В _libmdbx_ предложены средства для предотвращения, быстрого выхода из
-   > некомфортной ситуации и устранения её последствий. Подробности ниже.
-
-5. В _LMDB_ есть вероятность разрушения БД в режиме `WRITEMAP+MAPASYNC`.
-   В _libmdbx_ для `WRITEMAP+MAPASYNC` гарантируется как сохранность базы,
-   так и согласованность данных.
-   > Дополнительно, в качестве альтернативы, предложен режим `UTTERLY_NOSYNC`.
-   > Подробности ниже.
-
-
-#### Проблема долгих чтений
-*Следует отметить*, что проблема "сборки мусора" так или иначе
-существует во всех СУБД (Vacuum в PostgreSQL). Однако в случае _libmdbx_
-и LMDB она проявляется более остро, прежде всего из-за высокой
-производительности, а также из-за намеренного упрощения внутренних
-механизмов ради производительности.
-
-Понимание проблемы требует некоторых пояснений, которые
-изложены ниже, но могут быть сложны для быстрого восприятия.
-Поэтому, тезисно:
-
-* Изменение данных на фоне долгой операции чтения может
-  приводить к исчерпанию места в БД.
-
-* После чего любая попытка обновить данные будет приводить к
-  ошибке `MAP_FULL` до завершения долгой операции чтения.
-
-* Характерными примерами долгих чтений являются горячее
-  резервное копирования и отладка клиентского приложения при
-  активной транзакции чтения.
-
-* В оригинальной _LMDB_ после этого будет наблюдаться
-  устойчивая деградация производительности всех механизмов
-  обратной записи на диск (в I/O контроллере, в гипервизоре,
-  в ядре ОС).
-
-* В _libmdbx_ предусмотрен механизм аварийного прерывания таких
-  операций, а также режим `LIFO RECLAIM` устраняющий последующую
-  деградацию производительности.
-
-Операции чтения выполняются в контексте снимка данных (версии
-БД), который был актуальным на момент старта транзакции чтения. Такой
-читаемый снимок поддерживается неизменным до завершения операции. В свою
-очередь, это не позволяет повторно использовать страницы БД в
-последующих версиях (снимках БД).
-
-Другими словами, если обновление данных выполняется на фоне долгой
-операции чтения, то вместо повторного использования "старых" ненужных
-страниц будут выделяться новые, так как "старые" страницы составляют
-снимок БД, который еще используется долгой операцией чтения.
-
-В результате, при интенсивном изменении данных и достаточно длительной
-операции чтения, в БД могут быть исчерпаны свободные страницы, что не
-позволит создавать новые снимки/версии БД. Такая ситуация будет
-сохраняться до завершения операции чтения, которая использует старый
-снимок данных и препятствует повторному использованию страниц БД.
-
-Однако, на этом проблемы не заканчиваются. После описанной ситуации, все
-дополнительные страницы, которые были выделены пока переработка старых
-была невозможна, будут участвовать в цикле выделения/освобождения до
-конца жизни экземпляра БД. В оригинальной _LMDB_ этот цикл использования
-страниц работает по принципу [FIFO](https://ru.wikipedia.org/wiki/FIFO).
-Поэтому увеличение количества циркулирующий страниц, с точки зрения
-механизмов кэширования и/или обратной записи, выглядит как увеличение
-рабочего набор данных. Проще говоря, однократное попадание в ситуацию
-"уснувшего читателя" приводит к устойчивому эффекту вымывания I/O кэша
-при всех последующих изменениях данных.
-
-Для устранения описанных проблемы в _libmdbx_ сделаны существенные
-доработки, подробности ниже. Иллюстрации к проблеме "долгих чтений"
-можно найти в [слайдах презентации](http://www.slideshare.net/leoyuriev/lmdb).
-
-Там же приведен пример количественной оценки прироста производительности
-за счет эффективной работы [BBWC](https://en.wikipedia.org/wiki/BBWC)
-при включении `LIFO RECLAIM` в _libmdbx_.
-
-#### Сохранность данных в режиме асинхронной фиксации
-При работе в режиме `WRITEMAP+MAPSYNC` запись измененных страниц
-выполняется ядром ОС, что имеет ряд преимуществ. Так например, при крахе
-приложения, ядро ОС сохранит все изменения.
-
-Однако, при аварийном отключении питания или сбое в ядре ОС, на диске
-может быть сохранена только часть измененных страниц БД. При этом с
-большой вероятностью может оказаться, что будут сохранены мета-страницы
-со ссылками на страницы с новыми версиями данных, но не сами новые
-данные. В этом случае БД будет безвозвратна разрушена, даже если до
-аварии производилась полная синхронизация данных (посредством
-`mdbx_env_sync()`).
-
-В _libmdbx_ эта проблема устранена путем полной переработки
-пути записи данных:
-
-* В режиме `WRITEMAP+MAPSYNC` _libmdbx_ не обновляет
-  мета-страницы непосредственно, а поддерживает их теневые копии
-  с переносом изменений после фиксации данных.
-
-* При завершении транзакций, в зависимости от состояния
-  синхронности данных между диском и оперативной памятью,
-  _libmdbx_ помечает точки фиксации либо как сильные (strong),
-  либо как слабые (weak). Так например, в режиме
-  `WRITEMAP+MAPSYNC` завершаемые транзакции помечаются как
-  слабые, а при явной синхронизации данных - как сильные.
-
-* В _libmdbx_ поддерживается не две, а три отдельные мета-страницы.
-  Это позволяет выполнять фиксацию транзакций с формированием как
-  сильной, так и слабой точки фиксации, без потери двух предыдущих
-  точек фиксации (из которых одна может быть сильной, а вторая слабой).
-  В результате, _libmdbx_ позволяет в произвольном порядке чередовать
-  сильные и слабые точки фиксации без нарушения соответствующих
-  гарантий в случае неожиданной системной аварии во время фиксации.
-
-* При открытии БД выполняется автоматический откат к последней
-  сильной фиксации. Этим обеспечивается гарантия сохранности БД.
-
-Такая гарантия надежности не дается бесплатно. Для сохранности данных,
-страницы, формирующие крайний снимок с сильной фиксацией, не должны
-повторно использоваться (перезаписываться) до формирования следующей
-сильной точки фиксации. Таким образом, крайняя точка фиксации создает
-описанный выше эффект "долгого чтения". Разница же здесь в том, что при
-исчерпании свободных страниц ситуация будет автоматически исправлена,
-посредством записи изменений на диск и формирования новой сильной точки
-фиксации.
-
-Таким образом, в режиме безопасной асинхронной фиксации _libmdbx_ будет
-всегда использовать новые страницы до исчерпания места в БД или до
-явного формирования сильной точки фиксации посредством
-`mdbx_env_sync()`. При этом суммарный трафик записи на диск будет
-примерно такой же, как если бы отдельно фиксировалась каждая транзакция.
-
-В текущей версии _libmdbx_ вам предоставляется выбор между безопасным
-режимом (по умолчанию) асинхронной фиксации, и режимом `UTTERLY_NOSYNC`
-когда при системной аварии есть шанс полного разрушения БД как в LMDB.
-
-В последующих версиях _libmdbx_ будут предусмотрены средства для
-асинхронной записи данных на диск с автоматическим формированием сильных
-точек фиксации.
-
---------------------------------------------------------------------------------
-
-Сравнение производительности
-============================
-
-Все представленные ниже данные получены многократным прогоном тестов на
-ноутбуке Lenovo Carbon-2, i7-4600U 2.1 ГГц, 8 Гб ОЗУ, с SSD-диском
-SAMSUNG MZNTD512HAGL-000L1 (DXT23L0Q) 512 Гб.
-
-Исходный код бенчмарка [_IOArena_](https://github.com/pmwkaa/ioarena) и
-сценарии тестирования [доступны на
-github](https://github.com/pmwkaa/ioarena/tree/HL%2B%2B2015).
-
---------------------------------------------------------------------------------
-
-### Интегральная производительность
-
-Показана соотнесенная сумма ключевых показателей производительности в трёх
-бенчмарках:
-
-   - Чтение/Поиск на машине с 4-мя процессорами;
-
-   - Транзакции с [CRUD](https://ru.wikipedia.org/wiki/CRUD)-операциями
-     (вставка, чтение, обновление, удаление) в режиме **синхронной фиксации**
-     данных (fdatasync при завершении каждой транзакции или аналог);
-
-   - Транзакции с [CRUD](https://ru.wikipedia.org/wiki/CRUD)-операциями
-     (вставка, чтение, обновление, удаление) в режиме **отложенной фиксации**
-     данных (отложенная запись посредством файловой систем или аналог);
-
-*Бенчмарк в режиме асинхронной записи не включен по двум причинам:*
-
-  1. Такое сравнение не совсем правомочно, его следует делать с движками
-     ориентированными на хранение данных в памяти ([Tarantool](https://tarantool.io/), [Redis](https://redis.io/)).
-
-  2. Превосходство libmdbx становится еще более подавляющим, что мешает
-     восприятию информации.
-
-![Comparison #1: Integral Performance](https://raw.githubusercontent.com/wiki/leo-yuriev/libmdbx/img/perf-slide-1.png)
-
---------------------------------------------------------------------------------
-
-### Масштабируемость чтения
-
-Для каждого движка показана суммарная производительность при
-одновременном выполнении запросов чтения/поиска в 1-2-4-8 потоков на
-машине с 4-мя физическими процессорами.
-
-![Comparison #2: Read Scalability](https://raw.githubusercontent.com/wiki/leo-yuriev/libmdbx/img/perf-slide-2.png)
-
---------------------------------------------------------------------------------
-
-### Синхронная фиксация
-
- - Линейная шкала слева и темные прямоугольники соответствуют количеству
-   транзакций в секунду, усредненному за всё время теста.
-
- - Логарифмическая шкала справа и желтые интервальные отрезки
-   соответствуют времени выполнения транзакций. При этом каждый отрезок
-   показывает минимальное и максимальное время, затраченное на выполнение
-   транзакций, а крестиком отмечено среднеквадратичное значение.
-
-Выполняется **10.000 транзакций в режиме синхронной фиксации данных** на
-диске. При этом требуется гарантия, что при аварийном выключении питания
-(или другом подобном сбое) все данные будут консистентны и полностью
-соответствовать последней завершенной транзакции. В _libmdbx_ в этом
-режиме при фиксации каждой транзакции выполняется системный вызов
-[fdatasync](https://linux.die.net/man/2/fdatasync).
-
-В каждой транзакции выполняется комбинированная CRUD-операция (две
-вставки, одно чтение, одно обновление, одно удаление). Бенчмарк стартует
-на пустой базе, а при завершении, в результате выполняемых действий, в
-базе насчитывается 10.000 небольших key-value записей.
-
-![Comparison #3: Sync-write mode](https://raw.githubusercontent.com/wiki/leo-yuriev/libmdbx/img/perf-slide-3.png)
-
---------------------------------------------------------------------------------
-
-### Отложенная фиксация
-
- - Линейная шкала слева и темные прямоугольники соответствуют количеству
-   транзакций в секунду, усредненному за всё время теста.
-
- - Логарифмическая шкала справа и желтые интервальные отрезки
-   соответствуют времени выполнения транзакций. При этом каждый отрезок
-   показывает минимальное и максимальное время, затраченное на выполнение
-   транзакций, а крестиком отмечено среднеквадратичное значение.
-
-Выполняется **100.000 транзакций в режиме отложенной фиксации данных**
-на диске. При этом требуется гарантия, что при аварийном выключении
-питания (или другом подобном сбое) все данные будут консистентны на
-момент завершения одной из транзакций, но допускается потеря изменений
-из некоторого количества последних транзакций, что для многих движков
-предполагает включение
-[WAL](https://en.wikipedia.org/wiki/Write-ahead_logging) (write-ahead
-logging) либо журнала транзакций, который в свою очередь опирается на
-гарантию упорядоченности данных в журналируемой файловой системе.
-_libmdbx_ при этом не ведет WAL, а передает весь контроль файловой
-системе и ядру ОС.
-
-В каждой транзакции выполняется комбинированная CRUD-операция (две
-вставки, одно чтение, одно обновление, одно удаление). Бенчмарк стартует
-на пустой базе, а при завершении, в результате выполняемых действий, в
-базе насчитывается 100.000 небольших key-value записей.
-
-![Comparison #4: Lazy-write mode](https://raw.githubusercontent.com/wiki/leo-yuriev/libmdbx/img/perf-slide-4.png)
-
---------------------------------------------------------------------------------
-
-### Асинхронная фиксация
-
- - Линейная шкала слева и темные прямоугольники соответствуют количеству
-   транзакций в секунду, усредненному за всё время теста.
-
- - Логарифмическая шкала справа и желтые интервальные отрезки
-   соответствуют времени выполнения транзакций. При этом каждый отрезок
-   показывает минимальное и максимальное время, затраченное на выполнение
-   транзакций, а крестиком отмечено среднеквадратичное значение.
-
-Выполняется **1.000.000 транзакций в режиме асинхронной фиксации
-данных** на диске. При этом требуется гарантия, что при аварийном
-выключении питания (или другом подобном сбое) все данные будут
-консистентны на момент завершения одной из транзакций, но допускается
-потеря изменений из значительного количества последних транзакций. Во
-всех движках при этом включался режим предполагающий минимальную
-нагрузку на диск по записи, и соответственно минимальную гарантию
-сохранности данных. В _libmdbx_ при этом используется режим асинхронной
-записи измененных страниц на диск посредством ядра ОС и системного
-вызова [msync(MS_ASYNC)](https://linux.die.net/man/2/msync).
-
-В каждой транзакции выполняется комбинированная CRUD-операция (две
-вставки, одно чтение, одно обновление, одно удаление). Бенчмарк стартует
-на пустой базе, а при завершении, в результате выполняемых действий, в
-базе насчитывается 10.000 небольших key-value записей.
-
-![Comparison #5: Async-write mode](https://raw.githubusercontent.com/wiki/leo-yuriev/libmdbx/img/perf-slide-5.png)
-
---------------------------------------------------------------------------------
-
-### Потребление ресурсов
-
-Показана соотнесенная сумма использованных ресурсов в ходе бенчмарка в
-режиме отложенной фиксации:
-
-  - суммарное количество операций ввода-вывода (IOPS), как записи, так и
-  чтения.
-
-  - суммарное затраченное время процессора, как в режиме пользовательских
-  процессов, так и в режиме ядра ОС.
-
-  - использованное место на диске при завершении теста, после закрытия БД
-  из тестирующего процесса, но без ожидания всех внутренних операций
-  обслуживания (компактификации LSM и т.п.).
-
-Движок _ForestDB_ был исключен при оформлении результатов, так как
-относительно конкурентов многократно превысил потребление каждого из
-ресурсов (потратил процессорное время на генерацию IOPS для заполнения
-диска), что не позволяло наглядно сравнить показатели остальных движков
-на одной диаграмме.
-
-Все данные собирались посредством системного вызова
-[getrusage()](http://man7.org/linux/man-pages/man2/getrusage.2.html) и
-сканированием директорий с данными.
-
-![Comparison #6: Cost comparison](https://raw.githubusercontent.com/wiki/leo-yuriev/libmdbx/img/perf-slide-6.png)
-
---------------------------------------------------------------------------------
-
-```
-$ objdump -f -h -j .text libmdbx.so
-
-libmdbx.so:     file format elf64-x86-64
-architecture: i386:x86-64, flags 0x00000150:
-HAS_SYMS, DYNAMIC, D_PAGED
-start address 0x000030e0
-
-Sections:
-Idx Name          Size      VMA               LMA               File off  Algn
- 11 .text         00014d84  00000000000030e0  00000000000030e0  000030e0  2**4
-                  CONTENTS, ALLOC, LOAD, READONLY, CODE
-
-```

TODO.md

@@ -0,0 +1,44 @@
+TODO
+----
+
+ - [SWIG](https://www.swig.org/).
+ - Параллельная lto-сборка с устранением предупреждений.
+ - Интеграция c DTrace и аналогами.
+ - Новый стиль обработки ошибок с записью "трассы" и причин.
+ - Формирование отладочной информации посредством gdb.
+ - Поддержка WASM.
+ - Ранняя/не-отложенная очистка GC.
+ - Явная и автоматические уплотнение/дефрагментация.
+ - Нелинейная обработка GC.
+ - Перевести курсоры на двусвязный список вместо односвязного.
+ - Внутри `txn_renew()` вынести проверку когерентности mmap за/после изменение размера.
+ - [Migration guide from LMDB to MDBX](https://libmdbx.dqdkfa.ru/dead-github/issues/199).
+ - [Support for RAW devices](https://libmdbx.dqdkfa.ru/dead-github/issues/124).
+ - [Support MessagePack for Keys & Values](https://libmdbx.dqdkfa.ru/dead-github/issues/115).
+ - Packages for [Astra Linux](https://astralinux.ru/), [ALT Linux](https://www.altlinux.org/), [ROSA Linux](https://www.rosalinux.ru/), etc.
+
+Done
+----
+
+ - Рефакторинг gc-get/gc-put c переходом на "интервальные" списки.
+ - [Engage new terminology](https://libmdbx.dqdkfa.ru/dead-github/issues/137).
+ - [More flexible support of asynchronous runtime/framework(s)](https://libmdbx.dqdkfa.ru/dead-github/issues/200).
+ - [Move most of `mdbx_chk` functional to the library API](https://libmdbx.dqdkfa.ru/dead-github/issues/204).
+ - [Simple careful mode for working with corrupted DB](https://libmdbx.dqdkfa.ru/dead-github/issues/223).
+ - [Engage an "overlapped I/O" on Windows](https://libmdbx.dqdkfa.ru/dead-github/issues/224).
+ - [Large/Overflow pages accounting for dirty-room](https://libmdbx.dqdkfa.ru/dead-github/issues/192).
+ - [Get rid of dirty-pages list in MDBX_WRITEMAP mode](https://libmdbx.dqdkfa.ru/dead-github/issues/193).
+
+Cancelled
+--------
+
+ - [Replace SRW-lock on Windows to allow shrink DB with `MDBX_NOSTICKYTHREADS` option](https://libmdbx.dqdkfa.ru/dead-github/issues/210).
+   Доработка не может быть реализована, так как замена SRW-блокировки
+   лишает лишь предварительную проблему, но не главную. На Windows
+   уменьшение размера отображенного в память файла не поддерживается ядром
+   ОС. Для этого необходимо снять отображение, изменить размер файла и
+   затем отобразить обратно. В свою очередь, для это необходимо
+   приостановить работающие с БД потоки выполняющие транзакции чтения, либо
+   готовые к такому выполнению. Но в режиме MDBX_NOSTICKYTHREADS нет
+   возможности отслеживать работающие с БД потоки, а приостановка всех
+   потоков неприемлема для большинства приложений.

appveyor.yml

@@ -1,54 +0,0 @@
-version: 0.2.0.{build}
-
-environment:
-  matrix:
-    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2017
-      TOOLSET: v141
-    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2015
-      TOOLSET: v140
-    - APPVEYOR_BUILD_WORKER_IMAGE: Visual Studio 2013
-      TOOLSET: v120
-
-branches:
-  except:
-    - coverity_scan
-
-configuration:
-- Debug
-- Release
-
-platform:
-- x86
-- x64
-#- ARM
-
-build_script:
-- ps: >
-    msbuild "C:\projects\libmdbx\mdbx.sln" /verbosity:minimal
-    /logger:"C:\Program Files\AppVeyor\BuildAgent\Appveyor.MSBuildLogger.dll"
-    /property:PlatformToolset=$env:TOOLSET
-    /property:Configuration=$env:CONFIGURATION
-    /property:Platform=$env:PLATFORM
-
-test_script:
-- ps: |
-    if (($env:PLATFORM -eq "x86") -and (Test-Path "C:\projects\libmdbx\Win32\$env:CONFIGURATION\mdbx_test.exe" -PathType Leaf)) {
-      $mdbx_test = "C:\projects\libmdbx\Win32\$env:CONFIGURATION\mdbx_test.exe"
-      $mdbx_chk = "C:\projects\libmdbx\Win32\$env:CONFIGURATION\mdbx_chk.exe"
-    } elseif (($env:PLATFORM -ne "ARM") -and ($env:PLATFORM -ne "ARM64")) {
-      $mdbx_test = "C:\projects\libmdbx\$env:PLATFORM\$env:CONFIGURATION\mdbx_test.exe"
-      $mdbx_chk = "C:\projects\libmdbx\$env:PLATFORM\$env:CONFIGURATION\mdbx_chk.exe"
-    } else {
-      $mdbx_test = ""
-      $mdbx_chk = ""
-    }
-
-    if ($mdbx_test -ne "") {
-        & "$mdbx_test" --pathname=test.db --dont-cleanup-after basic | Tee-Object -file test.log | Select-Object -last 42
-        & "$mdbx_chk" -nvv test.db | Tee-Object -file chk.log | Select-Object -last 42
-    }
-
-on_failure:
-- ps: Push-AppveyorArtifact test.log
-- ps: Push-AppveyorArtifact test.db
-- ps: Push-AppveyorArtifact chk.log

cmake/profile.cmake

@@ -0,0 +1,58 @@
+# Copyright (c) 2012-2025 Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> ###############################################
+# SPDX-License-Identifier: Apache-2.0
+
+if(CMAKE_VERSION VERSION_LESS 3.8.2)
+  cmake_minimum_required(VERSION 3.0.2)
+elseif(CMAKE_VERSION VERSION_LESS 3.12)
+  cmake_minimum_required(VERSION 3.8.2)
+else()
+  cmake_minimum_required(VERSION 3.12)
+endif()
+
+cmake_policy(PUSH)
+cmake_policy(VERSION ${CMAKE_MINIMUM_REQUIRED_VERSION})
+
+unset(MEMCHECK_OPTION_NAME)
+if(NOT DEFINED ENABLE_MEMCHECK)
+  if(DEFINED MDBX_USE_VALGRIND)
+    set(MEMCHECK_OPTION_NAME "MDBX_USE_VALGRIND")
+  elseif(DEFINED ENABLE_VALGRIND)
+    set(MEMCHECK_OPTION_NAME "ENABLE_VALGRIND")
+  else()
+    set(MEMCHECK_OPTION_NAME "ENABLE_MEMCHECK")
+  endif()
+  if(MEMCHECK_OPTION_NAME STREQUAL "ENABLE_MEMCHECK")
+    option(ENABLE_MEMCHECK "Enable integration with valgrind, a memory analyzing tool" OFF)
+  elseif(${MEMCHECK_OPTION_NAME})
+    set(ENABLE_MEMCHECK ON)
+  else()
+    set(ENABLE_MEMCHECK OFF)
+  endif()
+endif()
+
+include(CheckLibraryExists)
+check_library_exists(gcov __gcov_flush "" HAVE_GCOV)
+
+option(ENABLE_GCOV "Enable integration with gcov, a code coverage program" OFF)
+
+option(ENABLE_GPROF "Enable integration with gprof, a performance analyzing tool" OFF)
+
+option(ENABLE_ASAN "Enable AddressSanitizer, a fast memory error detector based on compiler instrumentation" OFF)
+
+option(ENABLE_UBSAN
+       "Enable UndefinedBehaviorSanitizer, a fast undefined behavior detector based on compiler instrumentation" OFF)
+
+if(ENABLE_MEMCHECK)
+  if(CMAKE_CXX_COMPILER_LOADED)
+    include(CheckIncludeFileCXX)
+    check_include_file_cxx(valgrind/memcheck.h HAVE_VALGRIND_MEMCHECK_H)
+  else()
+    include(CheckIncludeFile)
+    check_include_file(valgrind/memcheck.h HAVE_VALGRIND_MEMCHECK_H)
+  endif()
+  if(NOT HAVE_VALGRIND_MEMCHECK_H)
+    message(FATAL_ERROR "${MEMCHECK_OPTION_NAME} option is set but valgrind/memcheck.h is not found")
+  endif()
+endif()
+
+cmake_policy(POP)

cmake/utils.cmake

@@ -0,0 +1,524 @@
+# Copyright (c) 2012-2025 Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> ###############################################
+# SPDX-License-Identifier: Apache-2.0
+
+if(CMAKE_VERSION VERSION_LESS 3.8.2)
+  cmake_minimum_required(VERSION 3.0.2)
+elseif(CMAKE_VERSION VERSION_LESS 3.12)
+  cmake_minimum_required(VERSION 3.8.2)
+else()
+  cmake_minimum_required(VERSION 3.12)
+endif()
+
+cmake_policy(PUSH)
+cmake_policy(VERSION ${CMAKE_MINIMUM_REQUIRED_VERSION})
+
+macro(add_option HIVE NAME DESCRIPTION DEFAULT)
+  list(APPEND ${HIVE}_BUILD_OPTIONS ${HIVE}_${NAME})
+  if(NOT ${DEFAULT} STREQUAL "AUTO")
+    option(${HIVE}_${NAME} "${DESCRIPTION}" ${DEFAULT})
+  elseif(NOT DEFINED ${HIVE}_${NAME})
+    set(${HIVE}_${NAME}_AUTO ON)
+  endif()
+endmacro()
+
+macro(set_if_undefined VARNAME)
+  if(NOT DEFINED "${VARNAME}")
+    set("${VARNAME}" ${ARGN})
+  endif()
+endmacro()
+
+macro(add_compile_flags languages)
+  foreach(_lang ${languages})
+    string(REPLACE ";" " " _flags "${ARGN}")
+    if(CMAKE_CXX_COMPILER_LOADED AND _lang STREQUAL "CXX")
+      set("${_lang}_FLAGS" "${${_lang}_FLAGS} ${_flags}")
+    endif()
+    if(CMAKE_C_COMPILER_LOADED AND _lang STREQUAL "C")
+      set("${_lang}_FLAGS" "${${_lang}_FLAGS} ${_flags}")
+    endif()
+  endforeach()
+  unset(_lang)
+  unset(_flags)
+endmacro(add_compile_flags)
+
+macro(remove_flag varname flag)
+  string(REGEX REPLACE "^(.*)( ${flag} )(.*)$" "\\1 \\3" ${varname} ${${varname}})
+  string(REGEX REPLACE "^((.+ )*)(${flag})(( .+)*)$" "\\1\\4" ${varname} ${${varname}})
+endmacro(remove_flag)
+
+macro(remove_compile_flag languages flag)
+  foreach(_lang ${languages})
+    if(CMAKE_CXX_COMPILER_LOADED AND _lang STREQUAL "CXX")
+      remove_flag(${_lang}_FLAGS ${flag})
+    endif()
+    if(CMAKE_C_COMPILER_LOADED AND _lang STREQUAL "C")
+      remove_flag(${_lang}_FLAGS ${flag})
+    endif()
+  endforeach()
+  unset(_lang)
+endmacro(remove_compile_flag)
+
+macro(set_source_files_compile_flags)
+  foreach(file ${ARGN})
+    get_filename_component(_file_ext ${file} EXT)
+    set(_lang "")
+    if("${_file_ext}" STREQUAL ".m")
+      set(_lang OBJC)
+      # CMake believes that Objective C is a flavor of C++, not C, and uses g++ compiler for .m files. LANGUAGE property
+      # forces CMake to use CC for ${file}
+      set_source_files_properties(${file} PROPERTIES LANGUAGE C)
+    elseif("${_file_ext}" STREQUAL ".mm")
+      set(_lang OBJCXX)
+    endif()
+
+    if(_lang)
+      get_source_file_property(_flags ${file} COMPILE_FLAGS)
+      if("${_flags}" STREQUAL "NOTFOUND")
+        set(_flags "${CMAKE_${_lang}_FLAGS}")
+      else()
+        set(_flags "${_flags} ${CMAKE_${_lang}_FLAGS}")
+      endif()
+      # message(STATUS "Set (${file} ${_flags}")
+      set_source_files_properties(${file} PROPERTIES COMPILE_FLAGS "${_flags}")
+    endif()
+  endforeach()
+  unset(_file_ext)
+  unset(_lang)
+endmacro(set_source_files_compile_flags)
+
+macro(semver_parse str)
+  set(_semver_ok FALSE)
+  set(_semver_err "")
+  set(_semver_major 0)
+  set(_semver_minor 0)
+  set(_semver_patch 0)
+  set(_semver_tweak_withdot "")
+  set(_semver_tweak "")
+  set(_semver_extra "")
+  set(_semver_prerelease_withdash "")
+  set(_semver_prerelease "")
+  set(_semver_buildmetadata_withplus "")
+  set(_semver_buildmetadata "")
+  if("${str}" MATCHES
+     "^v?(0|[1-9][0-9]*)\\.(0|[1-9][0-9]*)\\.(0|[1-9][0-9]*)(\\.(0|[1-9][0-9]*))?([-+]-*[0-9a-zA-Z]+.*)?$")
+    set(_semver_major ${CMAKE_MATCH_1})
+    set(_semver_minor ${CMAKE_MATCH_2})
+    set(_semver_patch ${CMAKE_MATCH_3})
+    set(_semver_tweak_withdot ${CMAKE_MATCH_4})
+    set(_semver_tweak ${CMAKE_MATCH_5})
+    set(_semver_extra "${CMAKE_MATCH_6}")
+    if("${_semver_extra}" STREQUAL "")
+      set(_semver_ok TRUE)
+    elseif("${_semver_extra}" MATCHES "^([.-][a-zA-Z0-9-]+)*(\\+[^+]+)?$")
+      set(_semver_prerelease_withdash "${CMAKE_MATCH_1}")
+      if(NOT "${_semver_prerelease_withdash}" STREQUAL "")
+        string(SUBSTRING "${_semver_prerelease_withdash}" 1 -1 _semver_prerelease)
+      endif()
+      set(_semver_buildmetadata_withplus "${CMAKE_MATCH_2}")
+      if(NOT "${_semver_buildmetadata_withplus}" STREQUAL "")
+        string(SUBSTRING "${_semver_buildmetadata_withplus}" 1 -1 _semver_buildmetadata)
+      endif()
+      set(_semver_ok TRUE)
+    else()
+      set(_semver_err
+          "Поля prerelease и/или buildmetadata (строка `-foo+bar` в составе `0.0.0[.0][-foo][+bar]`) не соответствуют SemVer-спецификации"
+      )
+    endif()
+  else()
+    set(_semver_err "Версионная отметка в целом не соответствует шаблону `0.0.0[.0][-foo][+bar]` SemVer-спецификации")
+  endif()
+endmacro(semver_parse)
+
+function(_semver_parse_probe str expect)
+  semver_parse(${str})
+  if(expect AND NOT _semver_ok)
+    message(FATAL_ERROR "semver_parse(${str}) expect SUCCESS, got ${_semver_ok}: ${_semver_err}")
+  elseif(NOT expect AND _semver_ok)
+    message(FATAL_ERROR "semver_parse(${str}) expect FAIL, got ${_semver_ok}")
+  endif()
+endfunction()
+
+function(semver_parse_selfcheck)
+  _semver_parse_probe("0.0.4" TRUE)
+  _semver_parse_probe("v1.2.3" TRUE)
+  _semver_parse_probe("10.20.30" TRUE)
+  _semver_parse_probe("10.20.30.42" TRUE)
+  _semver_parse_probe("1.1.2-prerelease+meta" TRUE)
+  _semver_parse_probe("1.1.2+meta" TRUE)
+  _semver_parse_probe("1.1.2+meta-valid" TRUE)
+  _semver_parse_probe("1.0.0-alpha" TRUE)
+  _semver_parse_probe("1.0.0-beta" TRUE)
+  _semver_parse_probe("1.0.0-alpha.beta" TRUE)
+  _semver_parse_probe("1.0.0-alpha.beta.1" TRUE)
+  _semver_parse_probe("1.0.0-alpha.1" TRUE)
+  _semver_parse_probe("1.0.0-alpha0.valid" TRUE)
+  _semver_parse_probe("1.0.0-alpha.0valid" TRUE)
+  _semver_parse_probe("1.0.0-alpha-a.b-c-somethinglong+build.1-aef.1-its-okay" TRUE)
+  _semver_parse_probe("1.0.0-rc.1+build.1" TRUE)
+  _semver_parse_probe("2.0.0-rc.1+build.123" TRUE)
+  _semver_parse_probe("1.2.3-beta" TRUE)
+  _semver_parse_probe("10.2.3-DEV-SNAPSHOT" TRUE)
+  _semver_parse_probe("1.2.3-SNAPSHOT-123" TRUE)
+  _semver_parse_probe("1.0.0" TRUE)
+  _semver_parse_probe("2.0.0" TRUE)
+  _semver_parse_probe("1.1.7" TRUE)
+  _semver_parse_probe("2.0.0+build.1848" TRUE)
+  _semver_parse_probe("2.0.1-alpha.1227" TRUE)
+  _semver_parse_probe("1.0.0-alpha+beta" TRUE)
+  _semver_parse_probe("1.2.3----RC-SNAPSHOT.12.9.1--.12+788" TRUE)
+  _semver_parse_probe("1.2.3----R-S.12.9.1--.12+meta" TRUE)
+  _semver_parse_probe("1.2.3----RC-SNAPSHOT.12.9.1--.12" TRUE)
+  _semver_parse_probe("1.0.0+0.build.1-rc.10000aaa-kk-0.1" TRUE)
+  _semver_parse_probe("99999999999999999999999.999999999999999999.99999999999999999" TRUE)
+  _semver_parse_probe("v1.0.0-0A.is.legal" TRUE)
+
+  _semver_parse_probe("1" FALSE)
+  _semver_parse_probe("1.2" FALSE)
+  # _semver_parse_probe("1.2.3-0123" FALSE) _semver_parse_probe("1.2.3-0123.0123" FALSE)
+  _semver_parse_probe("1.1.2+.123" FALSE)
+  _semver_parse_probe("+invalid" FALSE)
+  _semver_parse_probe("-invalid" FALSE)
+  _semver_parse_probe("-invalid+invalid" FALSE)
+  _semver_parse_probe("-invalid.01" FALSE)
+  _semver_parse_probe("alpha" FALSE)
+  _semver_parse_probe("alpha.beta" FALSE)
+  _semver_parse_probe("alpha.beta.1" FALSE)
+  _semver_parse_probe("alpha.1" FALSE)
+  _semver_parse_probe("alpha+beta" FALSE)
+  _semver_parse_probe("alpha_beta" FALSE)
+  _semver_parse_probe("alpha." FALSE)
+  _semver_parse_probe("alpha.." FALSE)
+  _semver_parse_probe("beta" FALSE)
+  _semver_parse_probe("1.0.0-alpha_beta" FALSE)
+  _semver_parse_probe("-alpha." FALSE)
+  _semver_parse_probe("1.0.0-alpha.." FALSE)
+  _semver_parse_probe("1.0.0-alpha..1" FALSE)
+  _semver_parse_probe("1.0.0-alpha...1" FALSE)
+  _semver_parse_probe("1.0.0-alpha....1" FALSE)
+  _semver_parse_probe("1.0.0-alpha.....1" FALSE)
+  _semver_parse_probe("1.0.0-alpha......1" FALSE)
+  _semver_parse_probe("1.0.0-alpha.......1" FALSE)
+  _semver_parse_probe("01.1.1" FALSE)
+  _semver_parse_probe("1.01.1" FALSE)
+  _semver_parse_probe("1.1.01" FALSE)
+  _semver_parse_probe("1.2" FALSE)
+  _semver_parse_probe("1.2.3.DEV" FALSE)
+  _semver_parse_probe("1.2-SNAPSHOT" FALSE)
+  _semver_parse_probe("1.2.31.2.3----RC-SNAPSHOT.12.09.1--..12+788" FALSE)
+  _semver_parse_probe("1.2-RC-SNAPSHOT" FALSE)
+  _semver_parse_probe("-1.0.3-gamma+b7718" FALSE)
+  _semver_parse_probe("+justmeta" FALSE)
+  _semver_parse_probe("9.8.7+meta+meta" FALSE)
+  _semver_parse_probe("9.8.7-whatever+meta+meta" FALSE)
+  _semver_parse_probe(
+    "99999999999999999999999.999999999999999999.99999999999999999----RC-SNAPSHOT.12.09.1--------------------------------..12"
+    FALSE)
+endfunction()
+
+macro(git_get_versioninfo source_root_directory)
+  set(_git_describe "")
+  set(_git_timestamp "")
+  set(_git_tree "")
+  set(_git_commit "")
+  set(_git_last_vtag "")
+  set(_git_trailing_commits 0)
+  set(_git_is_dirty FALSE)
+
+  execute_process(
+    COMMAND ${GIT} show --no-patch --format=%cI HEAD
+    OUTPUT_VARIABLE _git_timestamp
+    OUTPUT_STRIP_TRAILING_WHITESPACE
+    WORKING_DIRECTORY ${source_root_directory}
+    RESULT_VARIABLE _rc)
+  if(_rc OR "${_git_timestamp}" STREQUAL "%cI")
+    execute_process(
+      COMMAND ${GIT} show --no-patch --format=%ci HEAD
+      OUTPUT_VARIABLE _git_timestamp
+      OUTPUT_STRIP_TRAILING_WHITESPACE
+      WORKING_DIRECTORY ${source_root_directory}
+      RESULT_VARIABLE _rc)
+    if(_rc OR "${_git_timestamp}" STREQUAL "%ci")
+      message(FATAL_ERROR "Please install latest version of git (`show --no-patch --format=%cI HEAD` failed)")
+    endif()
+  endif()
+
+  execute_process(
+    COMMAND ${GIT} show --no-patch --format=%T HEAD
+    OUTPUT_VARIABLE _git_tree
+    OUTPUT_STRIP_TRAILING_WHITESPACE
+    WORKING_DIRECTORY ${source_root_directory}
+    RESULT_VARIABLE _rc)
+  if(_rc OR "${_git_tree}" STREQUAL "")
+    message(FATAL_ERROR "Please install latest version of git (`show --no-patch --format=%T HEAD` failed)")
+  endif()
+
+  execute_process(
+    COMMAND ${GIT} show --no-patch --format=%H HEAD
+    OUTPUT_VARIABLE _git_commit
+    OUTPUT_STRIP_TRAILING_WHITESPACE
+    WORKING_DIRECTORY ${source_root_directory}
+    RESULT_VARIABLE _rc)
+  if(_rc OR "${_git_commit}" STREQUAL "")
+    message(FATAL_ERROR "Please install latest version of git (`show --no-patch --format=%H HEAD` failed)")
+  endif()
+
+  execute_process(
+    COMMAND ${GIT} status --untracked-files=no --porcelain
+    OUTPUT_VARIABLE _git_status
+    OUTPUT_STRIP_TRAILING_WHITESPACE
+    WORKING_DIRECTORY ${source_root_directory}
+    RESULT_VARIABLE _rc)
+  if(_rc)
+    message(FATAL_ERROR "Please install latest version of git (`status --untracked-files=no --porcelain` failed)")
+  endif()
+  if(NOT "${_git_status}" STREQUAL "")
+    set(_git_commit "DIRTY-${_git_commit}")
+    set(_git_is_dirty TRUE)
+  endif()
+  unset(_git_status)
+
+  execute_process(
+    COMMAND ${GIT} describe --tags --abbrev=0 "--match=v[0-9]*"
+    OUTPUT_VARIABLE _git_last_vtag
+    OUTPUT_STRIP_TRAILING_WHITESPACE
+    WORKING_DIRECTORY ${source_root_directory}
+    RESULT_VARIABLE _rc)
+  if(_rc OR "${_git_last_vtag}" STREQUAL "")
+    execute_process(
+      COMMAND ${GIT} tag
+      OUTPUT_VARIABLE _git_tags_dump
+      OUTPUT_STRIP_TRAILING_WHITESPACE
+      WORKING_DIRECTORY ${source_root_directory}
+      RESULT_VARIABLE _rc)
+    execute_process(
+      COMMAND ${GIT} rev-list --count --no-merges --remove-empty HEAD
+      OUTPUT_VARIABLE _git_whole_count
+      OUTPUT_STRIP_TRAILING_WHITESPACE
+      WORKING_DIRECTORY ${source_root_directory}
+      RESULT_VARIABLE _rc)
+    if(_rc)
+      message(
+        FATAL_ERROR
+          "Please install latest version of git (`git rev-list --count --no-merges --remove-empty HEAD` failed)")
+    endif()
+    if(_git_whole_count GREATER 42 AND "${_git_tags_dump}" STREQUAL "")
+      message(FATAL_ERROR "Please fetch tags (`describe --tags --abbrev=0 --match=v[0-9]*` failed)")
+    else()
+      message(NOTICE "Falling back to version `0.0.0` (have you made an initial release?")
+    endif()
+    set(_git_last_vtag "0.0.0")
+    set(_git_trailing_commits ${_git_whole_count})
+    execute_process(
+      COMMAND ${GIT} describe --tags --dirty --long --always
+      OUTPUT_VARIABLE _git_describe
+      OUTPUT_STRIP_TRAILING_WHITESPACE
+      WORKING_DIRECTORY ${source_root_directory}
+      RESULT_VARIABLE _rc)
+    if(_rc OR "${_git_describe}" STREQUAL "")
+      execute_process(
+        COMMAND ${GIT} describe --tags --all --dirty --long --always
+        OUTPUT_VARIABLE _git_describe
+        OUTPUT_STRIP_TRAILING_WHITESPACE
+        WORKING_DIRECTORY ${source_root_directory}
+        RESULT_VARIABLE _rc)
+      if(_rc OR "${_git_describe}" STREQUAL "")
+        message(FATAL_ERROR "Please install latest version of git (`describe --tags --all --long` failed)")
+      endif()
+    endif()
+  else()
+    execute_process(
+      COMMAND ${GIT} describe --tags --dirty --long "--match=v[0-9]*"
+      OUTPUT_VARIABLE _git_describe
+      OUTPUT_STRIP_TRAILING_WHITESPACE
+      WORKING_DIRECTORY ${source_root_directory}
+      RESULT_VARIABLE _rc)
+    if(_rc OR "${_git_describe}" STREQUAL "")
+      message(FATAL_ERROR "Please install latest version of git (`describe --tags --long --match=v[0-9]*`)")
+    endif()
+    execute_process(
+      COMMAND ${GIT} rev-list --count "${_git_last_vtag}..HEAD"
+      OUTPUT_VARIABLE _git_trailing_commits
+      OUTPUT_STRIP_TRAILING_WHITESPACE
+      WORKING_DIRECTORY ${source_root_directory}
+      RESULT_VARIABLE _rc)
+    if(_rc OR "${_git_trailing_commits}" STREQUAL "")
+      message(FATAL_ERROR "Please install latest version of git (`rev-list --count ${_git_last_vtag}..HEAD` failed)")
+    endif()
+  endif()
+endmacro(git_get_versioninfo)
+
+macro(semver_provide name source_root_directory build_directory_for_json_output build_metadata parent_scope)
+  set(_semver "")
+  set(_git_describe "")
+  set(_git_timestamp "")
+  set(_git_tree "")
+  set(_git_commit "")
+  set(_version_from "")
+  set(_git_root FALSE)
+
+  find_program(GIT git)
+  if(GIT)
+    execute_process(
+      COMMAND ${GIT} rev-parse --show-toplevel
+      OUTPUT_VARIABLE _git_root
+      ERROR_VARIABLE _git_root_error
+      OUTPUT_STRIP_TRAILING_WHITESPACE
+      WORKING_DIRECTORY ${source_root_directory}
+      RESULT_VARIABLE _rc)
+    if(_rc OR "${_git_root}" STREQUAL "")
+      if(EXISTS "${source_root_directory}/.git")
+        message(ERROR "`git rev-parse --show-toplevel` failed '${_git_root_error}'")
+      else()
+        message(VERBOSE "`git rev-parse --show-toplevel` failed '${_git_root_error}'")
+      endif()
+    else()
+      set(_source_root "${source_root_directory}")
+      if(NOT CMAKE_VERSION VERSION_LESS 3.19)
+        file(REAL_PATH "${_git_root}" _git_root)
+        file(REAL_PATH "${_source_root}" _source_root)
+      endif()
+      if(_source_root STREQUAL _git_root AND EXISTS "${_git_root}/VERSION.json")
+        message(
+          FATAL_ERROR
+            "Несколько источников информации о версии, допустим только один из: репозиторий git, либо файл VERSION.json"
+        )
+      endif()
+    endif()
+  endif()
+
+  if(EXISTS "${source_root_directory}/VERSION.json")
+    set(_version_from "${source_root_directory}/VERSION.json")
+
+    if(CMAKE_VERSION VERSION_LESS 3.19)
+      message(FATAL_ERROR "Требуется CMake версии >= 3.19 для чтения VERSION.json")
+    endif()
+    file(
+      STRINGS "${_version_from}" _versioninfo_json NEWLINE_CONSUME
+      LIMIT_COUNT 9
+      LIMIT_INPUT 999
+      ENCODING UTF-8)
+    string(JSON _git_describe GET ${_versioninfo_json} git_describe)
+    string(JSON _git_timestamp GET "${_versioninfo_json}" "git_timestamp")
+    string(JSON _git_tree GET "${_versioninfo_json}" "git_tree")
+    string(JSON _git_commit GET "${_versioninfo_json}" "git_commit")
+    string(JSON _semver GET "${_versioninfo_json}" "semver")
+    unset(_json_object)
+    if(NOT _semver)
+      message(FATAL_ERROR "Unable to retrieve ${name} version from \"${_version_from}\" file.")
+    endif()
+    semver_parse("${_semver}")
+    if(NOT _semver_ok)
+      message(FATAL_ERROR "SemVer `${_semver}` from ${_version_from}: ${_semver_err}")
+    endif()
+  elseif(_git_root AND _source_root STREQUAL _git_root)
+    set(_version_from git)
+    git_get_versioninfo(${source_root_directory})
+    semver_parse(${_git_last_vtag})
+    if(NOT _semver_ok)
+      message(FATAL_ERROR "Git tag `${_git_last_vtag}`: ${_semver_err}")
+    endif()
+    if(_git_trailing_commits GREATER 0 AND "${_semver_tweak}" STREQUAL "")
+      set(_semver_tweak ${_git_trailing_commits})
+    endif()
+
+  elseif(GIT)
+    message(
+      FATAL_ERROR
+        "Нет источника информации о версии (${source_root_directory}), требуется один из: репозиторий git, либо VERSION.json"
+    )
+  else()
+    message(FATAL_ERROR "Требуется git для получения информации о версии")
+  endif()
+
+  if(NOT _git_describe
+     OR NOT _git_timestamp
+     OR NOT _git_tree
+     OR NOT _git_commit
+     OR "${_semver_major}" STREQUAL ""
+     OR "${_semver_minor}" STREQUAL ""
+     OR "${_semver_patch}" STREQUAL "")
+    message(ERROR "Unable to retrieve ${name} version from ${_version_from}.")
+  endif()
+
+  set(_semver "${_semver_major}.${_semver_minor}.${_semver_patch}")
+  if("${_semver_tweak}" STREQUAL "")
+    set(_semver_tweak 0)
+  elseif(_semver_tweak GREATER 0)
+    string(APPEND _semver ".${_semver_tweak}")
+  endif()
+  if(NOT "${_semver_prerelease}" STREQUAL "")
+    string(APPEND _semver "-${_semver_prerelease}")
+  endif()
+  if(_git_is_dirty)
+    string(APPEND _semver "-DIRTY")
+  endif()
+
+  set(_semver_complete "${_semver}")
+  if(NOT "${build_metadata}" STREQUAL "")
+    string(APPEND _semver_complete "+${build_metadata}")
+  endif()
+
+  set(${name}_VERSION "${_semver_complete}")
+  set(${name}_VERSION_PURE "${_semver}")
+  set(${name}_VERSION_MAJOR ${_semver_major})
+  set(${name}_VERSION_MINOR ${_semver_minor})
+  set(${name}_VERSION_PATCH ${_semver_patch})
+  set(${name}_VERSION_TWEAK ${_semver_tweak})
+  set(${name}_VERSION_PRERELEASE "${_semver_prerelease}")
+  set(${name}_GIT_DESCRIBE "${_git_describe}")
+  set(${name}_GIT_TIMESTAMP "${_git_timestamp}")
+  set(${name}_GIT_TREE "${_git_tree}")
+  set(${name}_GIT_COMMIT "${_git_commit}")
+
+  if(${parent_scope})
+    set(${name}_VERSION
+        "${_semver_complete}"
+        PARENT_SCOPE)
+    set(${name}_VERSION_PURE
+        "${_semver}"
+        PARENT_SCOPE)
+    set(${name}_VERSION_MAJOR
+        ${_semver_major}
+        PARENT_SCOPE)
+    set(${name}_VERSION_MINOR
+        ${_semver_minor}
+        PARENT_SCOPE)
+    set(${name}_VERSION_PATCH
+        ${_semver_patch}
+        PARENT_SCOPE)
+    set(${name}_VERSION_TWEAK
+        "${_semver_tweak}"
+        PARENT_SCOPE)
+    set(${name}_VERSION_PRERELEASE
+        "${_semver_prerelease}"
+        PARENT_SCOPE)
+    set(${name}_GIT_DESCRIBE
+        "${_git_describe}"
+        PARENT_SCOPE)
+    set(${name}_GIT_TIMESTAMP
+        "${_git_timestamp}"
+        PARENT_SCOPE)
+    set(${name}_GIT_TREE
+        "${_git_tree}"
+        PARENT_SCOPE)
+    set(${name}_GIT_COMMIT
+        "${_git_commit}"
+        PARENT_SCOPE)
+  endif()
+
+  if(_version_from STREQUAL "git")
+    string(
+      CONFIGURE
+        "{
+      \"git_describe\" : \"@_git_describe@\",
+      \"git_timestamp\" : \"@_git_timestamp@\",
+      \"git_tree\" : \"@_git_tree@\",
+      \"git_commit\" : \"@_git_commit@\",
+      \"semver\" : \"@_semver@\"\n}"
+        _versioninfo_json
+      @ONLY ESCAPE_QUOTES)
+    file(WRITE "${build_directory_for_json_output}/VERSION.json" "${_versioninfo_json}")
+  endif()
+endmacro(semver_provide)
+
+cmake_policy(POP)

conanfile.py

@@ -0,0 +1,323 @@
+import shutil
+import json
+import os
+import re
+import subprocess
+from conan.tools.files import rm
+from conan.tools.scm import Git
+from conan.tools.apple import is_apple_os
+from conan.tools.cmake import CMakeToolchain, CMake, cmake_layout, CMakeDeps
+from conan import ConanFile
+required_conan_version = '>=2.7'
+
+
+def semver_parse(s):
+    m = re.match('^v?(?P<major>0|[1-9]\d*)\.(?P<minor>0|[1-9]\d*)\.(?P<patch>0|[1-9]\d*)(\\.(?P<tweak>0|[1-9]\d*))?(?:-(?P<prerelease>(?:0|[1-9]\d*|\d*[a-zA-Z-][0-9a-zA-Z-]*)(?:\.(?:0|[1-9]\d*|\d*[a-zA-Z-][0-9a-zA-Z-]*))*))?(?:\+(?P<buildmetadata>[0-9a-zA-Z-]+(?:\.[0-9a-zA-Z-]+)*))?$', s)
+    return m.groupdict() if m else None
+
+
+def semver_string(semver):
+    s = str(semver['major']) + '.' + \
+        str(semver['minor']) + '.' + str(semver['patch'])
+    if not semver['tweak'] is None and semver['tweak'] != 0:
+        s += '.' + str(semver['tweak'])
+    if not semver['prerelease'] is None and semver['prerelease'] != '':
+        s += '-' + semver['prerelease']
+    return s
+
+
+def semver_string_with_buildmetadata(semver):
+    s = semver_string(semver)
+    if not semver['buildmetadata'] is None and semver['buildmetadata'] != '':
+        s += '+' + semver['buildmetadata']
+    return s
+
+
+class libmdbx(ConanFile):
+    name = 'mdbx'
+    package_type = 'library'
+    description = 'One of the fastest embeddable key-value ACID database without WAL. libmdbx surpasses the legendary LMDB in terms of reliability, features and performance.'
+    license = 'Apache-2.0'
+    author = 'Leo Yuriev <leo@yuriev.ru>'
+    homepage = 'https://libmdbx.dqdkfa.ru'
+    url = 'https://gitflic.ru/project/erthink/libmdbx.git'
+    topics = ('embedded-database', 'key-value', 'btree', 'LMDB', 'storage-engine',
+              'data-storage', 'nosql', 'ACID', 'MVCC', 'MDBX')
+    no_copy_source = True
+    test_type = 'explicit'
+    build_policy = 'missing'
+    revision_mode = 'scm'
+    languages = 'C', 'C++'
+    provides = 'libmdbx'
+    implements = ['auto_shared_fpic']
+    # upload_policy = 'skip'
+    # exports_sources = 'LICENSE', 'NOTICE', 'CMakeLists.txt', '*.h', '*.h++', '*.c', '*.c++', 'ntdll.def', 'man1/*', 'cmake/*', 'config.h.in'
+
+    settings = 'os', 'compiler', 'build_type', 'arch'
+    options = {
+        'mdbx.64bit_atomic': ['Auto', True, False],
+        'mdbx.64bit_cas': ['Auto', True, False],
+        'mdbx.apple.speed_insteadof_durability': ['Default', True, False],
+        'mdbx.avoid_msync': ['Auto', True, False],
+        'mdbx.build_cxx': ['Default', True, False],
+        'mdbx.build_tools': ['Default', True, False],
+        'mdbx.cacheline_size': ['Auto', 16, 32, 64, 128, 256],
+        'mdbx.disable_validation': ['Default', True, False],
+        'mdbx.enable_bigfoot': ['Default', True, False],
+        'mdbx.enable_dbi_lockfree': ['Default', True, False],
+        'mdbx.enable_dbi_sparse': ['Default', True, False],
+        'mdbx.enable_pgop_stat': ['Default', True, False],
+        'mdbx.enable_profgc': ['Default', True, False],
+        'mdbx.enable_refund': ['Default', True, False],
+        'mdbx.env_checkpid': ['Default', True, False],
+        'mdbx.force_assertions': ['Default', True, False],
+        'mdbx.have_builtin_cpu_supports': ['Auto', True, False],
+        'mdbx.locking': ['Auto', 'WindowsFileLocking', 'SystemV', 'POSIX1988', 'POSIX2001', 'POSIX2008'],
+        'mdbx.mmap_incoherent_file_write': ['Auto', True, False],
+        'mdbx.mmap_needs_jolt': ['Auto', True, False],
+        'mdbx.trust_rtc': ['Default', True, False],
+        'mdbx.txn_checkowner': ['Default', True, False],
+        'mdbx.unaligned_ok': ['Auto', True, False],
+        'mdbx.use_copyfilerange': ['Auto', True, False],
+        'mdbx.use_mincore': ['Auto', True, False],
+        'mdbx.use_ofdlocks': ['Auto', True, False],
+        'mdbx.use_sendfile': ['Auto', True, False],
+        'mdbx.without_msvc_crt': ['Default', True, False],
+        'shared': [True, False],
+    }
+    default_options = {
+        'mdbx.64bit_atomic': 'Auto',
+        'mdbx.64bit_cas': 'Auto',
+        'mdbx.apple.speed_insteadof_durability': 'Default',
+        'mdbx.avoid_msync': 'Auto',
+        'mdbx.build_cxx': 'Default',
+        'mdbx.build_tools': 'Default',
+        'mdbx.cacheline_size': 'Auto',
+        'mdbx.disable_validation': 'Default',
+        'mdbx.enable_bigfoot': 'Default',
+        'mdbx.enable_dbi_lockfree': 'Default',
+        'mdbx.enable_dbi_sparse': 'Default',
+        'mdbx.enable_pgop_stat': 'Default',
+        'mdbx.enable_profgc': 'Default',
+        'mdbx.enable_refund': 'Default',
+        'mdbx.env_checkpid': 'Default',
+        'mdbx.force_assertions': 'Default',
+        'mdbx.have_builtin_cpu_supports': 'Auto',
+        'mdbx.locking': 'Auto',
+        'mdbx.mmap_incoherent_file_write': 'Auto',
+        'mdbx.mmap_needs_jolt': 'Auto',
+        'mdbx.trust_rtc': 'Default',
+        'mdbx.txn_checkowner': 'Default',
+        'mdbx.unaligned_ok': 'Auto',
+        'mdbx.use_copyfilerange': 'Auto',
+        'mdbx.use_mincore': 'Auto',
+        'mdbx.use_ofdlocks': 'Auto',
+        'mdbx.use_sendfile': 'Auto',
+        'mdbx.without_msvc_crt': 'Default',
+        'shared': True,
+    }
+    options_description = {
+        'mdbx.64bit_atomic': 'Advanced: Assume 64-bit operations are atomic and not splitted to 32-bit halves. ',
+        'mdbx.64bit_cas': 'Advanced: Assume 64-bit atomic compare-and-swap operation is available. ',
+        'mdbx.apple.speed_insteadof_durability': 'Disable using `fcntl(F_FULLFSYNC)` for a performance reasons at the cost of durability on power failure. ',
+        'mdbx.avoid_msync': 'Disable in-memory database updating with consequent flush-to-disk/msync syscall in `MDBX_WRITEMAP` mode. ',
+        'mdbx.build_cxx': 'Build C++ portion. ',
+        'mdbx.build_tools': 'Build CLI tools (mdbx_chk/stat/dump/load/copy/drop). ',
+        'mdbx.cacheline_size': 'Advanced: CPU cache line size for data alignment to avoid cache line false-sharing. ',
+        'mdbx.disable_validation': 'Disable some checks to reduce an overhead and detection probability of database corruption to a values closer to the LMDB. ',
+        'mdbx.enable_bigfoot': 'Chunking long list of retired pages during huge transactions commit to avoid use sequences of pages. ',
+        'mdbx.enable_dbi_lockfree': 'Support for deferred releasing and a lockfree path to quickly open DBI handles. ',
+        'mdbx.enable_dbi_sparse': 'Support for sparse sets of DBI handles to reduce overhead when starting and processing transactions. ',
+        'mdbx.enable_pgop_stat': 'Gathering statistics for page operations. ',
+        'mdbx.enable_profgc': 'Profiling of GC search and updates. ',
+        'mdbx.enable_refund': 'Online database zero-cost auto-compactification during write-transactions. ',
+        'mdbx.env_checkpid': "Checking PID inside libmdbx's API against reuse database environment after the `fork()`. ",
+        'mdbx.force_assertions': 'Forces assertion checking even for release builds. ',
+        'mdbx.have_builtin_cpu_supports': 'Advanced: Assume the compiler and target system has `__builtin_cpu_supports()`. ',
+        'mdbx.locking': 'Advanced: Choices the locking implementation. ',
+        'mdbx.mmap_incoherent_file_write': "Advanced: Assume system don't have unified page cache and/or file write operations incoherent with memory-mapped files. ",
+        'mdbx.mmap_needs_jolt': 'Advanced: Assume system needs explicit syscall to sync/flush/write modified mapped memory. ',
+        'mdbx.trust_rtc': 'Advanced: Does a system have battery-backed Real-Time Clock or just a fake. ',
+        'mdbx.txn_checkowner': 'Checking transaction owner thread against misuse transactions from other threads. ',
+        'mdbx.unaligned_ok': 'Advanced: Assume a target CPU and/or the compiler support unaligned access. ',
+        'mdbx.use_copyfilerange': 'Advanced: Use `copy_file_range()` syscall. ',
+        'mdbx.use_mincore': "Use Unix' `mincore()` to determine whether database pages are resident in memory. ",
+        'mdbx.use_ofdlocks': 'Advanced: Use POSIX OFD-locks. ',
+        'mdbx.use_sendfile': 'Advancedc: Use `sendfile()` syscall. ',
+        'mdbx.without_msvc_crt': 'Avoid dependence from MSVC CRT and use ntdll.dll instead. ',
+    }
+
+    build_metadata = None
+
+    def config_options(self):
+        if self.settings.get_safe('os') != 'Linux':
+            self.options.rm_safe('mdbx.use_copyfilerange')
+            self.options.rm_safe('mdbx.use_sendfile')
+        if self.settings.get_safe('os') == 'Windows':
+            self.default_options['mdbx.avoid_msync'] = True
+            self.options.rm_safe('mdbx.env_checkpid')
+            self.options.rm_safe('mdbx.locking')
+            self.options.rm_safe('mdbx.mmap_incoherent_file_write')
+            self.options.rm_safe('mdbx.use_mincore')
+            self.options.rm_safe('mdbx.use_ofdlocks')
+        else:
+            self.options.rm_safe('mdbx.without_msvc_crt')
+        if is_apple_os(self):
+            self.options.rm_safe('mdbx.mmap_incoherent_file_write')
+        else:
+            self.options.rm_safe('mdbx.apple.speed_insteadof_durability')
+
+    def fetch_versioninfo_from_git(self):
+        git = Git(self, folder=self.recipe_folder)
+        git_timestamp = git.run('show --no-patch --format=%cI HEAD')
+        git_tree = git.run('show --no-patch --format=%T HEAD')
+        git_commit = git.run('show --no-patch --format=%H HEAD')
+        if git.run('rev-list --tags --count') == 0:
+            git.run('fetch --tags')
+        git_last_vtag = git.run('describe --tags --abbrev=0 --match=v[0-9]*')
+        if git_last_vtag == '':
+            git_describe = git.run('describe --all --long --always')
+            git_semver = semver_parse(
+                '0.0.0.' + git.run('rev-list --count --remove-empty --no-merges HEAD'))
+        else:
+            git_describe = git.run('describe --tags --long --match=v[0-9]*')
+            git_version = '.'.join(
+                map(str, re.split('[-v.]+', git.run('describe --tags --match=v[0-9]*'))[1:5]))
+            git_semver = semver_parse(git_last_vtag)
+            if git_semver['prerelease'] is None or git_semver['prerelease'] == '':
+                git_since_vtag = git.run(
+                    'rev-list ' + git_last_vtag + '.. --count')
+                if int(git_since_vtag) > 0:
+                    git_semver['tweak'] = int(git_since_vtag)
+                else:
+                    git_semver['tweak'] = None
+        info = {'git_describe': git_describe, 'git_timestamp': git_timestamp,
+                'git_tree': git_tree, 'git_commit': git_commit, 'semver': semver_string(git_semver)}
+        return info
+
+    def export_sources(self):
+        subprocess.run(['make', '-C', self.recipe_folder, 'DIST_DIR=' +
+                       self.export_sources_folder, '@dist-checked.tag'], check=True)
+        rm(self, 'Makefile', self.export_sources_folder)
+        rm(self, 'GNUmakefile', self.export_sources_folder)
+        # json.dump(self.fetch_versioninfo_from_git(), open(os.path.join(
+        #    self.export_sources_folder, 'VERSION.json'), 'w', encoding='utf-8'))
+
+    def source(self):
+        version_json_pathname = os.path.join(
+            self.export_sources_folder, 'VERSION.json')
+        version_json = json.load(
+            open(os.path.join(version_json_pathname), encoding='utf-8'))['semver']
+        if version_json != semver_string(semver_parse(self.version)):
+            self.output.error('Package/Recipe version "' + self.version +
+                              '" mismatch VERSION.json "' + version_json + '"')
+
+    def set_version(self):
+        if self.build_metadata is None and not self.version is None:
+            self.build_metadata = self.version
+            semver = semver_parse(self.build_metadata)
+            if semver:
+                self.build_metadata = semver['buildmetadata']
+            else:
+                self.build_metadata = re.match(
+                    '^[^0-9a-zA-Z]*([0-9a-zA-Z]+[-.0-9a-zA-Z]*)', self.build_metadata).group(1)
+        if self.build_metadata is None:
+            self.build_metadata = ''
+        version_json_pathname = os.path.join(
+            self.recipe_folder, 'VERSION.json')
+        if os.path.exists(version_json_pathname):
+            self.version = json.load(
+                open(version_json_pathname, encoding='utf-8'))['semver']
+            version_from = "'" + version_json_pathname + "'"
+        else:
+            self.version = self.fetch_versioninfo_from_git()['semver']
+            version_from = 'Git'
+        self.output.verbose('Fetch version from ' +
+                            version_from + ': ' + self.version)
+        if self.build_metadata != '':
+            self.version += '+' + self.build_metadata
+
+    def layout(self):
+        cmake_layout(self)
+
+    def handle_option(self, tc, name, define=False):
+        opt = self.options.get_safe(name)
+        if not opt is None:
+            value = str(opt).lower()
+            if value != 'auto' and value != 'default':
+                name = name.upper().replace('.', '_')
+                if define:
+                    if value == 'false' or value == 'no' or value == 'off':
+                        tc.preprocessor_definitions[name] = 0
+                    elif value == 'true' or value == 'yes' or value == 'on':
+                        tc.preprocessor_definitions[name] = 1
+                    else:
+                        tc.preprocessor_definitions[name] = int(opt)
+                    self.output.highlight(
+                        name + '=' + str(tc.preprocessor_definitions[name]) + ' (' + str(opt) + ')')
+                else:
+                    tc.cache_variables[name] = opt
+                    self.output.highlight(
+                        name + '=' + str(tc.cache_variables[name]) + ' (' + str(opt) + ')')
+
+    def generate(self):
+        tc = CMakeToolchain(self)
+        if self.build_metadata is None:
+            self.build_metadata = semver_parse(self.version)['buildmetadata']
+        if not self.build_metadata is None and self.build_metadata != '':
+            tc.variables['MDBX_BUILD_METADATA'] = self.build_metadata
+            self.output.highlight('MDBX_BUILD_METADATA is ' +
+                                  str(tc.variables['MDBX_BUILD_METADATA']))
+        self.handle_option(tc, 'mdbx.64bit_atomic', True)
+        self.handle_option(tc, 'mdbx.64bit_cas', True)
+        self.handle_option(tc, 'mdbx.apple.speed_insteadof_durability')
+        self.handle_option(tc, 'mdbx.avoid_msync')
+        self.handle_option(tc, 'mdbx.build_tools')
+        self.handle_option(tc, 'mdbx.build_cxx')
+        self.handle_option(tc, 'mdbx.cacheline_size', True)
+        self.handle_option(tc, 'mdbx.disable_validation')
+        self.handle_option(tc, 'mdbx.enable_bigfoot')
+        self.handle_option(tc, 'mdbx.enable_dbi_lockfree')
+        self.handle_option(tc, 'mdbx.enable_dbi_sparse')
+        self.handle_option(tc, 'mdbx.enable_pgop_stat')
+        self.handle_option(tc, 'mdbx.enable_profgc')
+        self.handle_option(tc, 'mdbx.enable_refund')
+        self.handle_option(tc, 'mdbx.env_checkpid')
+        self.handle_option(tc, 'mdbx.force_assertions')
+        self.handle_option(tc, 'mdbx.have_builtin_cpu_supports', True)
+        self.handle_option(tc, 'mdbx.mmap_incoherent_file_write', True)
+        self.handle_option(tc, 'mdbx.mmap_needs_jolt')
+        self.handle_option(tc, 'mdbx.trust_rtc')
+        self.handle_option(tc, 'mdbx.txn_checkowner')
+        self.handle_option(tc, 'mdbx.unaligned_ok', True)
+        self.handle_option(tc, 'mdbx.use_copyfilerange', True)
+        self.handle_option(tc, 'mdbx.use_mincore')
+        self.handle_option(tc, 'mdbx.use_ofdlocks')
+        self.handle_option(tc, 'mdbx.use_sendfile', True)
+        self.handle_option(tc, 'mdbx.without_msvc_crt')
+        opt = self.options.get_safe('mdbx.locking', 'auto')
+        if not opt is None:
+            value = str(opt).lower()
+            if value != 'auto' and value != 'default':
+                map = {'windowsfilelocking': -1, 'systemv': 5, 'posix1988': 1988,
+                       'posix2001': 2001, 'posix2008': 2008}
+                value = map[value]
+                tc.cache_variables['MDBX_LOCKING'] = value
+                self.output.highlight('MDBX_LOCKING=' +
+                                      str(tc.cache_variables['MDBX_LOCKING']))
+        tc.generate()
+
+    def build(self):
+        cmake = CMake(self)
+        cmake.configure()
+        cmake.build()
+
+    def package(self):
+        cmake = CMake(self)
+        cmake.install()
+
+    def package_info(self):
+        if self.options.shared:
+            self.cpp_info.libs = ['mdbx']
+        else:
+            self.cpp_info.libs = ['mdbx-static']

dll.vcxproj

@@ -1,181 +0,0 @@
-<?xml version="1.0" encoding="utf-8"?>
-<Project DefaultTargets="Build" ToolsVersion="14.0" xmlns="http://schemas.microsoft.com/developer/msbuild/2003">
-  <ItemGroup Label="ProjectConfigurations">
-    <ProjectConfiguration Include="Debug|Win32">
-      <Configuration>Debug</Configuration>
-      <Platform>Win32</Platform>
-    </ProjectConfiguration>
-    <ProjectConfiguration Include="Release|Win32">
-      <Configuration>Release</Configuration>
-      <Platform>Win32</Platform>
-    </ProjectConfiguration>
-    <ProjectConfiguration Include="Debug|x64">
-      <Configuration>Debug</Configuration>
-      <Platform>x64</Platform>
-    </ProjectConfiguration>
-    <ProjectConfiguration Include="Release|x64">
-      <Configuration>Release</Configuration>
-      <Platform>x64</Platform>
-    </ProjectConfiguration>
-  </ItemGroup>
-  <PropertyGroup Label="Globals">
-    <ProjectGuid>{6D19209B-ECE7-4B9C-941C-0AA2B484F199}</ProjectGuid>
-    <Keyword>Win32Proj</Keyword>
-    <WindowsTargetPlatformVersion>8.1</WindowsTargetPlatformVersion>
-  </PropertyGroup>
-  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.Default.props" />
-  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'" Label="Configuration">
-    <ConfigurationType>DynamicLibrary</ConfigurationType>
-    <UseDebugLibraries>true</UseDebugLibraries>
-    <PlatformToolset>v140</PlatformToolset>
-    <CharacterSet>MultiByte</CharacterSet>
-  </PropertyGroup>
-  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'" Label="Configuration">
-    <ConfigurationType>DynamicLibrary</ConfigurationType>
-    <UseDebugLibraries>false</UseDebugLibraries>
-    <PlatformToolset>v140</PlatformToolset>
-    <WholeProgramOptimization>true</WholeProgramOptimization>
-    <CharacterSet>MultiByte</CharacterSet>
-  </PropertyGroup>
-  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'" Label="Configuration">
-    <ConfigurationType>DynamicLibrary</ConfigurationType>
-    <UseDebugLibraries>true</UseDebugLibraries>
-    <PlatformToolset>v140</PlatformToolset>
-    <CharacterSet>MultiByte</CharacterSet>
-  </PropertyGroup>
-  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'" Label="Configuration">
-    <ConfigurationType>DynamicLibrary</ConfigurationType>
-    <UseDebugLibraries>false</UseDebugLibraries>
-    <PlatformToolset>v140</PlatformToolset>
-    <WholeProgramOptimization>true</WholeProgramOptimization>
-    <CharacterSet>MultiByte</CharacterSet>
-  </PropertyGroup>
-  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.props" />
-  <ImportGroup Label="ExtensionSettings">
-  </ImportGroup>
-  <ImportGroup Label="Shared">
-  </ImportGroup>
-  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
-    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
-  </ImportGroup>
-  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
-    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
-  </ImportGroup>
-  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
-    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
-  </ImportGroup>
-  <ImportGroup Label="PropertySheets" Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
-    <Import Project="$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props" Condition="exists('$(UserRootDir)\Microsoft.Cpp.$(Platform).user.props')" Label="LocalAppDataPlatform" />
-  </ImportGroup>
-  <PropertyGroup Label="UserMacros" />
-  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
-    <LinkIncremental>true</LinkIncremental>
-    <TargetName>mdbx</TargetName>
-    <OutDir>$(SolutionDir)$(Platform)\$(Configuration)\</OutDir>
-    <IntDir>$(SolutionDir)$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
-  </PropertyGroup>
-  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
-    <LinkIncremental>false</LinkIncremental>
-    <OutDir>$(SolutionDir)$(Platform)\$(Configuration)\</OutDir>
-    <TargetName>mdbx</TargetName>
-    <OutDir>$(SolutionDir)$(Platform)\$(Configuration)\</OutDir>
-    <IntDir>$(SolutionDir)$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
-  </PropertyGroup>
-  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
-    <LinkIncremental>true</LinkIncremental>
-    <TargetName>mdbx</TargetName>
-    <OutDir>$(SolutionDir)$(Platform)\$(Configuration)\</OutDir>
-    <IntDir>$(SolutionDir)$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
-  </PropertyGroup>
-  <PropertyGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
-    <LinkIncremental>false</LinkIncremental>
-    <TargetName>mdbx</TargetName>
-    <OutDir>$(SolutionDir)$(Platform)\$(Configuration)\</OutDir>
-    <IntDir>$(SolutionDir)$(Platform)\$(Configuration)\$(ProjectName)\</IntDir>
-  </PropertyGroup>
-  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|Win32'">
-    <ClCompile>
-      <PreprocessorDefinitions>WIN32;_DEBUG;_WINDOWS;_USRDLL;LIBMDBX_EXPORTS;%(PreprocessorDefinitions);MDBX_DEBUG=1</PreprocessorDefinitions>
-      <RuntimeLibrary>MultiThreadedDebugDLL</RuntimeLibrary>
-      <WarningLevel>EnableAllWarnings</WarningLevel>
-      <DebugInformationFormat>ProgramDatabase</DebugInformationFormat>
-      <Optimization>Disabled</Optimization>
-      <StringPooling>true</StringPooling>
-      <TreatWarningAsError>true</TreatWarningAsError>
-    </ClCompile>
-    <Link>
-      <TargetMachine>MachineX86</TargetMachine>
-      <GenerateDebugInformation>true</GenerateDebugInformation>
-      <SubSystem>Windows</SubSystem>
-    </Link>
-  </ItemDefinitionGroup>
-  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|Win32'">
-    <ClCompile>
-      <PreprocessorDefinitions>WIN32;NDEBUG;_WINDOWS;_USRDLL;LIBMDBX_EXPORTS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
-      <RuntimeLibrary>MultiThreadedDLL</RuntimeLibrary>
-      <WarningLevel>EnableAllWarnings</WarningLevel>
-      <DebugInformationFormat>ProgramDatabase</DebugInformationFormat>
-      <StringPooling>true</StringPooling>
-      <Optimization>Full</Optimization>
-      <InlineFunctionExpansion>AnySuitable</InlineFunctionExpansion>
-      <IntrinsicFunctions>true</IntrinsicFunctions>
-      <FavorSizeOrSpeed>Size</FavorSizeOrSpeed>
-      <OmitFramePointers>true</OmitFramePointers>
-      <WholeProgramOptimization>true</WholeProgramOptimization>
-    </ClCompile>
-    <Link>
-      <TargetMachine>MachineX86</TargetMachine>
-      <GenerateDebugInformation>true</GenerateDebugInformation>
-      <SubSystem>Windows</SubSystem>
-      <EnableCOMDATFolding>true</EnableCOMDATFolding>
-      <OptimizeReferences>true</OptimizeReferences>
-      <LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
-    </Link>
-  </ItemDefinitionGroup>
-  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
-    <ClCompile>
-      <DebugInformationFormat>ProgramDatabase</DebugInformationFormat>
-    </ClCompile>
-  </ItemDefinitionGroup>
-  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Debug|x64'">
-    <ClCompile>
-      <WarningLevel>EnableAllWarnings</WarningLevel>
-      <PreprocessorDefinitions>WIN64;_DEBUG;_WINDOWS;_USRDLL;LIBMDBX_EXPORTS;%(PreprocessorDefinitions);MDBX_DEBUG=1</PreprocessorDefinitions>
-      <RuntimeLibrary>MultiThreadedDebugDLL</RuntimeLibrary>
-      <StringPooling>true</StringPooling>
-      <TreatWarningAsError>true</TreatWarningAsError>
-    </ClCompile>
-  </ItemDefinitionGroup>
-  <ItemDefinitionGroup Condition="'$(Configuration)|$(Platform)'=='Release|x64'">
-    <ClCompile>
-      <PreprocessorDefinitions>WIN64;NDEBUG;_WINDOWS;_USRDLL;LIBMDBX_EXPORTS;%(PreprocessorDefinitions)</PreprocessorDefinitions>
-      <RuntimeLibrary>MultiThreadedDLL</RuntimeLibrary>
-      <StringPooling>true</StringPooling>
-      <Optimization>Full</Optimization>
-      <InlineFunctionExpansion>AnySuitable</InlineFunctionExpansion>
-      <IntrinsicFunctions>true</IntrinsicFunctions>
-      <FavorSizeOrSpeed>Size</FavorSizeOrSpeed>
-      <OmitFramePointers>true</OmitFramePointers>
-      <WholeProgramOptimization>true</WholeProgramOptimization>
-      <WarningLevel>EnableAllWarnings</WarningLevel>
-    </ClCompile>
-    <Link>
-      <LinkTimeCodeGeneration>UseLinkTimeCodeGeneration</LinkTimeCodeGeneration>
-    </Link>
-  </ItemDefinitionGroup>
-  <ItemGroup>
-    <ClCompile Include="src\lck-windows.c" />
-    <ClCompile Include="src\mdbx.c" />
-    <ClCompile Include="src\osal.c" />
-    <ClCompile Include="src\version.c" />
-  </ItemGroup>
-  <ItemGroup>
-    <ClInclude Include="mdbx.h" />
-    <ClInclude Include="src\bits.h" />
-    <ClInclude Include="src\defs.h" />
-    <ClInclude Include="src\osal.h" />
-  </ItemGroup>
-  <Import Project="$(VCTargetsPath)\Microsoft.Cpp.targets" />
-  <ImportGroup Label="ExtensionTargets">
-  </ImportGroup>
-</Project>

docs/_preface.md

@@ -0,0 +1,47 @@
+\page intro Introduction
+\section characteristics Characteristics
+
+Preface {#preface}
+------------------
+
+> For the most part, this section is a copy of the corresponding text
+> from LMDB description, but with some edits reflecting the improvements
+> and enhancements were made in MDBX.
+
+MDBX is a Btree-based database management library modeled loosely on the
+BerkeleyDB API, but much simplified. The entire database (aka "environment")
+is exposed in a memory map, and all data fetches return data directly from
+the mapped memory, so no malloc's or memcpy's occur during data fetches.
+As such, the library is extremely simple because it requires no page caching
+layer of its own, and it is extremely high performance and memory-efficient.
+It is also fully transactional with full ACID semantics, and when the memory
+map is read-only, the database integrity cannot be corrupted by stray pointer
+writes from application code.
+
+The library is fully thread-aware and supports concurrent read/write access
+from multiple processes and threads. Data pages use a copy-on-write strategy
+so no active data pages are ever overwritten, which also provides resistance
+to corruption and eliminates the need of any special recovery procedures
+after a system crash. Writes are fully serialized; only one write transaction
+may be active at a time, which guarantees that writers can never deadlock.
+The database structure is multi-versioned so readers run with no locks;
+writers cannot block readers, and readers don't block writers.
+
+Unlike other well-known database mechanisms which use either write-ahead
+transaction logs or append-only data writes, MDBX requires no maintenance
+during operation. Both write-ahead loggers and append-only databases require
+periodic checkpointing and/or compaction of their log or database files
+otherwise they grow without bound. MDBX tracks retired/freed pages within the
+database and re-uses them for new write operations, so the database size does
+not grow without bound in normal use. It is worth noting that the "next"
+version libmdbx (\ref MithrilDB) will solve this problem.
+
+The memory map can be used as a read-only or read-write map. It is read-only
+by default as this provides total immunity to corruption. Using read-write
+mode offers much higher write performance, but adds the possibility for stray
+application writes thru pointers to silently corrupt the database.
+Of course if your application code is known to be bug-free (...) then this is
+not an issue.
+
+If this is your first time using a transactional embedded key-value store,
+you may find the \ref starting section below to be helpful.

docs/_restrictions.md

@@ -0,0 +1,248 @@
+Restrictions & Caveats {#restrictions}
+======================
+In addition to those listed for some functions.
+
+
+## Long-lived read transactions {#long-lived-read}
+Avoid long-lived read transactions, especially in the scenarios with a
+high rate of write transactions. Long-lived read transactions prevents
+recycling pages retired/freed by newer write transactions, thus the
+database can grow quickly.
+
+Understanding the problem of long-lived read transactions requires some
+explanation, but can be difficult for quick perception. So is is
+reasonable to simplify this as follows:
+  1. Garbage collection problem exists in all databases one way or
+     another, e.g. VACUUM in PostgreSQL. But in MDBX it's even more
+     discernible because of high transaction rate and intentional
+     internals simplification in favor of performance.
+
+  2. MDBX employs [Multiversion concurrency control](https://en.wikipedia.org/wiki/Multiversion_concurrency_control)
+     on the [Copy-on-Write](https://en.wikipedia.org/wiki/Copy-on-write)
+     basis, that allows multiple readers runs in parallel with a write
+     transaction without blocking. An each write transaction needs free
+     pages to put the changed data, that pages will be placed in the new
+     b-tree snapshot at commit. MDBX efficiently recycling pages from
+     previous created unused snapshots, BUT this is impossible if anyone
+     a read transaction use such snapshot.
+
+  3. Thus massive altering of data during a parallel long read operation
+     will increase the process's work set and may exhaust entire free
+     database space.
+
+A good example of long readers is a hot backup to the slow destination
+or debugging of a client application while retaining an active read
+transaction. LMDB this results in `MDB_MAP_FULL` error and subsequent write
+performance degradation.
+
+MDBX mostly solve "long-lived" readers issue by offering to use a
+transaction parking-and-ousting approach by \ref mdbx_txn_park(),
+Handle-Slow-Readers \ref MDBX_hsr_func callback which allows to abort
+long-lived read transactions, and using the \ref MDBX_LIFORECLAIM mode
+which addresses subsequent performance degradation. The "next" version
+of libmdbx (aka \ref MithrilDB) will completely solve this.
+
+Nonetheless, situations that encourage lengthy read transactions while
+intensively updating data should be avoided. For example, you should
+avoid suspending/blocking processes/threads performing read
+transactions, including during debugging, and use transaction parking if
+necessary.
+
+You should also beware of aborting processes that perform reading
+transactions. Despite the fact that libmdbx automatically checks and
+cleans readers, as an a process aborting (especially with core dump) can
+take a long time, and checking readers cannot be performed too often due
+to performance degradation.
+
+This issue will be addressed in MithrilDB and one of libmdbx releases,
+presumably in 2025. To do this, nonlinear GC recycling will be
+implemented, without stopping garbage recycling on the old MVCC snapshot
+used by a long read transaction.
+
+After the planned implementation, any long-term reading transaction will
+still keep the used MVCC-snapshot (all the database pages forming it)
+from being recycled, but it will allow all unused MVCC snapshots to be
+recycled, both before and after the readable one. This will eliminate
+one of the main architectural flaws inherited from LMDB and caused the
+growth of a database in proportion to a volume of data changes made
+concurrently with a long-running read transaction.
+
+
+## Large data items
+
+MDBX allows you to store values up to 1 gigabyte in size, but this is
+not the main functionality for a key-value storage, but an additional
+feature that should not be abused. Such long values are stored in
+consecutive/adjacent DB pages, which has both pros and cons. This allows
+you to read long values directly without copying and without any
+overhead from a linear section of memory.
+
+On the other hand, when putting such values in the database, it is
+required to find a sufficient number of free consecutive/adjacent
+database pages, which can be very difficult and expensive, moreover
+sometimes impossible since b-tree tends to fragmentation. So, when
+placing very long values, the engine may need to process the entire GC,
+and in the absence of a sufficient sequence of free pages, increase the
+DB file. Thus, for long values, MDBX provides maximum read performance
+at the expense of write performance.
+
+Some aspects related to GC have been refined and improved in 2022 within
+the first releases of the 0.12.x series. In particular the search for
+free consecutive/adjacent pages through GC has been significantly
+speeded, including acceleration using NOEN/SSE2/AVX2/AVX512
+instructions.
+
+This issue will be addressed in MithrilDB and refined within one of
+0.15.x libmdbx releases, presumably at end of 2025.
+
+
+### Huge transactions
+
+A similar situation can be with huge transactions, in which a lot of
+database pages are retired. The retired pages should be put into GC as a
+list of page numbers for future reuse. But in huge transactions, such a
+list of retired page numbers can also be huge, i.e. it is a very long
+value and requires a long sequence of free pages to be saved. Thus, if
+you delete large amounts of information from the database in a single
+transaction, MDBX may need to increase the database file to save the
+list of pages to be retired.
+
+This issue was fixed in 2022 within the first releases of the 0.12.x
+series by `Big Foot` feature, which now is enabled by default.
+See \ref MDBX_ENABLE_BIGFOOT build-time option.
+
+The `Big Foot` feature which significantly reduces GC overhead for
+processing large lists of retired pages from huge transactions. Now
+libmdbx avoid creating large chunks of PNLs (page number lists) which
+required a long sequences of free pages, aka large/overflow pages. Thus
+avoiding searching, allocating and storing such sequences inside GC.
+
+
+## Space reservation
+An MDBX database configuration will often reserve considerable unused
+memory address space and maybe file size for future growth. This does
+not use actual memory or disk space, but users may need to understand
+the difference so they won't be scared off.
+
+However, on 64-bit systems with a relative small amount of RAM, such
+reservation can deplete system resources (trigger ENOMEM error, etc)
+when setting an inadequately large upper DB size using \ref
+mdbx_env_set_geometry() or \ref mdbx::env::geometry. So just avoid this.
+
+
+## Remote filesystems
+Do not use MDBX databases on remote filesystems, even between processes
+on the same host. This breaks file locks on some platforms, possibly
+memory map sync, and certainly sync between programs on different hosts.
+
+On the other hand, MDBX support the exclusive database operation over
+a network, and cooperative read-only access to the database placed on
+a read-only network shares.
+
+
+## Child processes
+Do not use opened \ref MDBX_env instance(s) in a child processes after `fork()`.
+It would be insane to call fork() and any MDBX-functions simultaneously
+from multiple threads. The best way is to prevent the presence of open
+MDBX-instances during `fork()`.
+
+The \ref MDBX_ENV_CHECKPID build-time option, which is ON by default on
+non-Windows platforms (i.e. where `fork()` is available), enables PID
+checking at a few critical points. But this does not give any guarantees,
+but only allows you to detect such errors a little sooner. Depending on
+the platform, you should expect an application crash and/or database
+corruption in such cases.
+
+On the other hand, MDBX allow calling \ref mdbx_env_close() in such cases to
+release resources, but no more and in general this is a wrong way.
+
+#### Since v0.13.1 and later
+
+Starting from the v0.13.1 release, the \ref mdbx_env_resurrect_after_fork()
+is available, which allows you to reuse an already open database
+environment in child processes, but strictly without inheriting any
+transactions from a parent process.
+
+
+## Read-only mode
+There is no pure read-only mode in a normal explicitly way, since
+readers need write access to LCK-file to be ones visible for writer.
+
+So MDBX always tries to open/create LCK-file for read-write, but switches
+to without-LCK mode on appropriate errors (`EROFS`, `EACCESS`, `EPERM`)
+if the read-only mode was requested by the \ref MDBX_RDONLY flag which is
+described below.
+
+
+## Troubleshooting the LCK-file
+1. A broken LCK-file can cause sync issues, including appearance of
+  wrong/inconsistent data for readers. When database opened in the
+  cooperative read-write mode the LCK-file requires to be mapped to
+  memory in read-write access. In this case it is always possible for
+  stray/malfunctioned application could writes thru pointers to
+  silently corrupt the LCK-file.
+
+  Unfortunately, there is no any portable way to prevent such
+  corruption, since the LCK-file is updated concurrently by
+  multiple processes in a lock-free manner and any locking is
+  unwise due to a large overhead.
+
+  \note Workaround: Just make all programs using the database close it;
+  the LCK-file is always reset on first open.
+
+2. Stale reader transactions left behind by an aborted program cause
+  further writes to grow the database quickly, and stale locks can
+  block further operation.
+  MDBX checks for stale readers while opening environment and before
+  growth the database. But in some cases, this may not be enough.
+
+  \note Workaround: Check for stale readers periodically, using the
+  \ref mdbx_reader_check() function or the mdbx_stat tool.
+
+3. Stale writers will be cleared automatically by MDBX on supported
+  platforms. But this is platform-specific, especially of
+  implementation of shared POSIX-mutexes and support for robust
+  mutexes. For instance there are no known issues on Linux, OSX,
+  Windows and FreeBSD.
+
+  \note Workaround: Otherwise just make all programs using the database
+  close it; the LCK-file is always reset on first open of the environment.
+
+
+## One thread - One transaction
+A thread can only use one transaction at a time, plus any nested
+read-write transactions in the non-writemap mode. Each transaction
+belongs to one thread. The \ref MDBX_NOSTICKYTHREADS flag changes this,
+see below.
+
+Do not start more than one transaction for a one thread. If you think
+about this, it's really strange to do something with two data snapshots
+at once, which may be different. MDBX checks and preventing this by
+returning corresponding error code (\ref MDBX_TXN_OVERLAPPING,
+\ref MDBX_BAD_RSLOT, \ref MDBX_BUSY) unless you using
+\ref MDBX_NOSTICKYTHREADS option on the environment.
+Nonetheless, with the `MDBX_NOSTICKYTHREADS` option, you must know
+exactly what you are doing, otherwise you will get deadlocks or reading
+an alien data.
+
+
+## Do not open twice
+Do not have open an MDBX database twice in the same process at the same
+time. By default MDBX prevent this in most cases by tracking databases
+opening and return \ref MDBX_BUSY if anyone LCK-file is already open.
+
+The reason for this is that when the "Open file description" locks (aka
+OFD-locks) are not available, MDBX uses POSIX locks on files, and these
+locks have issues if one process opens a file multiple times. If a single
+process opens the same environment multiple times, closing it once will
+remove all the locks held on it, and the other instances will be
+vulnerable to corruption from other processes.
+
+For compatibility with LMDB which allows multi-opening, MDBX can be
+configured at runtime by `mdbx_setup_debug(MDBX_DBG_LEGACY_MULTIOPEN, ...)`
+prior to calling other MDBX functions. In this way MDBX will track
+databases opening, detect multi-opening cases and then recover POSIX file
+locks as necessary. However, lock recovery can cause unexpected pauses,
+such as when another process opened the database in exclusive mode before
+the lock was restored - we have to wait until such a process releases the
+database, and so on.

docs/_starting.md

@@ -0,0 +1,245 @@
+Getting started {#starting}
+===============
+
+> This section is based on Bert Hubert's intro "LMDB Semantics", with
+> edits reflecting the improvements and enhancements were made in MDBX.
+> See Bert Hubert's [original](https://github.com/ahupowerdns/ahutils/blob/master/lmdb-semantics.md).
+
+Everything starts with an environment, created by \ref mdbx_env_create().
+Once created, this environment must also be opened with \ref mdbx_env_open(),
+and after use be closed by \ref mdbx_env_close(). At that a non-zero value
+of the last argument "mode" supposes MDBX will create database and directory
+if ones does not exist. In this case the non-zero "mode" argument specifies
+the file mode bits be applied when a new files are created by `open()` function.
+
+Within that directory, a lock file (aka LCK-file) and a storage file (aka
+DXB-file) will be generated. If you don't want to use a directory, you can
+pass the \ref MDBX_NOSUBDIR option, in which case the path you provided is used
+directly as the DXB-file, and another file with a "-lck" suffix added
+will be used for the LCK-file.
+
+Once the environment is open, a transaction can be created within it using
+\ref mdbx_txn_begin(). Transactions may be read-write or read-only, and read-write
+transactions may be nested. A transaction must only be used by one thread at
+a time. Transactions are always required, even for read-only access. The
+transaction provides a consistent view of the data.
+
+Once a transaction has been created, a database (i.e. key-value space inside
+the environment) can be opened within it using \ref mdbx_dbi_open(). If only one
+database will ever be used in the environment, a `NULL` can be passed as the
+database name. For named databases, the \ref MDBX_CREATE flag must be used to
+create the database if it doesn't already exist. Also, \ref mdbx_env_set_maxdbs()
+must be called after \ref mdbx_env_create() and before \ref mdbx_env_open() to set
+the maximum number of named databases you want to support.
+
+\note A single transaction can open multiple databases. Generally databases
+should only be opened once, by the first transaction in the process.
+
+Within a transaction, \ref mdbx_get() and \ref mdbx_put() can store single key-value
+pairs if that is all you need to do (but see \ref Cursors below if you want to do
+more).
+
+A key-value pair is expressed as two \ref MDBX_val structures. This struct that is
+exactly similar to POSIX's `struct iovec` and has two fields, `iov_len` and
+`iov_base`. The data is a `void` pointer to an array of `iov_len` bytes.
+\note The notable difference between MDBX and LMDB is that MDBX support zero
+length keys.
+
+Because MDBX is very efficient (and usually zero-copy), the data returned in
+an \ref MDBX_val structure may be memory-mapped straight from disk. In other words
+look but do not touch (or `free()` for that matter). Once a transaction is
+closed, the values can no longer be used, so make a copy if you need to keep
+them after that.
+
+## Cursors {#Cursors}
+To do more powerful things, we must use a cursor.
+
+Within the transaction, a cursor can be created with \ref mdbx_cursor_open().
+With this cursor we can store/retrieve/delete (multiple) values using
+\ref mdbx_cursor_get(), \ref mdbx_cursor_put() and \ref mdbx_cursor_del().
+
+The \ref mdbx_cursor_get() positions itself depending on the cursor operation
+requested, and for some operations, on the supplied key. For example, to list
+all key-value pairs in a database, use operation \ref MDBX_FIRST for the first
+call to \ref mdbx_cursor_get(), and \ref MDBX_NEXT on subsequent calls, until
+the end is hit.
+
+To retrieve all keys starting from a specified key value, use \ref MDBX_SET. For
+more cursor operations, see the \ref c_api reference.
+
+When using \ref mdbx_cursor_put(), either the function will position the cursor
+for you based on the key, or you can use operation \ref MDBX_CURRENT to use the
+current position of the cursor. \note Note that key must then match the current
+position's key.
+
+
+## Summarizing the opening
+
+So we have a cursor in a transaction which opened a database in an
+environment which is opened from a filesystem after it was separately
+created.
+
+Or, we create an environment, open it from a filesystem, create a transaction
+within it, open a database within that transaction, and create a cursor
+within all of the above.
+
+Got it?
+
+
+## Threads and processes
+
+Do not have open an database twice in the same process at the same time, MDBX
+will track and prevent this. Instead, share the MDBX environment that has
+opened the file across all threads. The reason for this is:
+ - When the "Open file description" locks (aka OFD-locks) are not available,
+   MDBX uses POSIX locks on files, and these locks have issues if one process
+   opens a file multiple times.
+ - If a single process opens the same environment multiple times, closing it
+   once will remove all the locks held on it, and the other instances will be
+   vulnerable to corruption from other processes.
+ + For compatibility with LMDB which allows multi-opening, MDBX can be
+   configured at runtime by \ref mdbx_setup_debug() with \ref MDBX_DBG_LEGACY_MULTIOPEN` option
+   prior to calling other MDBX functions. In this way MDBX will track
+   databases opening, detect multi-opening cases and then recover POSIX file
+   locks as necessary. However, lock recovery can cause unexpected pauses,
+   such as when another process opened the database in exclusive mode before
+   the lock was restored - we have to wait until such a process releases the
+   database, and so on.
+
+Do not use opened MDBX environment(s) after `fork()` in a child process(es),
+MDBX will check and prevent this at critical points. Instead, ensure there is
+no open MDBX-instance(s) during fork(), or at least close it immediately after
+`fork()` in the child process and reopen if required - for instance by using
+`pthread_atfork()`. The reason for this is:
+ - For competitive consistent reading, MDBX assigns a slot in the shared
+   table for each process that interacts with the database. This slot is
+   populated with process attributes, including the PID.
+ - After `fork()`, in order to remain connected to a database, the child
+   process must have its own such "slot", which can't be assigned in any
+   simple and robust way another than the regular.
+ - A write transaction from a parent process cannot continue in a child
+   process for obvious reasons.
+ - Moreover, in a multithreaded process at the fork() moment any number of
+   threads could run in critical and/or intermediate sections of MDBX code
+   with interaction and/or racing conditions with threads from other
+   process(es). For instance: shrinking a database or copying it to a pipe,
+   opening or closing environment, beginning or finishing a transaction,
+   and so on.
+ = Therefore, any solution other than simply close database (and reopen if
+   necessary) in a child process would be both extreme complicated and so
+   fragile.
+
+Do not start more than one transaction for a one thread. If you think
+about this, it's really strange to do something with two data snapshots
+at once, which may be different. MDBX checks and preventing this by
+returning corresponding error code (\ref MDBX_TXN_OVERLAPPING,
+\ref MDBX_BAD_RSLOT, \ref MDBX_BUSY) unless you using
+\ref MDBX_NOSTICKYTHREADS option on the environment. Nonetheless,
+with the \ref MDBX_NOSTICKYTHREADS option, you must know exactly what
+you are doing, otherwise you will get deadlocks or reading an alien
+data.
+
+Also note that a transaction is tied to one thread by default using
+Thread Local Storage. If you want to pass transactions across threads,
+you can use the \ref MDBX_NOSTICKYTHREADS option on the environment.
+Nevertheless, a write transaction must be committed or aborted in the
+same thread which it was started. MDBX checks this in a reasonable
+manner and return the \ref MDBX_THREAD_MISMATCH error in rules
+violation.
+
+
+## Transactions, rollbacks etc
+
+To actually get anything done, a transaction must be committed using
+\ref mdbx_txn_commit(). Alternatively, all of a transaction's operations
+can be discarded using \ref mdbx_txn_abort().
+
+\attention An important difference between MDBX and LMDB is that MDBX required
+that any opened cursors can be reused and must be freed explicitly, regardless
+ones was opened in a read-only or write transaction. The REASON for this is
+eliminates ambiguity which helps to avoid errors such as: use-after-free,
+double-free, i.e. memory corruption and segfaults.
+
+For read-only transactions, obviously there is nothing to commit to storage.
+\attention An another notable difference between MDBX and LMDB is that MDBX make
+handles opened for existing databases immediately available for other
+transactions, regardless this transaction will be aborted or reset. The
+REASON for this is to avoiding the requirement for multiple opening a same
+handles in concurrent read transactions, and tracking of such open but hidden
+handles until the completion of read transactions which opened them.
+
+In addition, as long as a transaction is open, a consistent view of the
+database is kept alive, which requires storage. A read-only transaction that
+no longer requires this consistent view should be terminated (committed or
+aborted) when the view is no longer needed (but see below for an
+optimization).
+
+There can be multiple simultaneously active read-only transactions but only
+one that can write. Once a single read-write transaction is opened, all
+further attempts to begin one will block until the first one is committed or
+aborted. This has no effect on read-only transactions, however, and they may
+continue to be opened at any time.
+
+
+## Duplicate keys aka Multi-values
+
+\ref mdbx_get() and \ref mdbx_put() respectively have no and only some support or
+multiple key-value pairs with identical keys. If there are multiple values
+for a key, \ref mdbx_get() will only return the first value.
+
+When multiple values for one key are required, pass the \ref MDBX_DUPSORT flag to
+\ref mdbx_dbi_open(). In an \ref MDBX_DUPSORT database, by default \ref mdbx_put() will
+not replace the value for a key if the key existed already. Instead it will add
+the new value to the key. In addition, \ref mdbx_del() will pay attention to the
+value field too, allowing for specific values of a key to be deleted.
+
+Finally, additional cursor operations become available for traversing through
+and retrieving duplicate values.
+
+
+## Some optimization
+
+If you frequently begin and abort read-only transactions, as an optimization,
+it is possible to only reset and renew a transaction.
+
+\ref mdbx_txn_reset() releases any old copies of data kept around for a read-only
+transaction. To reuse this reset transaction, call \ref mdbx_txn_renew() on it.
+Any cursors in this transaction can also be renewed using \ref mdbx_cursor_renew()
+or freed by \ref mdbx_cursor_close().
+
+To permanently free a transaction, reset or not, use \ref mdbx_txn_abort().
+
+
+## Cleaning up
+
+Any created cursors must be closed using \ref mdbx_cursor_close(). It is advisable
+to repeat:
+\note An important difference between MDBX and LMDB is that MDBX required that
+any opened cursors can be reused and must be freed explicitly, regardless
+ones was opened in a read-only or write transaction. The REASON for this is
+eliminates ambiguity which helps to avoid errors such as: use-after-free,
+double-free, i.e. memory corruption and segfaults.
+
+It is very rarely necessary to close a database handle, and in general they
+should just be left open. When you close a handle, it immediately becomes
+unavailable for all transactions in the environment. Therefore, you should
+avoid closing the handle while at least one transaction is using it.
+
+
+## Now read up on the full API!
+
+The full \ref c_api documentation lists further details below, like how to:
+
+- Configure database size and automatic size management: \ref mdbx_env_set_geometry().
+- Drop and clean a database: \ref mdbx_drop().
+- Detect and report errors: \ref c_err.
+- Optimize (bulk) loading speed: \ref MDBX_MULTIPLE, \ref MDBX_APPEND.
+- Reduce (temporarily) robustness to gain even more speed: \ref sync_modes.
+- Gather statistics about the database: \ref c_statinfo.
+- Sstimate size of range query result: \ref c_rqest.
+- Double performance by LIFO reclaiming on storages with write-back: \ref MDBX_LIFORECLAIM.
+- Use sequences and canary markers: \ref mdbx_dbi_sequence(), \ref MDBX_canary.
+- Use Handle-Slow-Readers callback to resolve a database full/overflow issues
+  due to long-lived read transactions: \ref mdbx_env_set_hsr().
+- Use exclusive mode: \ref MDBX_EXCLUSIVE.
+- Define custom sort orders (but this is recommended to be avoided).

docs/_toc.md

@@ -0,0 +1,37 @@
+The source code is availale on [Gitflic](https://gitflic.ru/project/erthink/libmdbx).
+Donations are welcome to ETH `0xD104d8f8B2dC312aaD74899F83EBf3EEBDC1EA3A`.
+Всё будет хорошо!
+
+> Questions, feedback and suggestions are welcome to the [Telegram' group](https://t.me/libmdbx) (archive [1](https://libmdbx.dqdkfa.ru/tg-archive/messages1.html),
+> [2](https://libmdbx.dqdkfa.ru/tg-archive/messages2.html), [3](https://libmdbx.dqdkfa.ru/tg-archive/messages3.html), [4](https://libmdbx.dqdkfa.ru/tg-archive/messages4.html),
+> [5](https://libmdbx.dqdkfa.ru/tg-archive/messages5.html), [6](https://libmdbx.dqdkfa.ru/tg-archive/messages6.html), [7](https://libmdbx.dqdkfa.ru/tg-archive/messages7.html)).
+> See the [ChangeLog](https://gitflic.ru/project/erthink/libmdbx/blob?file=ChangeLog.md) for `NEWS` and latest updates.
+
+\section toc Table of Contents
+
+This manual is divided into parts,
+each of which is divided into several sections.
+
+1. The \ref intro
+  - \ref characteristics
+  - \ref improvements
+  - \ref restrictions
+  - \ref performance
+2. \ref usage
+  - \ref getting
+  - \ref starting
+  - \ref bindings
+
+3. The `C/C++` API manual:
+  - The \ref c_api reference
+  - \ref c_crud_hints "Quick reference for Insert/Update/Delete operations"
+  - The \ref mdbx.h header file reference
+  - The \ref cxx_api reference
+  - The \ref mdbx.h++ header file reference
+
+Please do not hesitate to point out errors in the documentation,
+including creating [merge-request](https://gitflic.ru/project/erthink/libmdbx/merge-request) with corrections and improvements.
+
+---
+
+\section MithrilDB MithrilDB and Future

docs/header.html

@@ -0,0 +1,103 @@
+<!DOCTYPE html>
+<html xmlns="http://www.w3.org/1999/xhtml" lang="$langISO">
+<head>
+<meta http-equiv="Content-Type" content="text/xhtml;charset=UTF-8"/>
+<meta http-equiv="X-UA-Compatible" content="IE=11"/>
+<meta name="generator" content="Doxygen $doxygenversion"/>
+<meta name="viewport" content="width=device-width, initial-scale=1"/>
+<link rel="icon" href="favicon.ico">
+<link rel="icon" href="img/bear.png" type="image/png">
+<link rel="apple-touch-icon" href="img/bear.png">
+<meta property="og:type" content="article"/>
+<meta property="og:url" content="https://libmdbx.dqdkfa.ru/"/>
+<meta name="twitter:title" content="One of the fastest embeddable key-value engine"/>
+<meta name="twitter:description" content="MDBX surpasses the legendary LMDB in terms of reliability, features and performance. For now libmdbx is chosen by all modern Ethereum frontiers as a storage engine."/>
+<!--BEGIN PROJECT_NAME--><title>$projectname: $title</title><!--END PROJECT_NAME-->
+<!--BEGIN !PROJECT_NAME--><title>$title</title><!--END !PROJECT_NAME-->
+<!--BEGIN PROJECT_ICON-->
+<link rel="icon" href="$relpath^$projecticon" type="image/x-icon" />
+<!--END PROJECT_ICON-->
+<link href="$relpath^tabs.css" rel="stylesheet" type="text/css"/>
+<!--BEGIN DISABLE_INDEX-->
+  <!--BEGIN FULL_SIDEBAR-->
+<script type="text/javascript">var page_layout=1;</script>
+  <!--END FULL_SIDEBAR-->
+<!--END DISABLE_INDEX-->
+<script type="text/javascript" src="$relpath^jquery.js"></script>
+<script type="text/javascript" src="$relpath^dynsections.js"></script>
+<!--BEGIN COPY_CLIPBOARD-->
+<script type="text/javascript" src="$relpath^clipboard.js"></script>
+<!--END COPY_CLIPBOARD-->
+$treeview
+$search
+$mathjax
+$darkmode
+<link href="$relpath^$stylesheet" rel="stylesheet" type="text/css" />
+$extrastylesheet
+</head>
+<body>
+<!--BEGIN DISABLE_INDEX-->
+  <!--BEGIN FULL_SIDEBAR-->
+<div id="side-nav" class="ui-resizable side-nav-resizable"><!-- do not remove this div, it is closed by doxygen! -->
+  <!--END FULL_SIDEBAR-->
+<!--END DISABLE_INDEX-->
+
+<div id="top"><!-- do not remove this div, it is closed by doxygen! -->
+
+<!--BEGIN TITLEAREA-->
+<div id="titlearea">
+<table cellspacing="0" cellpadding="0">
+ <tbody>
+ <tr id="projectrow">
+  <!--BEGIN PROJECT_LOGO-->
+  <td id="projectlogo"><img alt="Logo" src="$relpath^$projectlogo"$logosize/></td>
+  <!--END PROJECT_LOGO-->
+  <!--BEGIN PROJECT_NAME-->
+  <td id="projectalign">
+   <div id="projectname">$projectname<!--BEGIN PROJECT_NUMBER--><span id="projectnumber">&#160;$projectnumber</span><!--END PROJECT_NUMBER-->
+   </div>
+   <!--BEGIN PROJECT_BRIEF--><div id="projectbrief">$projectbrief</div><!--END PROJECT_BRIEF-->
+  </td>
+  <!--END PROJECT_NAME-->
+  <!--BEGIN !PROJECT_NAME-->
+   <!--BEGIN PROJECT_BRIEF-->
+    <td>
+    <div id="projectbrief">$projectbrief</div>
+    </td>
+   <!--END PROJECT_BRIEF-->
+  <!--END !PROJECT_NAME-->
+  <!--BEGIN DISABLE_INDEX-->
+   <!--BEGIN SEARCHENGINE-->
+     <!--BEGIN !FULL_SIDEBAR-->
+    <td>$searchbox</td>
+     <!--END !FULL_SIDEBAR-->
+   <!--END SEARCHENGINE-->
+  <!--END DISABLE_INDEX-->
+ </tr>
+  <!--BEGIN SEARCHENGINE-->
+   <!--BEGIN FULL_SIDEBAR-->
+   <tr><td colspan="2">$searchbox</td></tr>
+   <!--END FULL_SIDEBAR-->
+  <!--END SEARCHENGINE-->
+ </tbody>
+</table>
+</div>
+<!--END TITLEAREA-->
+<!-- Yandex.Metrika counter -->
+<script type="text/javascript" >
+   (function(m,e,t,r,i,k,a){m[i]=m[i]||function(){(m[i].a=m[i].a||[]).push(arguments)};
+   m[i].l=1*new Date();
+   for (var j = 0; j < document.scripts.length; j++) {if (document.scripts[j].src === r) { return; }}
+   k=e.createElement(t),a=e.getElementsByTagName(t)[0],k.async=1,k.src=r,a.parentNode.insertBefore(k,a)})
+   (window, document, "script", "https://mc.yandex.ru/metrika/tag.js", "ym");
+
+   ym(99261645, "init", {
+        clickmap:true,
+        trackLinks:true,
+        accurateTrackBounce:true,
+        webvisor:true
+   });
+</script>
+<noscript><div><img src="https://mc.yandex.ru/watch/99261645" style="position:absolute; left:-9999px;" alt="" /></div></noscript>
+<!-- /Yandex.Metrika counter -->
+<!-- end header part -->

docs/ld+json

@@ -0,0 +1,27 @@
+<script type="application/ld+json">
+{
+  "@context": "https://schema.org",
+  "@type": "ItemList",
+  "itemListElement": [{
+    "@type": "ListItem",
+    "position": 1,
+    "name": "Группа в Telegram",
+    "url": "https://t.me/libmdbx"
+  },{
+    "@type": "ListItem",
+    "position": 2,
+    "name": "Исходный код",
+    "url": "https://gitflic.ru/project/erthink/libmdbx"
+  },{
+    "@type": "ListItem",
+    "position": 3,
+    "name": "C++ API",
+    "url": "https://libmdbx.dqdkfa.ru/group__cxx__api.html"
+  },{
+    "@type": "ListItem",
+    "position": 4,
+    "name": "Mirror on Github",
+    "url": "https://github.com/erthink/libmdbx"
+  }]
+}
+</script>

docs/manifest.webmanifest

@@ -0,0 +1,6 @@
+{
+  "icons": [
+    { "src": "favicon.ico", "type": "image/ico", "sizes": "32x32" },
+    { "src": "img/bear.png", "type": "image/png", "sizes": "256x256" }
+  ]
+}

docs/title

@@ -0,0 +1,2 @@
+<title>libmdbx: One of the fastest embeddable key-value engine</title>
+<meta name="description" content="libmdbx surpasses the legendary LMDB in terms of reliability, features and performance. For now libmdbx is chosen by all modern Ethereum frontiers as a storage engine.">

example/CMakeLists.txt

@@ -0,0 +1,6 @@
+set(TARGET mdbx_example)
+project(${TARGET})
+
+add_executable(${TARGET} example-mdbx.c)
+
+target_link_libraries(${TARGET} mdbx)

example/README.md

@@ -0,0 +1 @@
+See [example-mdbx.c](example-mdbx.c) as an example of using _libmdbx_, and do a line-by-line comparison of it with the [sample-bdb.txt](sample-bdb.txt) file.

example/example-mdbx.c

@@ -0,0 +1,154 @@
+/* MDBX usage example
+ *
+ * Do a line-by-line comparison of this and sample-bdb.txt
+ */
+
+/*
+ * Copyright 2015-2025 Leonid Yuriev <leo@yuriev.ru>.
+ * Copyright 2017 Ilya Shipitsin <chipitsine@gmail.com>.
+ * Copyright 2012-2015 Howard Chu, Symas Corp.
+ * All rights reserved.
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted only as authorized by the OpenLDAP
+ * Public License.
+ *
+ * A copy of this license is available in the file LICENSE in the
+ * top-level directory of the distribution or, alternatively, at
+ * <http://www.OpenLDAP.org/license.html>.
+ */
+
+#if (defined(__MINGW__) || defined(__MINGW32__) || defined(__MINGW64__)) && !defined(__USE_MINGW_ANSI_STDIO)
+#define __USE_MINGW_ANSI_STDIO 1
+#endif /* MinGW */
+
+#include "mdbx.h"
+
+#include <limits.h>
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+int main(int argc, char *argv[]) {
+  (void)argc;
+  (void)argv;
+
+  int rc;
+  MDBX_env *env = NULL;
+  MDBX_dbi dbi = 0;
+  MDBX_val key, data;
+  MDBX_txn *txn = NULL;
+  MDBX_cursor *cursor = NULL;
+  char sval[32];
+
+  printf("MDBX limits:\n");
+#if UINTPTR_MAX > 0xffffFFFFul || ULONG_MAX > 0xffffFFFFul
+  const double scale_factor = 1099511627776.0;
+  const char *const scale_unit = "TiB";
+#else
+  const double scale_factor = 1073741824.0;
+  const char *const scale_unit = "GiB";
+#endif
+  const size_t pagesize_min = mdbx_limits_pgsize_min();
+  const size_t pagesize_max = mdbx_limits_pgsize_max();
+  const size_t pagesize_default = mdbx_default_pagesize();
+
+  printf("\tPage size: a power of 2, minimum %zu, maximum %zu bytes,"
+         " default %zu bytes.\n",
+         pagesize_min, pagesize_max, pagesize_default);
+  printf("\tKey size: minimum %zu, maximum ≈¼ pagesize (%zu bytes for default"
+         " %zuK pagesize, %zu bytes for %zuK pagesize).\n",
+         (size_t)0, mdbx_limits_keysize_max(-1, MDBX_DB_DEFAULTS), pagesize_default / 1024,
+         mdbx_limits_keysize_max(pagesize_max, MDBX_DB_DEFAULTS), pagesize_max / 1024);
+  printf("\tValue size: minimum %zu, maximum %zu (0x%08zX) bytes for maps,"
+         " ≈¼ pagesize for multimaps (%zu bytes for default %zuK pagesize,"
+         " %zu bytes for %zuK pagesize).\n",
+         (size_t)0, mdbx_limits_valsize_max(pagesize_min, MDBX_DB_DEFAULTS),
+         mdbx_limits_valsize_max(pagesize_min, MDBX_DB_DEFAULTS), mdbx_limits_valsize_max(-1, MDBX_DUPSORT),
+         pagesize_default / 1024, mdbx_limits_valsize_max(pagesize_max, MDBX_DUPSORT), pagesize_max / 1024);
+  printf("\tWrite transaction size: up to %zu (0x%zX) pages (%f %s for default "
+         "%zuK pagesize, %f %s for %zuK pagesize).\n",
+         mdbx_limits_txnsize_max(pagesize_min) / pagesize_min, mdbx_limits_txnsize_max(pagesize_min) / pagesize_min,
+         mdbx_limits_txnsize_max(-1) / scale_factor, scale_unit, pagesize_default / 1024,
+         mdbx_limits_txnsize_max(pagesize_max) / scale_factor, scale_unit, pagesize_max / 1024);
+  printf("\tDatabase size: up to %zu pages (%f %s for default %zuK "
+         "pagesize, %f %s for %zuK pagesize).\n",
+         mdbx_limits_dbsize_max(pagesize_min) / pagesize_min, mdbx_limits_dbsize_max(-1) / scale_factor, scale_unit,
+         pagesize_default / 1024, mdbx_limits_dbsize_max(pagesize_max) / scale_factor, scale_unit, pagesize_max / 1024);
+  printf("\tMaximum sub-databases: %u.\n", MDBX_MAX_DBI);
+  printf("-----\n");
+
+  rc = mdbx_env_create(&env);
+  if (rc != MDBX_SUCCESS) {
+    fprintf(stderr, "mdbx_env_create: (%d) %s\n", rc, mdbx_strerror(rc));
+    goto bailout;
+  }
+  rc = mdbx_env_open(env, "./example-db", MDBX_NOSUBDIR | MDBX_LIFORECLAIM, 0664);
+  if (rc != MDBX_SUCCESS) {
+    fprintf(stderr, "mdbx_env_open: (%d) %s\n", rc, mdbx_strerror(rc));
+    goto bailout;
+  }
+
+  rc = mdbx_txn_begin(env, NULL, 0, &txn);
+  if (rc != MDBX_SUCCESS) {
+    fprintf(stderr, "mdbx_txn_begin: (%d) %s\n", rc, mdbx_strerror(rc));
+    goto bailout;
+  }
+  rc = mdbx_dbi_open(txn, NULL, 0, &dbi);
+  if (rc != MDBX_SUCCESS) {
+    fprintf(stderr, "mdbx_dbi_open: (%d) %s\n", rc, mdbx_strerror(rc));
+    goto bailout;
+  }
+
+  key.iov_len = sizeof(int);
+  key.iov_base = sval;
+  data.iov_len = sizeof(sval);
+  data.iov_base = sval;
+
+  sprintf(sval, "%03x %d foo bar", 32, 3141592);
+  rc = mdbx_put(txn, dbi, &key, &data, 0);
+  if (rc != MDBX_SUCCESS) {
+    fprintf(stderr, "mdbx_put: (%d) %s\n", rc, mdbx_strerror(rc));
+    goto bailout;
+  }
+  rc = mdbx_txn_commit(txn);
+  if (rc) {
+    fprintf(stderr, "mdbx_txn_commit: (%d) %s\n", rc, mdbx_strerror(rc));
+    goto bailout;
+  }
+  txn = NULL;
+
+  rc = mdbx_txn_begin(env, NULL, MDBX_TXN_RDONLY, &txn);
+  if (rc != MDBX_SUCCESS) {
+    fprintf(stderr, "mdbx_txn_begin: (%d) %s\n", rc, mdbx_strerror(rc));
+    goto bailout;
+  }
+  rc = mdbx_cursor_open(txn, dbi, &cursor);
+  if (rc != MDBX_SUCCESS) {
+    fprintf(stderr, "mdbx_cursor_open: (%d) %s\n", rc, mdbx_strerror(rc));
+    goto bailout;
+  }
+
+  int found = 0;
+  while ((rc = mdbx_cursor_get(cursor, &key, &data, MDBX_NEXT)) == 0) {
+    printf("key: %p %.*s, data: %p %.*s\n", key.iov_base, (int)key.iov_len, (char *)key.iov_base, data.iov_base,
+           (int)data.iov_len, (char *)data.iov_base);
+    found += 1;
+  }
+  if (rc != MDBX_NOTFOUND || found == 0) {
+    fprintf(stderr, "mdbx_cursor_get: (%d) %s\n", rc, mdbx_strerror(rc));
+    goto bailout;
+  } else {
+    rc = MDBX_SUCCESS;
+  }
+bailout:
+  if (cursor)
+    mdbx_cursor_close(cursor);
+  if (txn)
+    mdbx_txn_abort(txn);
+  if (dbi)
+    mdbx_dbi_close(env, dbi);
+  if (env)
+    mdbx_env_close(env);
+  return (rc != MDBX_SUCCESS) ? EXIT_FAILURE : EXIT_SUCCESS;
+}

example/sample-bdb.txt

@@ -1,10 +1,10 @@
-/* sample-bdb.txt - BerkeleyDB toy/sample
+/* BerkeleyDB toy/sample
  *
- * Do a line-by-line comparison of this and sample-mdb.txt
+ * Do a line-by-line comparison of this and example-mdbx.c
  */
 
 /*
- * Copyright 2015-2018 Leonid Yuriev <leo@yuriev.ru>.
+ * Copyright 2015-2025 Leonid Yuriev <leo@yuriev.ru>.
  * Copyright 2012-2015 Howard Chu, Symas Corp.
  * Copyright 2015,2016 Peter-Service R&D LLC.
  * All rights reserved.

libmdbx.config

@@ -1,2 +0,0 @@
-// Add predefined macros for your project here. For example:
-// #define THE_ANSWER 42

libmdbx.creator

@@ -1 +0,0 @@
-[General]

libmdbx.files

@@ -1,58 +0,0 @@
-CMakeLists.txt
-README-RU.md
-pcrf_test/CMakeLists.txt
-src/tools/CMakeLists.txt
-test/CMakeLists.txt
-tutorial/CMakeLists.txt
-tutorial/sample-mdbx.c
-AUTHORS
-LICENSE
-Makefile
-README.md
-TODO.md
-mdbx.h
-src/bits.h
-src/defs.h
-src/lck-posix.c
-src/lck-windows.c
-src/mdbx.c
-src/osal.c
-src/osal.h
-src/tools/mdbx_chk.c
-src/tools/mdbx_copy.1
-src/tools/mdbx_copy.c
-src/tools/mdbx_dump.1
-src/tools/mdbx_dump.c
-src/tools/mdbx_load.1
-src/tools/mdbx_load.c
-src/tools/mdbx_stat.1
-src/tools/mdbx_stat.c
-src/tools/wingetopt.c
-src/tools/wingetopt.h
-src/version.c
-test/actor.cc
-test/base.h
-test/chrono.cc
-test/chrono.h
-test/config.h
-test/dead.cc
-test/hill.cc
-test/jitter.cc
-test/keygen.cc
-test/keygen.h
-test/log.cc
-test/log.h
-test/main.cc
-test/config.cc
-test/cases.cc
-test/osal-unix.cc
-test/osal-windows.cc
-test/osal.h
-test/test.cc
-test/test.h
-test/try.cc
-test/utils.cc
-test/utils.h
-tutorial/README.md
-tutorial/sample-bdb.txt
-tutorial/sample-mdb.txt

libmdbx.includes

@@ -1,6 +0,0 @@
-.
-src
-src/tools
-test
-pcrf_test
-tutorial

mdbx.sln

@@ -1,97 +0,0 @@
-
-Microsoft Visual Studio Solution File, Format Version 12.00
-# Visual Studio 14
-VisualStudioVersion = 14.0.25420.1
-MinimumVisualStudioVersion = 10.0.40219.1
-Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "test", "test\test.vcxproj", "{30E29CE6-E6FC-4D32-AA07-46A55FAF3A31}"
-EndProject
-Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "dll", "dll.vcxproj", "{6D19209B-ECE7-4B9C-941C-0AA2B484F199}"
-EndProject
-Project("{2150E333-8FDC-42A3-9474-1A3956D46DE8}") = "tools", "tools", "{0A147F9F-22D5-44E6-B389-218CFFB0C524}"
-EndProject
-Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "mdbx_load", "src\tools\mdbx_load.vcxproj", "{15030120-5F7F-48F9-ABE5-DFC814F2A4BB}"
-EndProject
-Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "mdbx_dump", "src\tools\mdbx_dump.vcxproj", "{15030120-5F7F-48F9-ABE5-DFC814F2A4BC}"
-EndProject
-Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "mdbx_copy", "src\tools\mdbx_copy.vcxproj", "{15030120-5F7F-48F9-ABE5-DFC814F2A4BD}"
-EndProject
-Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "mdbx_chk", "src\tools\mdbx_chk.vcxproj", "{15030120-5F7F-48F9-ABE5-DFC814F2A4BE}"
-EndProject
-Project("{8BC9CEB8-8B4A-11D0-8D11-00A0C91BC942}") = "mdbx_stat", "src\tools\mdbx_stat.vcxproj", "{15030120-5F7F-48F9-ABE5-DFC814F2A4BF}"
-EndProject
-Global
-	GlobalSection(SolutionConfigurationPlatforms) = preSolution
-		Debug|x64 = Debug|x64
-		Debug|x86 = Debug|x86
-		Release|x64 = Release|x64
-		Release|x86 = Release|x86
-	EndGlobalSection
-	GlobalSection(ProjectConfigurationPlatforms) = postSolution
-		{30E29CE6-E6FC-4D32-AA07-46A55FAF3A31}.Debug|x64.ActiveCfg = Debug|x64
-		{30E29CE6-E6FC-4D32-AA07-46A55FAF3A31}.Debug|x64.Build.0 = Debug|x64
-		{30E29CE6-E6FC-4D32-AA07-46A55FAF3A31}.Debug|x86.ActiveCfg = Debug|Win32
-		{30E29CE6-E6FC-4D32-AA07-46A55FAF3A31}.Debug|x86.Build.0 = Debug|Win32
-		{30E29CE6-E6FC-4D32-AA07-46A55FAF3A31}.Release|x64.ActiveCfg = Release|x64
-		{30E29CE6-E6FC-4D32-AA07-46A55FAF3A31}.Release|x64.Build.0 = Release|x64
-		{30E29CE6-E6FC-4D32-AA07-46A55FAF3A31}.Release|x86.ActiveCfg = Release|Win32
-		{30E29CE6-E6FC-4D32-AA07-46A55FAF3A31}.Release|x86.Build.0 = Release|Win32
-		{6D19209B-ECE7-4B9C-941C-0AA2B484F199}.Debug|x64.ActiveCfg = Debug|x64
-		{6D19209B-ECE7-4B9C-941C-0AA2B484F199}.Debug|x64.Build.0 = Debug|x64
-		{6D19209B-ECE7-4B9C-941C-0AA2B484F199}.Debug|x86.ActiveCfg = Debug|Win32
-		{6D19209B-ECE7-4B9C-941C-0AA2B484F199}.Debug|x86.Build.0 = Debug|Win32
-		{6D19209B-ECE7-4B9C-941C-0AA2B484F199}.Release|x64.ActiveCfg = Release|x64
-		{6D19209B-ECE7-4B9C-941C-0AA2B484F199}.Release|x64.Build.0 = Release|x64
-		{6D19209B-ECE7-4B9C-941C-0AA2B484F199}.Release|x86.ActiveCfg = Release|Win32
-		{6D19209B-ECE7-4B9C-941C-0AA2B484F199}.Release|x86.Build.0 = Release|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BB}.Debug|x64.ActiveCfg = Debug|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BB}.Debug|x64.Build.0 = Debug|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BB}.Debug|x86.ActiveCfg = Debug|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BB}.Debug|x86.Build.0 = Debug|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BB}.Release|x64.ActiveCfg = Release|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BB}.Release|x64.Build.0 = Release|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BB}.Release|x86.ActiveCfg = Release|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BB}.Release|x86.Build.0 = Release|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BC}.Debug|x64.ActiveCfg = Debug|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BC}.Debug|x64.Build.0 = Debug|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BC}.Debug|x86.ActiveCfg = Debug|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BC}.Debug|x86.Build.0 = Debug|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BC}.Release|x64.ActiveCfg = Release|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BC}.Release|x64.Build.0 = Release|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BC}.Release|x86.ActiveCfg = Release|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BC}.Release|x86.Build.0 = Release|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BD}.Debug|x64.ActiveCfg = Debug|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BD}.Debug|x64.Build.0 = Debug|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BD}.Debug|x86.ActiveCfg = Debug|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BD}.Debug|x86.Build.0 = Debug|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BD}.Release|x64.ActiveCfg = Release|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BD}.Release|x64.Build.0 = Release|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BD}.Release|x86.ActiveCfg = Release|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BD}.Release|x86.Build.0 = Release|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BE}.Debug|x64.ActiveCfg = Debug|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BE}.Debug|x64.Build.0 = Debug|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BE}.Debug|x86.ActiveCfg = Debug|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BE}.Debug|x86.Build.0 = Debug|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BE}.Release|x64.ActiveCfg = Release|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BE}.Release|x64.Build.0 = Release|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BE}.Release|x86.ActiveCfg = Release|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BE}.Release|x86.Build.0 = Release|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BF}.Debug|x64.ActiveCfg = Debug|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BF}.Debug|x64.Build.0 = Debug|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BF}.Debug|x86.ActiveCfg = Debug|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BF}.Debug|x86.Build.0 = Debug|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BF}.Release|x64.ActiveCfg = Release|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BF}.Release|x64.Build.0 = Release|x64
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BF}.Release|x86.ActiveCfg = Release|Win32
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BF}.Release|x86.Build.0 = Release|Win32
-	EndGlobalSection
-	GlobalSection(SolutionProperties) = preSolution
-		HideSolutionNode = FALSE
-	EndGlobalSection
-	GlobalSection(NestedProjects) = preSolution
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BB} = {0A147F9F-22D5-44E6-B389-218CFFB0C524}
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BC} = {0A147F9F-22D5-44E6-B389-218CFFB0C524}
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BD} = {0A147F9F-22D5-44E6-B389-218CFFB0C524}
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BE} = {0A147F9F-22D5-44E6-B389-218CFFB0C524}
-		{15030120-5F7F-48F9-ABE5-DFC814F2A4BF} = {0A147F9F-22D5-44E6-B389-218CFFB0C524}
-	EndGlobalSection
-EndGlobal

packages/buildroot/0001-package-libmdbx-new-package-library-database.patch

@@ -0,0 +1,173 @@
+From 349c08cf21b66ecea851340133a1b845c25675f7 Mon Sep 17 00:00:00 2001
+From: =?UTF-8?q?=D0=9B=D0=B5=D0=BE=D0=BD=D0=B8=D0=B4=20=D0=AE=D1=80=D1=8C?=
+ =?UTF-8?q?=D0=B5=D0=B2=20=28Leonid=20Yuriev=29?= <leo@yuriev.ru>
+Date: Tue, 22 Apr 2025 14:38:49 +0300
+Subject: [PATCH] package/libmdbx: new package (library/database).
+MIME-Version: 1.0
+Content-Type: text/plain; charset=UTF-8
+Content-Transfer-Encoding: 8bit
+
+This patch adds libmdbx:
+ - libmdbx is one of the fastest compact embeddable key-value ACID database.
+ - libmdbx has a specific set of properties and capabilities,
+   focused on creating unique lightweight solutions.
+ - libmdbx surpasses the legendary LMDB (Lightning Memory-Mapped Database)
+   in terms of reliability, features and performance.
+ - more information at https://libmdbx.dqdkfa.ru
+
+The 0.13.6 "Бузина" (Elderberry) is stable release of _libmdbx_ branch with new superior features.
+
+The complete ChangeLog: https://gitflic.ru/project/erthink/libmdbx/blob?file=ChangeLog.md
+
+Signed-off-by: Леонид Юрьев (Leonid Yuriev) <leo@yuriev.ru>
+---
+ DEVELOPERS                   |  3 +++
+ package/Config.in            |  1 +
+ package/libmdbx/Config.in    | 45 ++++++++++++++++++++++++++++++++++++
+ package/libmdbx/libmdbx.hash |  6 +++++
+ package/libmdbx/libmdbx.mk   | 42 +++++++++++++++++++++++++++++++++
+ 5 files changed, 97 insertions(+)
+ create mode 100644 package/libmdbx/Config.in
+ create mode 100644 package/libmdbx/libmdbx.hash
+ create mode 100644 package/libmdbx/libmdbx.mk
+
+diff --git a/DEVELOPERS b/DEVELOPERS
+index 9ab1e125f4..758ff6a2d5 100644
+--- a/DEVELOPERS
++++ b/DEVELOPERS
+@@ -1482,6 +1482,9 @@ N:	Leon Anavi <leon.anavi@konsulko.com>
+ F:	board/olimex/a10_olinuxino
+ F:	configs/olimex_a10_olinuxino_lime_defconfig
+ 
++N:	Leonid Yuriev <leo@yuriev.ru>
++F:	package/libmdbx/
++
+ N:	Lionel Flandrin <lionel@svkt.org>
+ F:	package/python-babel/
+ F:	package/python-daemonize/
+diff --git a/package/Config.in b/package/Config.in
+index 016a99ed1a..a6f95bfaa9 100644
+--- a/package/Config.in
++++ b/package/Config.in
+@@ -1372,6 +1372,7 @@ menu "Database"
+ 	source "package/kompexsqlite/Config.in"
+ 	source "package/leveldb/Config.in"
+ 	source "package/libgit2/Config.in"
++	source "package/libmdbx/Config.in"
+ 	source "package/libodb/Config.in"
+ 	source "package/libodb-boost/Config.in"
+ 	source "package/libodb-mysql/Config.in"
+diff --git a/package/libmdbx/Config.in b/package/libmdbx/Config.in
+new file mode 100644
+index 0000000000..a9a4ac45c5
+--- /dev/null
++++ b/package/libmdbx/Config.in
+@@ -0,0 +1,45 @@
++config BR2_PACKAGE_LIBMDBX
++	bool "libmdbx"
++	depends on BR2_USE_MMU
++	depends on BR2_TOOLCHAIN_HAS_SYNC_4
++	depends on BR2_TOOLCHAIN_HAS_THREADS
++	depends on BR2_TOOLCHAIN_GCC_AT_LEAST_4_4
++	help
++	  One of the fastest compact key-value ACID database
++	  without WAL. libmdbx has a specific set of properties
++	  and capabilities, focused on creating unique lightweight
++	  solutions.
++
++	  libmdbx surpasses the legendary LMDB in terms of
++	  reliability, features and performance.
++
++	  https://libmdbx.dqdkfa.ru
++
++if BR2_PACKAGE_LIBMDBX
++
++config BR2_PACKAGE_LIBMDBX_TOOLS
++	bool "install tools"
++	help
++	  Install libmdbx tools for checking, dump, restore
++	  and show statistics of databases.
++
++config BR2_PACKAGE_LIBMDBX_CXX
++	bool "C++ API"
++	depends on BR2_INSTALL_LIBSTDCPP
++	depends on BR2_TOOLCHAIN_GCC_AT_LEAST_4_8
++	depends on !BR2_TOOLCHAIN_HAS_GCC_BUG_64735
++	help
++	  Enable modern C++11/14/17/20 API for libmdbx.
++
++comment "libmdbx C++ support needs a toolchain w/ C++11, gcc >= 4.8 w/o bug#64735"
++	depends on !BR2_INSTALL_LIBSTDCPP || \
++		!BR2_TOOLCHAIN_GCC_AT_LEAST_4_8 || \
++		BR2_TOOLCHAIN_HAS_GCC_BUG_64735
++
++endif
++
++comment "libmdbx needs MMU, a toolchain w/ threads, gcc >= 4.4 w/ 4-byte atomics"
++	depends on BR2_USE_MMU
++	depends on !BR2_TOOLCHAIN_HAS_THREADS || \
++		!BR2_TOOLCHAIN_HAS_SYNC_4 || \
++		!BR2_TOOLCHAIN_GCC_AT_LEAST_4_4
+diff --git a/package/libmdbx/libmdbx.hash b/package/libmdbx/libmdbx.hash
+new file mode 100644
+index 0000000000..ae5266716b
+--- /dev/null
++++ b/package/libmdbx/libmdbx.hash
+@@ -0,0 +1,6 @@
++# Hashes from: https://libmdbx.dqdkfa.ru/release/SHA256SUMS
++sha256  57db987de6f7ccc66a66ae28a7bda9f9fbb48ac5fb9279bcca92fd5de13075d1  libmdbx-amalgamated-0.13.6.tar.xz
++
++# Locally calculated
++sha256  0d542e0c8804e39aa7f37eb00da5a762149dc682d7829451287e11b938e94594  LICENSE
++sha256  651f71b46c6bb0046d2122df7f9def9cb24f4dc28c5b11cef059f66565cda30f  NOTICE
+diff --git a/package/libmdbx/libmdbx.mk b/package/libmdbx/libmdbx.mk
+new file mode 100644
+index 0000000000..571757262e
+--- /dev/null
++++ b/package/libmdbx/libmdbx.mk
+@@ -0,0 +1,42 @@
++################################################################################
++#
++# libmdbx
++#
++################################################################################
++
++LIBMDBX_VERSION = 0.13.6
++LIBMDBX_SOURCE = libmdbx-amalgamated-$(LIBMDBX_VERSION).tar.xz
++LIBMDBX_SITE = https://libmdbx.dqdkfa.ru/release
++LIBMDBX_SUPPORTS_IN_SOURCE_BUILD = NO
++LIBMDBX_LICENSE = Apache-2.0
++LIBMDBX_LICENSE_FILES = LICENSE NOTICE
++LIBMDBX_REDISTRIBUTE = YES
++LIBMDBX_STRIP_COMPONENTS = 0
++LIBMDBX_INSTALL_STAGING = YES
++
++# Set CMAKE_BUILD_TYPE to Release to remove -Werror and avoid a build failure
++# with glibc < 2.12
++LIBMDBX_CONF_OPTS = \
++	-DCMAKE_BUILD_TYPE=Release \
++	-DMDBX_INSTALL_MANPAGES=OFF \
++	-DBUILD_FOR_NATIVE_CPU=OFF \
++	-DMDBX_BUILD_CXX=$(if $(BR2_PACKAGE_LIBMDBX_CXX),ON,OFF) \
++	-DMDBX_BUILD_TOOLS=$(if $(BR2_PACKAGE_LIBMDBX_TOOLS),ON,OFF)
++
++ifeq ($(BR2_STATIC_LIBS)$(BR2_SHARED_STATIC_LIBS),y)
++LIBMDBX_CONF_OPTS += -DMDBX_INSTALL_STATIC=ON
++else
++LIBMDBX_CONF_OPTS += -DMDBX_INSTALL_STATIC=OFF
++endif
++
++ifeq ($(BR2_SHARED_LIBS)$(BR2_SHARED_STATIC_LIBS),y)
++LIBMDBX_CONF_OPTS += \
++	-DMDBX_BUILD_SHARED_LIBRARY=ON \
++	-DMDBX_LINK_TOOLS_NONSTATIC=ON
++else
++LIBMDBX_CONF_OPTS += \
++	-DMDBX_BUILD_SHARED_LIBRARY=OFF \
++	-DMDBX_LINK_TOOLS_NONSTATIC=OFF
++endif
++
++$(eval $(cmake-package))
+-- 
+2.49.0
+

packages/rpm/CMakeLists.txt

@@ -1,184 +0,0 @@
-cmake_minimum_required(VERSION 2.8.7)
-set(TARGET mdbx)
-project(${TARGET})
-
-set(MDBX_VERSION_MAJOR 0)
-set(MDBX_VERSION_MINOR 2)
-set(MDBX_VERSION_RELEASE 0)
-set(MDBX_VERSION_REVISION 0)
-
-set(MDBX_VERSION_STRING ${MDBX_VERSION_MAJOR}.${MDBX_VERSION_MINOR}.${MDBX_VERSION_RELEASE})
-
-enable_language(C)
-enable_language(CXX)
-
-set(CMAKE_CXX_STANDARD 11)
-set(CMAKE_CXX_STANDARD_REQUIRED on)
-
-add_definitions(-DNDEBUG=1 -DMDBX_DEBUG=0 -DLIBMDBX_EXPORTS=1 -D_GNU_SOURCE=1)
-
-find_package(Threads REQUIRED)
-
-get_directory_property(hasParent PARENT_DIRECTORY)
-if(hasParent)
-  set(STANDALONE_BUILD 0)
-else()
-  set(STANDALONE_BUILD 1)
-  enable_testing()
-
-  if (CMAKE_C_COMPILER_ID MATCHES GNU)
-    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -O2")
-    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -g3")
-    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Wall")
-    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Werror")
-    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -Wextra")
-    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -ffunction-sections")
-    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fPIC")
-    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -fvisibility=hidden")
-    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -std=gnu11")
-    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} -pthread")
-  endif()
-
-  if (CMAKE_CXX_COMPILER_ID MATCHES GNU)
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -W")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wall")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Werror")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wpointer-arith")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-sign-compare")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wformat-security")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Woverloaded-virtual")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wwrite-strings")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fmax-errors=20")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fvisibility=hidden")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-unused-parameter -Wunused-function -Wunused-variable -Wunused-value -Wmissing-declarations")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-missing-field-initializers")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wcast-qual")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -ggdb")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -fno-omit-frame-pointer")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-strict-aliasing")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -finline-functions-called-once")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -Wno-packed-bitfield-compat")
-
-    set(CMAKE_CXX_FLAGS_DEBUG "-O0 -g3")
-    set(CMAKE_CXX_FLAGS_RELWITHDEBINFO "-O3 -g3")
-  endif()
-
-  if (COVERAGE)
-    if (NOT "${CMAKE_BUILD_TYPE}" STREQUAL "Debug")
-      message(FATAL_ERROR "Coverage requires -DCMAKE_BUILD_TYPE=Debug Current value=${CMAKE_BUILD_TYPE}")
-    endif()
-
-    message(STATUS "Setting coverage compiler flags")
-    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -g -ggdb3 -O0 --coverage -fprofile-arcs -ftest-coverage")
-    set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS}     -g -ggdb3 -O0 --coverage -fprofile-arcs -ftest-coverage")
-    add_definitions(-DCOVERAGE_TEST)
-  endif()
-
-  if (NOT TRAVIS)
-    set(CMAKE_CXX_FLAGS_DEBUG "${CMAKE_CXX_FLAGS_DEBUG} -fsanitize=address -fsanitize=leak -fstack-protector-strong -static-libasan")
-  endif()
-endif()
-
-set(${TARGET}_SRC
-  mdbx.h
-  src/bits.h
-  src/defs.h
-  src/lck-posix.c
-  src/mdbx.c
-  src/osal.c
-  src/osal.h
-  src/version.c
-  )
-
-add_library(${TARGET}_STATIC STATIC
-  ${${TARGET}_SRC}
-  )
-
-add_library(${TARGET} ALIAS ${TARGET}_STATIC)
-
-add_library(${TARGET}_SHARED SHARED
-  ${${TARGET}_SRC}
-  )
-
-set_target_properties(${TARGET}_SHARED PROPERTIES
-  VERSION ${MDBX_VERSION_STRING}
-  SOVERSION ${MDBX_VERSION_MAJOR}.${MDBX_VERSION_MINOR}
-  OUTPUT_NAME ${TARGET}
-  CLEAN_DIRECT_OUTPUT 1
-  )
-
-set_target_properties(${TARGET}_STATIC PROPERTIES
-  VERSION ${MDBX_VERSION_STRING}
-  SOVERSION ${MDBX_VERSION_MAJOR}.${MDBX_VERSION_MINOR}
-  OUTPUT_NAME ${TARGET}
-  CLEAN_DIRECT_OUTPUT 1
-  )
-
-target_include_directories(${TARGET}_STATIC PUBLIC
-  ${CMAKE_CURRENT_SOURCE_DIR})
-target_include_directories(${TARGET}_SHARED PUBLIC
-  ${CMAKE_CURRENT_SOURCE_DIR})
-
-target_link_libraries(${TARGET}_STATIC ${CMAKE_THREAD_LIBS_INIT})
-target_link_libraries(${TARGET}_SHARED ${CMAKE_THREAD_LIBS_INIT})
-if(UNIX AND NOT APPLE)
-  target_link_libraries(${TARGET}_STATIC rt)
-  target_link_libraries(${TARGET}_SHARED rt)
-endif()
-
-install(TARGETS ${TARGET}_STATIC DESTINATION ${CMAKE_INSTALL_PREFIX}/lib64 COMPONENT mdbx)
-install(TARGETS ${TARGET}_SHARED DESTINATION ${CMAKE_INSTALL_PREFIX}/lib64 COMPONENT mdbx)
-install(FILES mdbx.h DESTINATION ${CMAKE_INSTALL_PREFIX}/include COMPONENT mdbx-devel)
-
-add_subdirectory(src/tools)
-add_subdirectory(test)
-add_subdirectory(test/pcrf)
-add_subdirectory(tutorial)
-
-##############################################################################
-
-set(CPACK_GENERATOR "RPM")
-set(CPACK_RPM_COMPONENT_INSTALL ON)
-
-# Version
-if (NOT "$ENV{BUILD_NUMBER}" STREQUAL "")
-  set(CPACK_PACKAGE_RELEASE $ENV{BUILD_NUMBER})
-else()
-  if (NOT "$ENV{CI_PIPELINE_ID}" STREQUAL "")
-    set(CPACK_PACKAGE_RELEASE $ENV{CI_PIPELINE_ID})
-  else()
-    set(CPACK_PACKAGE_RELEASE 1)
-  endif()
-endif()
-set(CPACK_RPM_PACKAGE_RELEASE ${CPACK_PACKAGE_RELEASE})
-
-set(CPACK_PACKAGE_VERSION ${MDBX_VERSION_STRING})
-set(CPACK_PACKAGE_VERSION_FULL ${CPACK_PACKAGE_VERSION}-${CPACK_PACKAGE_RELEASE})
-
-set(CPACK_RPM_mdbx-devel_PACKAGE_REQUIRES "mdbx = ${CPACK_PACKAGE_VERSION}")
-
-set(CPACK_RPM_SPEC_INSTALL_POST "/bin/true")
-set(CPACK_RPM_mdbx_PACKAGE_NAME mdbx)
-set(CPACK_RPM_mdbx-devel_PACKAGE_NAME mdbx-devel)
-set(CPACK_PACKAGE_DESCRIPTION_SUMMARY "The revised and extended descendant of Symas LMDB")
-
-set(CPACK_PACKAGE_VENDOR "???")
-set(CPACK_PACKAGE_CONTACT "Vladimir Romanov")
-set(CPACK_PACKAGE_RELOCATABLE false)
-set(CPACK_RPM_PACKAGE_ARCHITECTURE "x86_64")
-set(CPACK_RPM_PACKAGE_REQUIRES "")
-set(CPACK_RPM_PACKAGE_GROUP "Applications/Database")
-
-set(CPACK_RPM_mdbx_FILE_NAME "${CPACK_RPM_mdbx_PACKAGE_NAME}-${CPACK_PACKAGE_VERSION_FULL}.${CPACK_RPM_PACKAGE_ARCHITECTURE}.rpm")
-set(CPACK_RPM_mdbx-devel_FILE_NAME "${CPACK_RPM_mdbx-devel_PACKAGE_NAME}-${CPACK_PACKAGE_VERSION_FULL}.${CPACK_RPM_PACKAGE_ARCHITECTURE}.rpm")
-
-set(CPACK_RPM_EXCLUDE_FROM_AUTO_FILELIST_ADDITION
-  /usr/local
-  /usr/local/bin
-  /usr/local/lib64
-  /usr/local/include
-  /usr/local/man
-  /usr/local/man/man1
-  )
-
-include(CPack)

packages/rpm/build.sh

@@ -1,18 +0,0 @@
-#!/bin/bash
-set -e
-CONFIG=$1
-
-if [[ -z "${CONFIG}" ]]; then
-    CONFIG=Debug
-fi
-if [[ -r /opt/rh/devtoolset-6/enable ]]; then
-    source /opt/rh/devtoolset-6/enable
-fi
-#rm -f -r build || true
-mkdir -p cmake-build-${CONFIG}
-pushd  cmake-build-${CONFIG} &> /dev/null
-if [[ ! -r Makefile ]]; then
-    cmake .. -DCMAKE_BUILD_TYPE=${CONFIG}
-fi
-make -j8 || exit 1
-popd &> /dev/null

packages/rpm/package.sh

@@ -1,25 +0,0 @@
-#!/bin/bash
-set -e
-
-CONFIG=$1
-
-if [[ -z "${CONFIG}" ]]; then
-    CONFIG=Debug
-fi
-
-DIRNAME=`dirname ${BASH_SOURCE[0]}`
-DIRNAME=`readlink --canonicalize ${DIRNAME}`
-
-if [[ -r /opt/rh/devtoolset-6/enable ]]; then
-    source /opt/rh/devtoolset-6/enable
-fi
-
-mkdir -p cmake-build-${CONFIG}
-pushd  cmake-build-${CONFIG} &> /dev/null
-if [[ ! -r Makefile ]]; then
-    cmake .. -DCMAKE_BUILD_TYPE=${CONFIG}
-fi
-rm -f *.rpm
-make -j8 package || exit 1
-rm -f *-Unspecified.rpm
-popd &> /dev/null

src/alloy.c

@@ -0,0 +1,58 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#define xMDBX_ALLOY 1  /* alloyed build */
+#include "internals.h" /* must be included first */
+
+#include "api-cold.c"
+#include "api-copy.c"
+#include "api-cursor.c"
+#include "api-dbi.c"
+#include "api-env.c"
+#include "api-extra.c"
+#include "api-key-transform.c"
+#include "api-misc.c"
+#include "api-opts.c"
+#include "api-range-estimate.c"
+#include "api-txn-data.c"
+#include "api-txn.c"
+#include "audit.c"
+#include "chk.c"
+#include "cogs.c"
+#include "coherency.c"
+#include "cursor.c"
+#include "dbi.c"
+#include "dpl.c"
+#include "dxb.c"
+#include "env.c"
+#include "gc-get.c"
+#include "gc-put.c"
+#include "global.c"
+#include "lck-posix.c"
+#include "lck-windows.c"
+#include "lck.c"
+#include "logging_and_debug.c"
+#include "meta.c"
+#include "mvcc-readers.c"
+#include "node.c"
+#include "osal.c"
+#include "page-get.c"
+#include "page-iov.c"
+#include "page-ops.c"
+#include "pnl.c"
+#include "refund.c"
+#include "rkl.c"
+#include "spill.c"
+#include "table.c"
+#include "tls.c"
+#include "tree-ops.c"
+#include "tree-search.c"
+#include "txl.c"
+#include "txn-basal.c"
+#include "txn-nested.c"
+#include "txn-ro.c"
+#include "txn.c"
+#include "utils.c"
+#include "version.c"
+#include "walk.c"
+#include "windows-import.c"

src/api-cold.c

@@ -0,0 +1,543 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+__cold size_t mdbx_default_pagesize(void) {
+  size_t pagesize = globals.sys_pagesize;
+  ENSURE(nullptr, is_powerof2(pagesize));
+  pagesize = (pagesize >= MDBX_MIN_PAGESIZE) ? pagesize : MDBX_MIN_PAGESIZE;
+  pagesize = (pagesize <= MDBX_MAX_PAGESIZE) ? pagesize : MDBX_MAX_PAGESIZE;
+  return pagesize;
+}
+
+__cold intptr_t mdbx_limits_dbsize_min(intptr_t pagesize) {
+  if (pagesize < 1)
+    pagesize = (intptr_t)mdbx_default_pagesize();
+  else if (unlikely(pagesize < (intptr_t)MDBX_MIN_PAGESIZE || pagesize > (intptr_t)MDBX_MAX_PAGESIZE ||
+                    !is_powerof2((size_t)pagesize)))
+    return -1;
+
+  return MIN_PAGENO * pagesize;
+}
+
+__cold intptr_t mdbx_limits_dbsize_max(intptr_t pagesize) {
+  if (pagesize < 1)
+    pagesize = (intptr_t)mdbx_default_pagesize();
+  else if (unlikely(pagesize < (intptr_t)MDBX_MIN_PAGESIZE || pagesize > (intptr_t)MDBX_MAX_PAGESIZE ||
+                    !is_powerof2((size_t)pagesize)))
+    return -1;
+
+  STATIC_ASSERT(MAX_MAPSIZE < INTPTR_MAX);
+  const uint64_t limit = (1 + (uint64_t)MAX_PAGENO) * pagesize;
+  return (limit < MAX_MAPSIZE) ? (intptr_t)limit : (intptr_t)MAX_MAPSIZE;
+}
+
+__cold intptr_t mdbx_limits_txnsize_max(intptr_t pagesize) {
+  if (pagesize < 1)
+    pagesize = (intptr_t)mdbx_default_pagesize();
+  else if (unlikely(pagesize < (intptr_t)MDBX_MIN_PAGESIZE || pagesize > (intptr_t)MDBX_MAX_PAGESIZE ||
+                    !is_powerof2((size_t)pagesize)))
+    return -1;
+
+  STATIC_ASSERT(MAX_MAPSIZE < INTPTR_MAX);
+  const uint64_t pgl_limit = pagesize * (uint64_t)(PAGELIST_LIMIT / MDBX_GOLD_RATIO_DBL);
+  const uint64_t map_limit = (uint64_t)(MAX_MAPSIZE / MDBX_GOLD_RATIO_DBL);
+  return (pgl_limit < map_limit) ? (intptr_t)pgl_limit : (intptr_t)map_limit;
+}
+
+__cold intptr_t mdbx_limits_keysize_max(intptr_t pagesize, MDBX_db_flags_t flags) {
+  if (pagesize < 1)
+    pagesize = (intptr_t)mdbx_default_pagesize();
+  if (unlikely(pagesize < (intptr_t)MDBX_MIN_PAGESIZE || pagesize > (intptr_t)MDBX_MAX_PAGESIZE ||
+               !is_powerof2((size_t)pagesize)))
+    return -1;
+
+  return keysize_max(pagesize, flags);
+}
+
+__cold int mdbx_env_get_maxkeysize_ex(const MDBX_env *env, MDBX_db_flags_t flags) {
+  if (unlikely(!env || env->signature.weak != env_signature))
+    return -1;
+
+  return (int)mdbx_limits_keysize_max((intptr_t)env->ps, flags);
+}
+
+__cold int mdbx_env_get_maxkeysize(const MDBX_env *env) { return mdbx_env_get_maxkeysize_ex(env, MDBX_DUPSORT); }
+
+__cold intptr_t mdbx_limits_keysize_min(MDBX_db_flags_t flags) { return keysize_min(flags); }
+
+__cold intptr_t mdbx_limits_valsize_max(intptr_t pagesize, MDBX_db_flags_t flags) {
+  if (pagesize < 1)
+    pagesize = (intptr_t)mdbx_default_pagesize();
+  if (unlikely(pagesize < (intptr_t)MDBX_MIN_PAGESIZE || pagesize > (intptr_t)MDBX_MAX_PAGESIZE ||
+               !is_powerof2((size_t)pagesize)))
+    return -1;
+
+  return valsize_max(pagesize, flags);
+}
+
+__cold int mdbx_env_get_maxvalsize_ex(const MDBX_env *env, MDBX_db_flags_t flags) {
+  if (unlikely(!env || env->signature.weak != env_signature))
+    return -1;
+
+  return (int)mdbx_limits_valsize_max((intptr_t)env->ps, flags);
+}
+
+__cold intptr_t mdbx_limits_valsize_min(MDBX_db_flags_t flags) { return valsize_min(flags); }
+
+__cold intptr_t mdbx_limits_pairsize4page_max(intptr_t pagesize, MDBX_db_flags_t flags) {
+  if (pagesize < 1)
+    pagesize = (intptr_t)mdbx_default_pagesize();
+  if (unlikely(pagesize < (intptr_t)MDBX_MIN_PAGESIZE || pagesize > (intptr_t)MDBX_MAX_PAGESIZE ||
+               !is_powerof2((size_t)pagesize)))
+    return -1;
+
+  if (flags & (MDBX_DUPSORT | MDBX_DUPFIXED | MDBX_INTEGERDUP | MDBX_REVERSEDUP))
+    return BRANCH_NODE_MAX(pagesize) - NODESIZE;
+
+  return LEAF_NODE_MAX(pagesize) - NODESIZE;
+}
+
+__cold int mdbx_env_get_pairsize4page_max(const MDBX_env *env, MDBX_db_flags_t flags) {
+  if (unlikely(!env || env->signature.weak != env_signature))
+    return -1;
+
+  return (int)mdbx_limits_pairsize4page_max((intptr_t)env->ps, flags);
+}
+
+__cold intptr_t mdbx_limits_valsize4page_max(intptr_t pagesize, MDBX_db_flags_t flags) {
+  if (pagesize < 1)
+    pagesize = (intptr_t)mdbx_default_pagesize();
+  if (unlikely(pagesize < (intptr_t)MDBX_MIN_PAGESIZE || pagesize > (intptr_t)MDBX_MAX_PAGESIZE ||
+               !is_powerof2((size_t)pagesize)))
+    return -1;
+
+  if (flags & (MDBX_DUPSORT | MDBX_DUPFIXED | MDBX_INTEGERDUP | MDBX_REVERSEDUP))
+    return valsize_max(pagesize, flags);
+
+  return PAGESPACE(pagesize);
+}
+
+__cold int mdbx_env_get_valsize4page_max(const MDBX_env *env, MDBX_db_flags_t flags) {
+  if (unlikely(!env || env->signature.weak != env_signature))
+    return -1;
+
+  return (int)mdbx_limits_valsize4page_max((intptr_t)env->ps, flags);
+}
+
+/*----------------------------------------------------------------------------*/
+
+static size_t estimate_rss(size_t database_bytes) {
+  return database_bytes + database_bytes / 64 + (512 + MDBX_WORDBITS * 16) * MEGABYTE;
+}
+
+__cold int mdbx_env_warmup(const MDBX_env *env, const MDBX_txn *txn, MDBX_warmup_flags_t flags,
+                           unsigned timeout_seconds_16dot16) {
+  if (unlikely(env == nullptr && txn == nullptr))
+    return LOG_IFERR(MDBX_EINVAL);
+  if (unlikely(flags > (MDBX_warmup_force | MDBX_warmup_oomsafe | MDBX_warmup_lock | MDBX_warmup_touchlimit |
+                        MDBX_warmup_release)))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  if (txn) {
+    int err = check_txn(txn, MDBX_TXN_FINISHED | MDBX_TXN_ERROR);
+    if (unlikely(err != MDBX_SUCCESS))
+      return LOG_IFERR(err);
+  }
+  if (env) {
+    int err = check_env(env, false);
+    if (unlikely(err != MDBX_SUCCESS))
+      return LOG_IFERR(err);
+    if (txn && unlikely(txn->env != env))
+      return LOG_IFERR(MDBX_EINVAL);
+  } else {
+    env = txn->env;
+  }
+
+  const uint64_t timeout_monotime = (timeout_seconds_16dot16 && (flags & MDBX_warmup_force))
+                                        ? osal_monotime() + osal_16dot16_to_monotime(timeout_seconds_16dot16)
+                                        : 0;
+
+  if (flags & MDBX_warmup_release)
+    munlock_all(env);
+
+  pgno_t used_pgno;
+  if (txn) {
+    used_pgno = txn->geo.first_unallocated;
+  } else {
+    const troika_t troika = meta_tap(env);
+    used_pgno = meta_recent(env, &troika).ptr_v->geometry.first_unallocated;
+  }
+  const size_t used_range = pgno_align2os_bytes(env, used_pgno);
+  const pgno_t mlock_pgno = bytes2pgno(env, used_range);
+
+  int rc = MDBX_SUCCESS;
+  if (flags & MDBX_warmup_touchlimit) {
+    const size_t estimated_rss = estimate_rss(used_range);
+#if defined(_WIN32) || defined(_WIN64)
+    SIZE_T current_ws_lower, current_ws_upper;
+    if (GetProcessWorkingSetSize(GetCurrentProcess(), &current_ws_lower, &current_ws_upper) &&
+        current_ws_lower < estimated_rss) {
+      const SIZE_T ws_lower = estimated_rss;
+      const SIZE_T ws_upper =
+          (MDBX_WORDBITS == 32 && ws_lower > MEGABYTE * 2048) ? ws_lower : ws_lower + MDBX_WORDBITS * MEGABYTE * 32;
+      if (!SetProcessWorkingSetSize(GetCurrentProcess(), ws_lower, ws_upper)) {
+        rc = (int)GetLastError();
+        WARNING("SetProcessWorkingSetSize(%zu, %zu) error %d", ws_lower, ws_upper, rc);
+      }
+    }
+#endif /* Windows */
+#ifdef RLIMIT_RSS
+    struct rlimit rss;
+    if (getrlimit(RLIMIT_RSS, &rss) == 0 && rss.rlim_cur < estimated_rss) {
+      rss.rlim_cur = estimated_rss;
+      if (rss.rlim_max < estimated_rss)
+        rss.rlim_max = estimated_rss;
+      if (setrlimit(RLIMIT_RSS, &rss)) {
+        rc = errno;
+        WARNING("setrlimit(%s, {%zu, %zu}) error %d", "RLIMIT_RSS", (size_t)rss.rlim_cur, (size_t)rss.rlim_max, rc);
+      }
+    }
+#endif /* RLIMIT_RSS */
+#ifdef RLIMIT_MEMLOCK
+    if (flags & MDBX_warmup_lock) {
+      struct rlimit memlock;
+      if (getrlimit(RLIMIT_MEMLOCK, &memlock) == 0 && memlock.rlim_cur < estimated_rss) {
+        memlock.rlim_cur = estimated_rss;
+        if (memlock.rlim_max < estimated_rss)
+          memlock.rlim_max = estimated_rss;
+        if (setrlimit(RLIMIT_MEMLOCK, &memlock)) {
+          rc = errno;
+          WARNING("setrlimit(%s, {%zu, %zu}) error %d", "RLIMIT_MEMLOCK", (size_t)memlock.rlim_cur,
+                  (size_t)memlock.rlim_max, rc);
+        }
+      }
+    }
+#endif /* RLIMIT_MEMLOCK */
+    (void)estimated_rss;
+  }
+
+#if defined(MLOCK_ONFAULT) &&                                                                                          \
+    ((defined(_GNU_SOURCE) && __GLIBC_PREREQ(2, 27)) || (defined(__ANDROID_API__) && __ANDROID_API__ >= 30)) &&        \
+    (defined(__linux__) || defined(__gnu_linux__))
+  if ((flags & MDBX_warmup_lock) != 0 && globals.linux_kernel_version >= 0x04040000 &&
+      atomic_load32(&env->mlocked_pgno, mo_AcquireRelease) < mlock_pgno) {
+    if (mlock2(env->dxb_mmap.base, used_range, MLOCK_ONFAULT)) {
+      rc = errno;
+      WARNING("mlock2(%zu, %s) error %d", used_range, "MLOCK_ONFAULT", rc);
+    } else {
+      update_mlcnt(env, mlock_pgno, true);
+      rc = MDBX_SUCCESS;
+    }
+    if (rc != EINVAL)
+      flags -= MDBX_warmup_lock;
+  }
+#endif /* MLOCK_ONFAULT */
+
+  int err = MDBX_ENOSYS;
+  err = dxb_set_readahead(env, used_pgno, true, true);
+  if (err != MDBX_SUCCESS && rc == MDBX_SUCCESS)
+    rc = err;
+
+  if ((flags & MDBX_warmup_force) != 0 && (rc == MDBX_SUCCESS || rc == MDBX_ENOSYS)) {
+    const volatile uint8_t *ptr = env->dxb_mmap.base;
+    size_t offset = 0, unused = 42;
+#if !(defined(_WIN32) || defined(_WIN64))
+    if (flags & MDBX_warmup_oomsafe) {
+      const int null_fd = open("/dev/null", O_WRONLY);
+      if (unlikely(null_fd < 0))
+        rc = errno;
+      else {
+        struct iovec iov[MDBX_AUXILARY_IOV_MAX];
+        for (;;) {
+          unsigned i;
+          for (i = 0; i < MDBX_AUXILARY_IOV_MAX && offset < used_range; ++i) {
+            iov[i].iov_base = (void *)(ptr + offset);
+            iov[i].iov_len = 1;
+            offset += globals.sys_pagesize;
+          }
+          if (unlikely(writev(null_fd, iov, i) < 0)) {
+            rc = errno;
+            if (rc == EFAULT)
+              rc = ENOMEM;
+            break;
+          }
+          if (offset >= used_range) {
+            rc = MDBX_SUCCESS;
+            break;
+          }
+          if (timeout_seconds_16dot16 && osal_monotime() > timeout_monotime) {
+            rc = MDBX_RESULT_TRUE;
+            break;
+          }
+        }
+        close(null_fd);
+      }
+    } else
+#endif /* Windows */
+      for (;;) {
+        unused += ptr[offset];
+        offset += globals.sys_pagesize;
+        if (offset >= used_range) {
+          rc = MDBX_SUCCESS;
+          break;
+        }
+        if (timeout_seconds_16dot16 && osal_monotime() > timeout_monotime) {
+          rc = MDBX_RESULT_TRUE;
+          break;
+        }
+      }
+    (void)unused;
+  }
+
+  if ((flags & MDBX_warmup_lock) != 0 && (rc == MDBX_SUCCESS || rc == MDBX_ENOSYS) &&
+      atomic_load32(&env->mlocked_pgno, mo_AcquireRelease) < mlock_pgno) {
+#if defined(_WIN32) || defined(_WIN64)
+    if (VirtualLock(env->dxb_mmap.base, used_range)) {
+      update_mlcnt(env, mlock_pgno, true);
+      rc = MDBX_SUCCESS;
+    } else {
+      rc = (int)GetLastError();
+      WARNING("%s(%zu) error %d", "VirtualLock", used_range, rc);
+    }
+#elif defined(_POSIX_MEMLOCK_RANGE)
+    if (mlock(env->dxb_mmap.base, used_range) == 0) {
+      update_mlcnt(env, mlock_pgno, true);
+      rc = MDBX_SUCCESS;
+    } else {
+      rc = errno;
+      WARNING("%s(%zu) error %d", "mlock", used_range, rc);
+    }
+#else
+    rc = MDBX_ENOSYS;
+#endif
+  }
+
+  return LOG_IFERR(rc);
+}
+
+/*----------------------------------------------------------------------------*/
+
+__cold int mdbx_env_get_fd(const MDBX_env *env, mdbx_filehandle_t *arg) {
+  int rc = check_env(env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(!arg))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  *arg = env->lazy_fd;
+  return MDBX_SUCCESS;
+}
+
+__cold int mdbx_env_set_flags(MDBX_env *env, MDBX_env_flags_t flags, bool onoff) {
+  int rc = check_env(env, false);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(flags & ((env->flags & ENV_ACTIVE) ? ~ENV_CHANGEABLE_FLAGS : ~ENV_USABLE_FLAGS)))
+    return LOG_IFERR(MDBX_EPERM);
+
+  if (unlikely(env->flags & MDBX_RDONLY))
+    return LOG_IFERR(MDBX_EACCESS);
+
+  const bool lock_needed = (env->flags & ENV_ACTIVE) && !env_owned_wrtxn(env);
+  bool should_unlock = false;
+  if (lock_needed) {
+    rc = lck_txn_lock(env, false);
+    if (unlikely(rc != MDBX_SUCCESS))
+      return LOG_IFERR(rc);
+    should_unlock = true;
+  }
+
+  if (onoff)
+    env->flags = combine_durability_flags(env->flags, flags);
+  else
+    env->flags &= ~flags;
+
+  if (should_unlock)
+    lck_txn_unlock(env);
+  return MDBX_SUCCESS;
+}
+
+__cold int mdbx_env_get_flags(const MDBX_env *env, unsigned *flags) {
+  int rc = check_env(env, false);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(!flags))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  *flags = env->flags & ENV_USABLE_FLAGS;
+  return MDBX_SUCCESS;
+}
+
+__cold int mdbx_env_set_userctx(MDBX_env *env, void *ctx) {
+  int rc = check_env(env, false);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  env->userctx = ctx;
+  return MDBX_SUCCESS;
+}
+
+__cold void *mdbx_env_get_userctx(const MDBX_env *env) { return env ? env->userctx : nullptr; }
+
+__cold int mdbx_env_set_assert(MDBX_env *env, MDBX_assert_func *func) {
+  int rc = check_env(env, false);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+#if MDBX_DEBUG
+  env->assert_func = func;
+  return MDBX_SUCCESS;
+#else
+  (void)func;
+  return LOG_IFERR(MDBX_ENOSYS);
+#endif
+}
+
+__cold int mdbx_env_set_hsr(MDBX_env *env, MDBX_hsr_func *hsr) {
+  int rc = check_env(env, false);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  env->hsr_callback = hsr;
+  return MDBX_SUCCESS;
+}
+
+__cold MDBX_hsr_func *mdbx_env_get_hsr(const MDBX_env *env) {
+  return likely(env && env->signature.weak == env_signature) ? env->hsr_callback : nullptr;
+}
+
+#if defined(_WIN32) || defined(_WIN64)
+__cold int mdbx_env_get_pathW(const MDBX_env *env, const wchar_t **arg) {
+  int rc = check_env(env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(!arg))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  *arg = env->pathname.specified;
+  return MDBX_SUCCESS;
+}
+#endif /* Windows */
+
+__cold int mdbx_env_get_path(const MDBX_env *env, const char **arg) {
+  int rc = check_env(env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(!arg))
+    return LOG_IFERR(MDBX_EINVAL);
+
+#if defined(_WIN32) || defined(_WIN64)
+  if (!env->pathname_char) {
+    *arg = nullptr;
+    DWORD flags = /* WC_ERR_INVALID_CHARS */ 0x80;
+    size_t mb_len =
+        WideCharToMultiByte(CP_THREAD_ACP, flags, env->pathname.specified, -1, nullptr, 0, nullptr, nullptr);
+    rc = mb_len ? MDBX_SUCCESS : (int)GetLastError();
+    if (rc == ERROR_INVALID_FLAGS) {
+      mb_len = WideCharToMultiByte(CP_THREAD_ACP, flags = 0, env->pathname.specified, -1, nullptr, 0, nullptr, nullptr);
+      rc = mb_len ? MDBX_SUCCESS : (int)GetLastError();
+    }
+    if (unlikely(rc != MDBX_SUCCESS))
+      return LOG_IFERR(rc);
+
+    char *const mb_pathname = osal_malloc(mb_len);
+    if (!mb_pathname)
+      return LOG_IFERR(MDBX_ENOMEM);
+    if (mb_len != (size_t)WideCharToMultiByte(CP_THREAD_ACP, flags, env->pathname.specified, -1, mb_pathname,
+                                              (int)mb_len, nullptr, nullptr)) {
+      rc = (int)GetLastError();
+      osal_free(mb_pathname);
+      return LOG_IFERR(rc);
+    }
+    if (env->pathname_char ||
+        InterlockedCompareExchangePointer((PVOID volatile *)&env->pathname_char, mb_pathname, nullptr))
+      osal_free(mb_pathname);
+  }
+  *arg = env->pathname_char;
+#else
+  *arg = env->pathname.specified;
+#endif /* Windows */
+  return MDBX_SUCCESS;
+}
+
+/*------------------------------------------------------------------------------
+ * Legacy API */
+
+#ifndef LIBMDBX_NO_EXPORTS_LEGACY_API
+
+LIBMDBX_API int mdbx_txn_begin(MDBX_env *env, MDBX_txn *parent, MDBX_txn_flags_t flags, MDBX_txn **ret) {
+  return __inline_mdbx_txn_begin(env, parent, flags, ret);
+}
+
+LIBMDBX_API int mdbx_txn_commit(MDBX_txn *txn) { return __inline_mdbx_txn_commit(txn); }
+
+LIBMDBX_API __cold int mdbx_env_stat(const MDBX_env *env, MDBX_stat *stat, size_t bytes) {
+  return __inline_mdbx_env_stat(env, stat, bytes);
+}
+
+LIBMDBX_API __cold int mdbx_env_info(const MDBX_env *env, MDBX_envinfo *info, size_t bytes) {
+  return __inline_mdbx_env_info(env, info, bytes);
+}
+
+LIBMDBX_API int mdbx_dbi_flags(const MDBX_txn *txn, MDBX_dbi dbi, unsigned *flags) {
+  return __inline_mdbx_dbi_flags(txn, dbi, flags);
+}
+
+LIBMDBX_API __cold int mdbx_env_sync(MDBX_env *env) { return __inline_mdbx_env_sync(env); }
+
+LIBMDBX_API __cold int mdbx_env_sync_poll(MDBX_env *env) { return __inline_mdbx_env_sync_poll(env); }
+
+LIBMDBX_API __cold int mdbx_env_close(MDBX_env *env) { return __inline_mdbx_env_close(env); }
+
+LIBMDBX_API __cold int mdbx_env_set_mapsize(MDBX_env *env, size_t size) {
+  return __inline_mdbx_env_set_mapsize(env, size);
+}
+
+LIBMDBX_API __cold int mdbx_env_set_maxdbs(MDBX_env *env, MDBX_dbi dbs) {
+  return __inline_mdbx_env_set_maxdbs(env, dbs);
+}
+
+LIBMDBX_API __cold int mdbx_env_get_maxdbs(const MDBX_env *env, MDBX_dbi *dbs) {
+  return __inline_mdbx_env_get_maxdbs(env, dbs);
+}
+
+LIBMDBX_API __cold int mdbx_env_set_maxreaders(MDBX_env *env, unsigned readers) {
+  return __inline_mdbx_env_set_maxreaders(env, readers);
+}
+
+LIBMDBX_API __cold int mdbx_env_get_maxreaders(const MDBX_env *env, unsigned *readers) {
+  return __inline_mdbx_env_get_maxreaders(env, readers);
+}
+
+LIBMDBX_API __cold int mdbx_env_set_syncbytes(MDBX_env *env, size_t threshold) {
+  return __inline_mdbx_env_set_syncbytes(env, threshold);
+}
+
+LIBMDBX_API __cold int mdbx_env_get_syncbytes(const MDBX_env *env, size_t *threshold) {
+  return __inline_mdbx_env_get_syncbytes(env, threshold);
+}
+
+LIBMDBX_API __cold int mdbx_env_set_syncperiod(MDBX_env *env, unsigned seconds_16dot16) {
+  return __inline_mdbx_env_set_syncperiod(env, seconds_16dot16);
+}
+
+LIBMDBX_API __cold int mdbx_env_get_syncperiod(const MDBX_env *env, unsigned *seconds_16dot16) {
+  return __inline_mdbx_env_get_syncperiod(env, seconds_16dot16);
+}
+
+LIBMDBX_API __cold uint64_t mdbx_key_from_int64(const int64_t i64) { return __inline_mdbx_key_from_int64(i64); }
+
+LIBMDBX_API __cold uint32_t mdbx_key_from_int32(const int32_t i32) { return __inline_mdbx_key_from_int32(i32); }
+
+LIBMDBX_API __cold intptr_t mdbx_limits_pgsize_min(void) { return __inline_mdbx_limits_pgsize_min(); }
+
+LIBMDBX_API __cold intptr_t mdbx_limits_pgsize_max(void) { return __inline_mdbx_limits_pgsize_max(); }
+
+#endif /* LIBMDBX_NO_EXPORTS_LEGACY_API */

src/api-copy.c

@@ -0,0 +1,880 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \note Please refer to the COPYRIGHT file for explanations license change,
+/// credits and acknowledgments.
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+typedef struct compacting_context {
+  MDBX_env *env;
+  MDBX_txn *txn;
+  MDBX_copy_flags_t flags;
+  pgno_t first_unallocated;
+  osal_condpair_t condpair;
+  volatile unsigned head;
+  volatile unsigned tail;
+  uint8_t *write_buf[2];
+  size_t write_len[2];
+  /* Error code.  Never cleared if set.  Both threads can set nonzero
+   * to fail the copy.  Not mutex-protected, expects atomic int. */
+  volatile int error;
+  mdbx_filehandle_t fd;
+} ctx_t;
+
+__cold static int compacting_walk_tree(ctx_t *ctx, tree_t *tree);
+
+/* Dedicated writer thread for compacting copy. */
+__cold static THREAD_RESULT THREAD_CALL compacting_write_thread(void *arg) {
+  ctx_t *const ctx = arg;
+
+#if defined(EPIPE) && !(defined(_WIN32) || defined(_WIN64))
+  sigset_t sigset;
+  sigemptyset(&sigset);
+  sigaddset(&sigset, SIGPIPE);
+  ctx->error = pthread_sigmask(SIG_BLOCK, &sigset, nullptr);
+#endif /* EPIPE */
+
+  osal_condpair_lock(&ctx->condpair);
+  while (!ctx->error) {
+    while (ctx->tail == ctx->head && !ctx->error) {
+      int err = osal_condpair_wait(&ctx->condpair, true);
+      if (err != MDBX_SUCCESS) {
+        ctx->error = err;
+        goto bailout;
+      }
+    }
+    const unsigned toggle = ctx->tail & 1;
+    size_t wsize = ctx->write_len[toggle];
+    if (wsize == 0) {
+      ctx->tail += 1;
+      break /* EOF */;
+    }
+    ctx->write_len[toggle] = 0;
+    uint8_t *ptr = ctx->write_buf[toggle];
+    if (!ctx->error) {
+      int err = osal_write(ctx->fd, ptr, wsize);
+      if (err != MDBX_SUCCESS) {
+#if defined(EPIPE) && !(defined(_WIN32) || defined(_WIN64))
+        if (err == EPIPE) {
+          /* Collect the pending SIGPIPE,
+           * otherwise at least OS X gives it to the process on thread-exit. */
+          int unused;
+          sigwait(&sigset, &unused);
+        }
+#endif /* EPIPE */
+        ctx->error = err;
+        goto bailout;
+      }
+    }
+    ctx->tail += 1;
+    osal_condpair_signal(&ctx->condpair, false);
+  }
+bailout:
+  osal_condpair_unlock(&ctx->condpair);
+  return (THREAD_RESULT)0;
+}
+
+/* Give buffer and/or MDBX_EOF to writer thread, await unused buffer. */
+__cold static int compacting_toggle_write_buffers(ctx_t *ctx) {
+  osal_condpair_lock(&ctx->condpair);
+  eASSERT(ctx->env, ctx->head - ctx->tail < 2 || ctx->error);
+  ctx->head += 1;
+  osal_condpair_signal(&ctx->condpair, true);
+  while (!ctx->error && ctx->head - ctx->tail == 2 /* both buffers in use */) {
+    if (ctx->flags & MDBX_CP_THROTTLE_MVCC)
+      mdbx_txn_park(ctx->txn, false);
+    int err = osal_condpair_wait(&ctx->condpair, false);
+    if (err == MDBX_SUCCESS && (ctx->flags & MDBX_CP_THROTTLE_MVCC) != 0)
+      err = mdbx_txn_unpark(ctx->txn, false);
+    if (err != MDBX_SUCCESS)
+      ctx->error = err;
+  }
+  osal_condpair_unlock(&ctx->condpair);
+  return ctx->error;
+}
+
+static int compacting_put_bytes(ctx_t *ctx, const void *src, size_t bytes, pgno_t pgno, pgno_t npages) {
+  assert(pgno == 0 || bytes > PAGEHDRSZ);
+  while (bytes > 0) {
+    const size_t side = ctx->head & 1;
+    const size_t left = MDBX_ENVCOPY_WRITEBUF - ctx->write_len[side];
+    if (left < (pgno ? PAGEHDRSZ : 1)) {
+      int err = compacting_toggle_write_buffers(ctx);
+      if (unlikely(err != MDBX_SUCCESS))
+        return err;
+      continue;
+    }
+    const size_t chunk = (bytes < left) ? bytes : left;
+    void *const dst = ctx->write_buf[side] + ctx->write_len[side];
+    if (src) {
+      memcpy(dst, src, chunk);
+      if (pgno) {
+        assert(chunk > PAGEHDRSZ);
+        page_t *mp = dst;
+        mp->pgno = pgno;
+        if (mp->txnid == 0)
+          mp->txnid = ctx->txn->txnid;
+        if (mp->flags == P_LARGE) {
+          assert(bytes <= pgno2bytes(ctx->env, npages));
+          mp->pages = npages;
+        }
+        pgno = 0;
+      }
+      src = ptr_disp(src, chunk);
+    } else
+      memset(dst, 0, chunk);
+    bytes -= chunk;
+    ctx->write_len[side] += chunk;
+  }
+  return MDBX_SUCCESS;
+}
+
+static int compacting_put_page(ctx_t *ctx, const page_t *mp, const size_t head_bytes, const size_t tail_bytes,
+                               const pgno_t npages) {
+  if (tail_bytes) {
+    assert(head_bytes + tail_bytes <= ctx->env->ps);
+    assert(npages == 1 && (page_type(mp) == P_BRANCH || page_type(mp) == P_LEAF));
+  } else {
+    assert(head_bytes <= pgno2bytes(ctx->env, npages));
+    assert((npages == 1 && page_type(mp) == (P_LEAF | P_DUPFIX)) || page_type(mp) == P_LARGE);
+  }
+
+  const pgno_t pgno = ctx->first_unallocated;
+  ctx->first_unallocated += npages;
+  int err = compacting_put_bytes(ctx, mp, head_bytes, pgno, npages);
+  if (unlikely(err != MDBX_SUCCESS))
+    return err;
+  err = compacting_put_bytes(ctx, nullptr, pgno2bytes(ctx->env, npages) - (head_bytes + tail_bytes), 0, 0);
+  if (unlikely(err != MDBX_SUCCESS))
+    return err;
+  return compacting_put_bytes(ctx, ptr_disp(mp, ctx->env->ps - tail_bytes), tail_bytes, 0, 0);
+}
+
+__cold static int compacting_walk(ctx_t *ctx, MDBX_cursor *mc, pgno_t *const parent_pgno, txnid_t parent_txnid) {
+  mc->top = 0;
+  mc->ki[0] = 0;
+  int rc = page_get(mc, *parent_pgno, &mc->pg[0], parent_txnid);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return rc;
+
+  rc = tree_search_finalize(mc, nullptr, Z_FIRST);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return rc;
+
+  /* Make cursor pages writable */
+  const intptr_t deep_limit = mc->top + 1;
+  void *const buf = osal_malloc(pgno2bytes(ctx->env, deep_limit + 1));
+  if (buf == nullptr)
+    return MDBX_ENOMEM;
+
+  void *ptr = buf;
+  for (intptr_t i = 0; i <= mc->top; i++) {
+    page_copy(ptr, mc->pg[i], ctx->env->ps);
+    mc->pg[i] = ptr;
+    ptr = ptr_disp(ptr, ctx->env->ps);
+  }
+  /* This is writable space for a leaf page. Usually not needed. */
+  page_t *const leaf = ptr;
+
+  while (mc->top >= 0) {
+    page_t *mp = mc->pg[mc->top];
+    const size_t nkeys = page_numkeys(mp);
+    if (is_leaf(mp)) {
+      if (!(mc->flags & z_inner) /* may have nested N_TREE or N_BIG nodes */) {
+        for (size_t i = 0; i < nkeys; i++) {
+          node_t *node = page_node(mp, i);
+          if (node_flags(node) == N_BIG) {
+            /* Need writable leaf */
+            if (mp != leaf) {
+              mc->pg[mc->top] = leaf;
+              page_copy(leaf, mp, ctx->env->ps);
+              mp = leaf;
+              node = page_node(mp, i);
+            }
+
+            const pgr_t lp = page_get_large(mc, node_largedata_pgno(node), mp->txnid);
+            if (unlikely((rc = lp.err) != MDBX_SUCCESS))
+              goto bailout;
+            const size_t datasize = node_ds(node);
+            const pgno_t npages = largechunk_npages(ctx->env, datasize);
+            poke_pgno(node_data(node), ctx->first_unallocated);
+            rc = compacting_put_page(ctx, lp.page, PAGEHDRSZ + datasize, 0, npages);
+            if (unlikely(rc != MDBX_SUCCESS))
+              goto bailout;
+          } else if (node_flags(node) & N_TREE) {
+            if (!MDBX_DISABLE_VALIDATION && unlikely(node_ds(node) != sizeof(tree_t))) {
+              ERROR("%s/%d: %s %u", "MDBX_CORRUPTED", MDBX_CORRUPTED, "invalid dupsort sub-tree node size",
+                    (unsigned)node_ds(node));
+              rc = MDBX_CORRUPTED;
+              goto bailout;
+            }
+
+            /* Need writable leaf */
+            if (mp != leaf) {
+              mc->pg[mc->top] = leaf;
+              page_copy(leaf, mp, ctx->env->ps);
+              mp = leaf;
+              node = page_node(mp, i);
+            }
+
+            tree_t *nested = nullptr;
+            if (node_flags(node) & N_DUP) {
+              rc = cursor_dupsort_setup(mc, node, mp);
+              if (likely(rc == MDBX_SUCCESS)) {
+                nested = &mc->subcur->nested_tree;
+                rc = compacting_walk(ctx, &mc->subcur->cursor, &nested->root, mp->txnid);
+              }
+            } else {
+              cASSERT(mc, (mc->flags & z_inner) == 0 && mc->subcur == 0);
+              cursor_couple_t *couple = container_of(mc, cursor_couple_t, outer);
+              nested = &couple->inner.nested_tree;
+              memcpy(nested, node_data(node), sizeof(tree_t));
+              rc = compacting_walk_tree(ctx, nested);
+            }
+            if (unlikely(rc != MDBX_SUCCESS))
+              goto bailout;
+            memcpy(node_data(node), nested, sizeof(tree_t));
+          }
+        }
+      }
+    } else {
+      mc->ki[mc->top]++;
+      if (mc->ki[mc->top] < nkeys) {
+        for (;;) {
+          const node_t *node = page_node(mp, mc->ki[mc->top]);
+          rc = page_get(mc, node_pgno(node), &mp, mp->txnid);
+          if (unlikely(rc != MDBX_SUCCESS))
+            goto bailout;
+          mc->top += 1;
+          if (unlikely(mc->top >= deep_limit)) {
+            rc = MDBX_CURSOR_FULL;
+            goto bailout;
+          }
+          mc->ki[mc->top] = 0;
+          if (!is_branch(mp)) {
+            mc->pg[mc->top] = mp;
+            break;
+          }
+          /* Whenever we advance to a sibling branch page,
+           * we must proceed all the way down to its first leaf. */
+          page_copy(mc->pg[mc->top], mp, ctx->env->ps);
+        }
+        continue;
+      }
+    }
+
+    const pgno_t pgno = ctx->first_unallocated;
+    if (likely(!is_dupfix_leaf(mp))) {
+      rc = compacting_put_page(ctx, mp, PAGEHDRSZ + mp->lower, ctx->env->ps - (PAGEHDRSZ + mp->upper), 1);
+    } else {
+      rc = compacting_put_page(ctx, mp, PAGEHDRSZ + page_numkeys(mp) * mp->dupfix_ksize, 0, 1);
+    }
+    if (unlikely(rc != MDBX_SUCCESS))
+      goto bailout;
+
+    if (mc->top) {
+      /* Update parent if there is one */
+      node_set_pgno(page_node(mc->pg[mc->top - 1], mc->ki[mc->top - 1]), pgno);
+      cursor_pop(mc);
+    } else {
+      /* Otherwise we're done */
+      *parent_pgno = pgno;
+      break;
+    }
+  }
+
+bailout:
+  osal_free(buf);
+  return rc;
+}
+
+__cold static int compacting_walk_tree(ctx_t *ctx, tree_t *tree) {
+  if (unlikely(tree->root == P_INVALID))
+    return MDBX_SUCCESS; /* empty db */
+
+  cursor_couple_t couple;
+  memset(&couple, 0, sizeof(couple));
+  couple.inner.cursor.signature = ~cur_signature_live;
+  kvx_t kvx = {.clc = {.k = {.lmin = INT_MAX}, .v = {.lmin = INT_MAX}}};
+  int rc = cursor_init4walk(&couple, ctx->txn, tree, &kvx);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return rc;
+
+  couple.outer.checking |= z_ignord | z_pagecheck;
+  couple.inner.cursor.checking |= z_ignord | z_pagecheck;
+  if (!tree->mod_txnid)
+    tree->mod_txnid = ctx->txn->txnid;
+  return compacting_walk(ctx, &couple.outer, &tree->root, tree->mod_txnid);
+}
+
+__cold static void compacting_fixup_meta(MDBX_env *env, meta_t *meta) {
+  eASSERT(env, meta->trees.gc.mod_txnid || meta->trees.gc.root == P_INVALID);
+  eASSERT(env, meta->trees.main.mod_txnid || meta->trees.main.root == P_INVALID);
+
+  /* Calculate filesize taking in account shrink/growing thresholds */
+  if (meta->geometry.first_unallocated != meta->geometry.now) {
+    meta->geometry.now = meta->geometry.first_unallocated;
+    const size_t aligner = pv2pages(meta->geometry.grow_pv ? meta->geometry.grow_pv : meta->geometry.shrink_pv);
+    if (aligner) {
+      const pgno_t aligned = pgno_align2os_pgno(env, meta->geometry.first_unallocated + aligner -
+                                                         meta->geometry.first_unallocated % aligner);
+      meta->geometry.now = aligned;
+    }
+  }
+
+  if (meta->geometry.now < meta->geometry.lower)
+    meta->geometry.now = meta->geometry.lower;
+  if (meta->geometry.now > meta->geometry.upper)
+    meta->geometry.now = meta->geometry.upper;
+
+  /* Update signature */
+  assert(meta->geometry.now >= meta->geometry.first_unallocated);
+  meta_sign_as_steady(meta);
+}
+
+/* Make resizable */
+__cold static void meta_make_sizeable(meta_t *meta) {
+  meta->geometry.lower = MIN_PAGENO;
+  if (meta->geometry.grow_pv == 0) {
+    const pgno_t step = 1 + (meta->geometry.upper - meta->geometry.lower) / 42;
+    meta->geometry.grow_pv = pages2pv(step);
+  }
+  if (meta->geometry.shrink_pv == 0) {
+    const pgno_t step = pv2pages(meta->geometry.grow_pv) << 1;
+    meta->geometry.shrink_pv = pages2pv(step);
+  }
+}
+
+__cold static int copy_with_compacting(MDBX_env *env, MDBX_txn *txn, mdbx_filehandle_t fd, uint8_t *buffer,
+                                       const bool dest_is_pipe, const MDBX_copy_flags_t flags) {
+  const size_t meta_bytes = pgno2bytes(env, NUM_METAS);
+  uint8_t *const data_buffer = buffer + ceil_powerof2(meta_bytes, globals.sys_pagesize);
+  meta_t *const meta = meta_init_triplet(env, buffer);
+  meta_set_txnid(env, meta, txn->txnid);
+
+  if (flags & MDBX_CP_FORCE_DYNAMIC_SIZE)
+    meta_make_sizeable(meta);
+
+  /* copy canary sequences if present */
+  if (txn->canary.v) {
+    meta->canary = txn->canary;
+    meta->canary.v = constmeta_txnid(meta);
+  }
+
+  if (txn->dbs[MAIN_DBI].root == P_INVALID) {
+    /* When the DB is empty, handle it specially to
+     * fix any breakage like page leaks from ITS#8174. */
+    meta->trees.main.flags = txn->dbs[MAIN_DBI].flags;
+    compacting_fixup_meta(env, meta);
+    if (dest_is_pipe) {
+      if (flags & MDBX_CP_THROTTLE_MVCC)
+        mdbx_txn_park(txn, false);
+      int rc = osal_write(fd, buffer, meta_bytes);
+      if (likely(rc == MDBX_SUCCESS) && (flags & MDBX_CP_THROTTLE_MVCC) != 0)
+        rc = mdbx_txn_unpark(txn, false);
+      if (unlikely(rc != MDBX_SUCCESS))
+        return rc;
+    }
+  } else {
+    /* Count free pages + GC pages. */
+    cursor_couple_t couple;
+    int rc = cursor_init(&couple.outer, txn, FREE_DBI);
+    if (unlikely(rc != MDBX_SUCCESS))
+      return rc;
+    pgno_t gc_npages = txn->dbs[FREE_DBI].branch_pages + txn->dbs[FREE_DBI].leaf_pages + txn->dbs[FREE_DBI].large_pages;
+    MDBX_val key, data;
+    rc = outer_first(&couple.outer, &key, &data);
+    while (rc == MDBX_SUCCESS) {
+      const pnl_t pnl = data.iov_base;
+      if (unlikely(data.iov_len % sizeof(pgno_t) || data.iov_len < MDBX_PNL_SIZEOF(pnl))) {
+        ERROR("%s/%d: %s %zu", "MDBX_CORRUPTED", MDBX_CORRUPTED, "invalid GC-record length", data.iov_len);
+        return MDBX_CORRUPTED;
+      }
+      if (unlikely(!pnl_check(pnl, txn->geo.first_unallocated))) {
+        ERROR("%s/%d: %s", "MDBX_CORRUPTED", MDBX_CORRUPTED, "invalid GC-record content");
+        return MDBX_CORRUPTED;
+      }
+      gc_npages += MDBX_PNL_GETSIZE(pnl);
+      rc = outer_next(&couple.outer, &key, &data, MDBX_NEXT);
+    }
+    if (unlikely(rc != MDBX_NOTFOUND))
+      return rc;
+
+    meta->geometry.first_unallocated = txn->geo.first_unallocated - gc_npages;
+    meta->trees.main = txn->dbs[MAIN_DBI];
+
+    ctx_t ctx;
+    memset(&ctx, 0, sizeof(ctx));
+    rc = osal_condpair_init(&ctx.condpair);
+    if (unlikely(rc != MDBX_SUCCESS))
+      return rc;
+
+    memset(data_buffer, 0, 2 * (size_t)MDBX_ENVCOPY_WRITEBUF);
+    ctx.write_buf[0] = data_buffer;
+    ctx.write_buf[1] = data_buffer + (size_t)MDBX_ENVCOPY_WRITEBUF;
+    ctx.first_unallocated = NUM_METAS;
+    ctx.env = env;
+    ctx.fd = fd;
+    ctx.txn = txn;
+    ctx.flags = flags;
+
+    osal_thread_t thread;
+    int thread_err = osal_thread_create(&thread, compacting_write_thread, &ctx);
+    if (likely(thread_err == MDBX_SUCCESS)) {
+      if (dest_is_pipe) {
+        if (!meta->trees.main.mod_txnid)
+          meta->trees.main.mod_txnid = txn->txnid;
+        compacting_fixup_meta(env, meta);
+        if (flags & MDBX_CP_THROTTLE_MVCC)
+          mdbx_txn_park(txn, false);
+        rc = osal_write(fd, buffer, meta_bytes);
+        if (likely(rc == MDBX_SUCCESS) && (flags & MDBX_CP_THROTTLE_MVCC) != 0)
+          rc = mdbx_txn_unpark(txn, false);
+      }
+      if (likely(rc == MDBX_SUCCESS))
+        rc = compacting_walk_tree(&ctx, &meta->trees.main);
+      if (ctx.write_len[ctx.head & 1])
+        /* toggle to flush non-empty buffers */
+        compacting_toggle_write_buffers(&ctx);
+
+      if (likely(rc == MDBX_SUCCESS) && unlikely(meta->geometry.first_unallocated != ctx.first_unallocated)) {
+        if (ctx.first_unallocated > meta->geometry.first_unallocated) {
+          ERROR("the source DB %s: post-compactification used pages %" PRIaPGNO " %c expected %" PRIaPGNO,
+                "has double-used pages or other corruption", ctx.first_unallocated, '>',
+                meta->geometry.first_unallocated);
+          rc = MDBX_CORRUPTED; /* corrupted DB */
+        }
+        if (ctx.first_unallocated < meta->geometry.first_unallocated) {
+          WARNING("the source DB %s: post-compactification used pages %" PRIaPGNO " %c expected %" PRIaPGNO,
+                  "has page leak(s)", ctx.first_unallocated, '<', meta->geometry.first_unallocated);
+          if (dest_is_pipe)
+            /* the root within already written meta-pages is wrong */
+            rc = MDBX_CORRUPTED;
+        }
+        /* fixup meta */
+        meta->geometry.first_unallocated = ctx.first_unallocated;
+      }
+
+      /* toggle with empty buffers to exit thread's loop */
+      eASSERT(env, (ctx.write_len[ctx.head & 1]) == 0);
+      compacting_toggle_write_buffers(&ctx);
+      thread_err = osal_thread_join(thread);
+      eASSERT(env, (ctx.tail == ctx.head && ctx.write_len[ctx.head & 1] == 0) || ctx.error);
+      osal_condpair_destroy(&ctx.condpair);
+    }
+    if (unlikely(thread_err != MDBX_SUCCESS))
+      return thread_err;
+    if (unlikely(rc != MDBX_SUCCESS))
+      return rc;
+    if (unlikely(ctx.error != MDBX_SUCCESS))
+      return ctx.error;
+    if (!dest_is_pipe)
+      compacting_fixup_meta(env, meta);
+  }
+
+  if (flags & MDBX_CP_THROTTLE_MVCC)
+    mdbx_txn_park(txn, false);
+
+  /* Extend file if required */
+  if (meta->geometry.now != meta->geometry.first_unallocated) {
+    const size_t whole_size = pgno2bytes(env, meta->geometry.now);
+    if (!dest_is_pipe)
+      return osal_ftruncate(fd, whole_size);
+
+    const size_t used_size = pgno2bytes(env, meta->geometry.first_unallocated);
+    memset(data_buffer, 0, (size_t)MDBX_ENVCOPY_WRITEBUF);
+    for (size_t offset = used_size; offset < whole_size;) {
+      const size_t chunk =
+          ((size_t)MDBX_ENVCOPY_WRITEBUF < whole_size - offset) ? (size_t)MDBX_ENVCOPY_WRITEBUF : whole_size - offset;
+      int rc = osal_write(fd, data_buffer, chunk);
+      if (unlikely(rc != MDBX_SUCCESS))
+        return rc;
+      offset += chunk;
+    }
+  }
+  return MDBX_SUCCESS;
+}
+
+//----------------------------------------------------------------------------
+
+__cold static int copy_asis(MDBX_env *env, MDBX_txn *txn, mdbx_filehandle_t fd, uint8_t *buffer,
+                            const bool dest_is_pipe, const MDBX_copy_flags_t flags) {
+  bool should_unlock = false;
+  if ((txn->flags & MDBX_TXN_RDONLY) != 0 && (flags & MDBX_CP_RENEW_TXN) != 0) {
+    /* Try temporarily block writers until we snapshot the meta pages */
+    int err = lck_txn_lock(env, true);
+    if (likely(err == MDBX_SUCCESS))
+      should_unlock = true;
+    else if (unlikely(err != MDBX_BUSY))
+      return err;
+  }
+
+  jitter4testing(false);
+  int rc = MDBX_SUCCESS;
+  const size_t meta_bytes = pgno2bytes(env, NUM_METAS);
+  troika_t troika = meta_tap(env);
+  /* Make a snapshot of meta-pages,
+   * but writing ones after the data was flushed */
+retry_snap_meta:
+  memcpy(buffer, env->dxb_mmap.base, meta_bytes);
+  const meta_ptr_t recent = meta_recent(env, &troika);
+  meta_t *headcopy = /* LY: get pointer to the snapshot copy */
+      ptr_disp(buffer, ptr_dist(recent.ptr_c, env->dxb_mmap.base));
+  jitter4testing(false);
+  if (txn->flags & MDBX_TXN_RDONLY) {
+    if (recent.txnid != txn->txnid) {
+      if (flags & MDBX_CP_RENEW_TXN)
+        rc = mdbx_txn_renew(txn);
+      else {
+        rc = MDBX_MVCC_RETARDED;
+        for (size_t n = 0; n < NUM_METAS; ++n) {
+          meta_t *const meta = page_meta(ptr_disp(buffer, pgno2bytes(env, n)));
+          if (troika.txnid[n] == txn->txnid && ((/* is_steady */ (troika.fsm >> n) & 1) || rc != MDBX_SUCCESS)) {
+            rc = MDBX_SUCCESS;
+            headcopy = meta;
+          } else if (troika.txnid[n] > txn->txnid)
+            meta_set_txnid(env, meta, 0);
+        }
+      }
+    }
+    if (should_unlock)
+      lck_txn_unlock(env);
+    else {
+      troika_t snap = meta_tap(env);
+      if (memcmp(&troika, &snap, sizeof(troika_t)) && rc == MDBX_SUCCESS) {
+        troika = snap;
+        goto retry_snap_meta;
+      }
+    }
+  }
+  if (unlikely(rc != MDBX_SUCCESS))
+    return rc;
+
+  if (txn->flags & MDBX_TXN_RDONLY)
+    eASSERT(env, meta_txnid(headcopy) == txn->txnid);
+  if (flags & MDBX_CP_FORCE_DYNAMIC_SIZE)
+    meta_make_sizeable(headcopy);
+  /* Update signature to steady */
+  meta_sign_as_steady(headcopy);
+
+  /* Copy the data */
+  const size_t whole_size = pgno_align2os_bytes(env, txn->geo.end_pgno);
+  const size_t used_size = pgno2bytes(env, txn->geo.first_unallocated);
+  jitter4testing(false);
+
+  if (flags & MDBX_CP_THROTTLE_MVCC)
+    mdbx_txn_park(txn, false);
+
+  if (dest_is_pipe)
+    rc = osal_write(fd, buffer, meta_bytes);
+
+  uint8_t *const data_buffer = buffer + ceil_powerof2(meta_bytes, globals.sys_pagesize);
+#if MDBX_USE_COPYFILERANGE
+  static bool copyfilerange_unavailable;
+#if (defined(__linux__) || defined(__gnu_linux__))
+  if (globals.linux_kernel_version >= 0x05030000 && globals.linux_kernel_version < 0x05130000)
+    copyfilerange_unavailable = true;
+#endif /* linux */
+  bool not_the_same_filesystem = false;
+  if (!copyfilerange_unavailable) {
+    struct statfs statfs_info;
+    if (fstatfs(fd, &statfs_info) || statfs_info.f_type == /* ECRYPTFS_SUPER_MAGIC */ 0xf15f)
+      /* avoid use copyfilerange_unavailable() to ecryptfs due bugs */
+      not_the_same_filesystem = true;
+  }
+#endif /* MDBX_USE_COPYFILERANGE */
+
+  for (size_t offset = meta_bytes; rc == MDBX_SUCCESS && offset < used_size;) {
+    if (flags & MDBX_CP_THROTTLE_MVCC) {
+      rc = mdbx_txn_unpark(txn, false);
+      if (unlikely(rc != MDBX_SUCCESS))
+        break;
+    }
+
+#if MDBX_USE_SENDFILE
+    static bool sendfile_unavailable;
+    if (dest_is_pipe && likely(!sendfile_unavailable)) {
+      off_t in_offset = offset;
+      const ssize_t written = sendfile(fd, env->lazy_fd, &in_offset, used_size - offset);
+      if (likely(written > 0)) {
+        offset = in_offset;
+        if (flags & MDBX_CP_THROTTLE_MVCC)
+          rc = mdbx_txn_park(txn, false);
+        continue;
+      }
+      rc = MDBX_ENODATA;
+      if (written == 0 || ignore_enosys_and_eagain(rc = errno) != MDBX_RESULT_TRUE)
+        break;
+      sendfile_unavailable = true;
+    }
+#endif /* MDBX_USE_SENDFILE */
+
+#if MDBX_USE_COPYFILERANGE
+    if (!dest_is_pipe && !not_the_same_filesystem && likely(!copyfilerange_unavailable)) {
+      off_t in_offset = offset, out_offset = offset;
+      ssize_t bytes_copied = copy_file_range(env->lazy_fd, &in_offset, fd, &out_offset, used_size - offset, 0);
+      if (likely(bytes_copied > 0)) {
+        offset = in_offset;
+        if (flags & MDBX_CP_THROTTLE_MVCC)
+          rc = mdbx_txn_park(txn, false);
+        continue;
+      }
+      rc = MDBX_ENODATA;
+      if (bytes_copied == 0)
+        break;
+      rc = errno;
+      if (rc == EXDEV || rc == /* workaround for ecryptfs bug(s),
+                                  maybe useful for others FS */
+                             EINVAL)
+        not_the_same_filesystem = true;
+      else if (ignore_enosys_and_eagain(rc) == MDBX_RESULT_TRUE)
+        copyfilerange_unavailable = true;
+      else
+        break;
+    }
+#endif /* MDBX_USE_COPYFILERANGE */
+
+    /* fallback to portable */
+    const size_t chunk =
+        ((size_t)MDBX_ENVCOPY_WRITEBUF < used_size - offset) ? (size_t)MDBX_ENVCOPY_WRITEBUF : used_size - offset;
+    /* copy to avoid EFAULT in case swapped-out */
+    memcpy(data_buffer, ptr_disp(env->dxb_mmap.base, offset), chunk);
+    if (flags & MDBX_CP_THROTTLE_MVCC)
+      mdbx_txn_park(txn, false);
+    rc = osal_write(fd, data_buffer, chunk);
+    offset += chunk;
+  }
+
+  /* Extend file if required */
+  if (likely(rc == MDBX_SUCCESS) && whole_size != used_size) {
+    if (!dest_is_pipe)
+      rc = osal_ftruncate(fd, whole_size);
+    else {
+      memset(data_buffer, 0, (size_t)MDBX_ENVCOPY_WRITEBUF);
+      for (size_t offset = used_size; rc == MDBX_SUCCESS && offset < whole_size;) {
+        const size_t chunk =
+            ((size_t)MDBX_ENVCOPY_WRITEBUF < whole_size - offset) ? (size_t)MDBX_ENVCOPY_WRITEBUF : whole_size - offset;
+        rc = osal_write(fd, data_buffer, chunk);
+        offset += chunk;
+      }
+    }
+  }
+
+  return rc;
+}
+
+//----------------------------------------------------------------------------
+
+__cold static int copy2fd(MDBX_txn *txn, mdbx_filehandle_t fd, MDBX_copy_flags_t flags) {
+  if (unlikely(txn->flags & MDBX_TXN_DIRTY))
+    return MDBX_BAD_TXN;
+
+  int rc = MDBX_SUCCESS;
+  if (txn->flags & MDBX_TXN_RDONLY) {
+    if (flags & MDBX_CP_THROTTLE_MVCC) {
+      rc = mdbx_txn_park(txn, true);
+      if (unlikely(rc != MDBX_SUCCESS))
+        return rc;
+    }
+  } else if (unlikely(flags & (MDBX_CP_THROTTLE_MVCC | MDBX_CP_RENEW_TXN)))
+    return MDBX_EINVAL;
+
+  const int dest_is_pipe = osal_is_pipe(fd);
+  if (MDBX_IS_ERROR(dest_is_pipe))
+    return dest_is_pipe;
+
+  if (!dest_is_pipe) {
+    rc = osal_fseek(fd, 0);
+    if (unlikely(rc != MDBX_SUCCESS))
+      return rc;
+  }
+
+  MDBX_env *const env = txn->env;
+  const size_t buffer_size =
+      pgno_align2os_bytes(env, NUM_METAS) +
+      ceil_powerof2(((flags & MDBX_CP_COMPACT) ? 2 * (size_t)MDBX_ENVCOPY_WRITEBUF : (size_t)MDBX_ENVCOPY_WRITEBUF),
+                    globals.sys_pagesize);
+
+  uint8_t *buffer = nullptr;
+  rc = osal_memalign_alloc(globals.sys_pagesize, buffer_size, (void **)&buffer);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return rc;
+
+  if (!dest_is_pipe) {
+    /* Firstly write a stub to meta-pages.
+     * Now we sure to incomplete copy will not be used. */
+    memset(buffer, -1, pgno2bytes(env, NUM_METAS));
+    rc = osal_write(fd, buffer, pgno2bytes(env, NUM_METAS));
+  }
+
+  if (likely(rc == MDBX_SUCCESS))
+    rc = mdbx_txn_unpark(txn, false);
+  if (likely(rc == MDBX_SUCCESS)) {
+    memset(buffer, 0, pgno2bytes(env, NUM_METAS));
+    rc = ((flags & MDBX_CP_COMPACT) ? copy_with_compacting : copy_asis)(env, txn, fd, buffer, dest_is_pipe, flags);
+
+    if (likely(rc == MDBX_SUCCESS))
+      rc = mdbx_txn_unpark(txn, false);
+  }
+
+  if (txn->flags & MDBX_TXN_RDONLY) {
+    if (flags & MDBX_CP_THROTTLE_MVCC)
+      mdbx_txn_park(txn, true);
+    else if (flags & MDBX_CP_DISPOSE_TXN)
+      mdbx_txn_reset(txn);
+  }
+
+  if (!dest_is_pipe) {
+    if (likely(rc == MDBX_SUCCESS) && (flags & MDBX_CP_DONT_FLUSH) == 0)
+      rc = osal_fsync(fd, MDBX_SYNC_DATA | MDBX_SYNC_SIZE);
+
+    /* Write actual meta */
+    if (likely(rc == MDBX_SUCCESS))
+      rc = osal_pwrite(fd, buffer, pgno2bytes(env, NUM_METAS), 0);
+
+    if (likely(rc == MDBX_SUCCESS) && (flags & MDBX_CP_DONT_FLUSH) == 0)
+      rc = osal_fsync(fd, MDBX_SYNC_DATA | MDBX_SYNC_IODQ);
+  }
+
+  osal_memalign_free(buffer);
+  return rc;
+}
+
+__cold static int copy2pathname(MDBX_txn *txn, const pathchar_t *dest_path, MDBX_copy_flags_t flags) {
+  if (unlikely(!dest_path || *dest_path == '\0'))
+    return MDBX_EINVAL;
+
+  /* The destination path must exist, but the destination file must not.
+   * We don't want the OS to cache the writes, since the source data is
+   * already in the OS cache. */
+  mdbx_filehandle_t newfd = INVALID_HANDLE_VALUE;
+  int rc = osal_openfile(MDBX_OPEN_COPY, txn->env, dest_path, &newfd,
+#if defined(_WIN32) || defined(_WIN64)
+                         (mdbx_mode_t)-1
+#else
+                         S_IRUSR | S_IWUSR | S_IRGRP | S_IWGRP
+#endif
+  );
+
+#if defined(_WIN32) || defined(_WIN64)
+  /* no locking required since the file opened with ShareMode == 0 */
+#else
+  if (rc == MDBX_SUCCESS) {
+    MDBX_STRUCT_FLOCK lock_op;
+    memset(&lock_op, 0, sizeof(lock_op));
+    lock_op.l_type = F_WRLCK;
+    lock_op.l_whence = SEEK_SET;
+    lock_op.l_start = 0;
+    lock_op.l_len = OFF_T_MAX;
+    if (MDBX_FCNTL(newfd, MDBX_F_SETLK, &lock_op))
+      rc = errno;
+  }
+
+#if defined(LOCK_EX) && (!defined(__ANDROID_API__) || __ANDROID_API__ >= 24)
+  if (rc == MDBX_SUCCESS && flock(newfd, LOCK_EX | LOCK_NB)) {
+    const int err_flock = errno, err_fs = osal_check_fs_local(newfd, 0);
+    if (err_flock != EAGAIN || err_fs != MDBX_EREMOTE) {
+      ERROR("%s flock(%" MDBX_PRIsPATH ") error %d, remote-fs check status %d", "unexpected", dest_path, err_flock,
+            err_fs);
+      rc = err_flock;
+    } else {
+      WARNING("%s flock(%" MDBX_PRIsPATH ") error %d, remote-fs check status %d", "ignore", dest_path, err_flock,
+              err_fs);
+    }
+  }
+#endif /* LOCK_EX && ANDROID_API >= 24 */
+
+#endif /* Windows / POSIX */
+
+  if (rc == MDBX_SUCCESS)
+    rc = copy2fd(txn, newfd, flags);
+
+  if (newfd != INVALID_HANDLE_VALUE) {
+    int err = osal_closefile(newfd);
+    if (rc == MDBX_SUCCESS && err != rc)
+      rc = err;
+    if (rc != MDBX_SUCCESS)
+      (void)osal_removefile(dest_path);
+  }
+  return rc;
+}
+
+//----------------------------------------------------------------------------
+
+__cold int mdbx_txn_copy2fd(MDBX_txn *txn, mdbx_filehandle_t fd, MDBX_copy_flags_t flags) {
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED);
+  if (likely(rc == MDBX_SUCCESS))
+    rc = copy2fd(txn, fd, flags);
+  if (flags & MDBX_CP_DISPOSE_TXN)
+    mdbx_txn_abort(txn);
+  return LOG_IFERR(rc);
+}
+
+__cold int mdbx_env_copy2fd(MDBX_env *env, mdbx_filehandle_t fd, MDBX_copy_flags_t flags) {
+  if (unlikely(flags & (MDBX_CP_DISPOSE_TXN | MDBX_CP_RENEW_TXN)))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = check_env(env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  MDBX_txn *txn = nullptr;
+  rc = mdbx_txn_begin(env, nullptr, MDBX_TXN_RDONLY, &txn);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  rc = copy2fd(txn, fd, flags | MDBX_CP_DISPOSE_TXN | MDBX_CP_RENEW_TXN);
+  mdbx_txn_abort(txn);
+  return LOG_IFERR(rc);
+}
+
+__cold int mdbx_txn_copy2pathname(MDBX_txn *txn, const char *dest_path, MDBX_copy_flags_t flags) {
+#if defined(_WIN32) || defined(_WIN64)
+  wchar_t *dest_pathW = nullptr;
+  int rc = osal_mb2w(dest_path, &dest_pathW);
+  if (likely(rc == MDBX_SUCCESS)) {
+    rc = mdbx_txn_copy2pathnameW(txn, dest_pathW, flags);
+    osal_free(dest_pathW);
+  }
+  return LOG_IFERR(rc);
+}
+
+__cold int mdbx_txn_copy2pathnameW(MDBX_txn *txn, const wchar_t *dest_path, MDBX_copy_flags_t flags) {
+#endif /* Windows */
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED);
+  if (likely(rc == MDBX_SUCCESS))
+    rc = copy2pathname(txn, dest_path, flags);
+  if (flags & MDBX_CP_DISPOSE_TXN)
+    mdbx_txn_abort(txn);
+  return LOG_IFERR(rc);
+}
+
+__cold int mdbx_env_copy(MDBX_env *env, const char *dest_path, MDBX_copy_flags_t flags) {
+#if defined(_WIN32) || defined(_WIN64)
+  wchar_t *dest_pathW = nullptr;
+  int rc = osal_mb2w(dest_path, &dest_pathW);
+  if (likely(rc == MDBX_SUCCESS)) {
+    rc = mdbx_env_copyW(env, dest_pathW, flags);
+    osal_free(dest_pathW);
+  }
+  return LOG_IFERR(rc);
+}
+
+__cold int mdbx_env_copyW(MDBX_env *env, const wchar_t *dest_path, MDBX_copy_flags_t flags) {
+#endif /* Windows */
+  if (unlikely(flags & (MDBX_CP_DISPOSE_TXN | MDBX_CP_RENEW_TXN)))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = check_env(env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  MDBX_txn *txn = nullptr;
+  rc = mdbx_txn_begin(env, nullptr, MDBX_TXN_RDONLY, &txn);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  rc = copy2pathname(txn, dest_path, flags | MDBX_CP_DISPOSE_TXN | MDBX_CP_RENEW_TXN);
+  mdbx_txn_abort(txn);
+  return LOG_IFERR(rc);
+}

src/api-cursor.c

@@ -0,0 +1,755 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+MDBX_cursor *mdbx_cursor_create(void *context) {
+  cursor_couple_t *couple = osal_calloc(1, sizeof(cursor_couple_t));
+  if (unlikely(!couple))
+    return nullptr;
+
+  VALGRIND_MAKE_MEM_UNDEFINED(couple, sizeof(cursor_couple_t));
+  couple->outer.signature = cur_signature_ready4dispose;
+  couple->outer.next = &couple->outer;
+  couple->userctx = context;
+  cursor_reset(couple);
+  VALGRIND_MAKE_MEM_DEFINED(&couple->outer.backup, sizeof(couple->outer.backup));
+  VALGRIND_MAKE_MEM_DEFINED(&couple->outer.tree, sizeof(couple->outer.tree));
+  VALGRIND_MAKE_MEM_DEFINED(&couple->outer.clc, sizeof(couple->outer.clc));
+  VALGRIND_MAKE_MEM_DEFINED(&couple->outer.dbi_state, sizeof(couple->outer.dbi_state));
+  VALGRIND_MAKE_MEM_DEFINED(&couple->outer.subcur, sizeof(couple->outer.subcur));
+  VALGRIND_MAKE_MEM_DEFINED(&couple->outer.txn, sizeof(couple->outer.txn));
+  return &couple->outer;
+}
+
+int mdbx_cursor_renew(MDBX_txn *txn, MDBX_cursor *mc) {
+  return likely(mc) ? mdbx_cursor_bind(txn, mc, (kvx_t *)mc->clc - txn->env->kvs) : LOG_IFERR(MDBX_EINVAL);
+}
+
+int mdbx_cursor_reset(MDBX_cursor *mc) {
+  int rc = cursor_check(mc, MDBX_TXN_FINISHED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cursor_reset((cursor_couple_t *)mc);
+  return MDBX_SUCCESS;
+}
+
+int mdbx_cursor_bind(MDBX_txn *txn, MDBX_cursor *mc, MDBX_dbi dbi) {
+  if (unlikely(!mc))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  if (unlikely(mc->signature != cur_signature_ready4dispose && mc->signature != cur_signature_live)) {
+    int rc = (mc->signature == cur_signature_wait4eot) ? MDBX_EINVAL : MDBX_EBADSIGN;
+    return LOG_IFERR(rc);
+  }
+
+  int rc = check_txn(txn, MDBX_TXN_FINISHED | MDBX_TXN_HAS_CHILD);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(dbi == FREE_DBI && !(txn->flags & MDBX_TXN_RDONLY)))
+    return LOG_IFERR(MDBX_EACCESS);
+
+  rc = dbi_check(txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(mc->backup)) /* Cursor from parent transaction */
+    LOG_IFERR(MDBX_EINVAL);
+
+  if (mc->signature == cur_signature_live) {
+    if (mc->txn == txn && cursor_dbi(mc) == dbi)
+      return MDBX_SUCCESS;
+    rc = mdbx_cursor_unbind(mc);
+    if (unlikely(rc != MDBX_SUCCESS))
+      return (rc == MDBX_BAD_TXN) ? MDBX_EINVAL : rc;
+  }
+  cASSERT(mc, mc->next == mc);
+
+  rc = cursor_init(mc, txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  mc->next = txn->cursors[dbi];
+  txn->cursors[dbi] = mc;
+  txn->flags |= txn_may_have_cursors;
+  return MDBX_SUCCESS;
+}
+
+int mdbx_cursor_unbind(MDBX_cursor *mc) {
+  if (unlikely(!mc))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  if (unlikely(mc->signature != cur_signature_live))
+    return (mc->signature == cur_signature_ready4dispose) ? MDBX_SUCCESS : LOG_IFERR(MDBX_EBADSIGN);
+
+  if (unlikely(mc->backup)) /* Cursor from parent transaction */
+    /* TODO: реализовать при переходе на двусвязный список курсоров */
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = check_txn(mc->txn, MDBX_TXN_FINISHED | MDBX_TXN_HAS_CHILD);
+  if (unlikely(rc != MDBX_SUCCESS)) {
+    for (const MDBX_txn *txn = mc->txn; rc == MDBX_BAD_TXN && check_txn(txn, MDBX_TXN_FINISHED) == MDBX_SUCCESS;
+         txn = txn->nested)
+      if (dbi_state(txn, cursor_dbi(mc)) == 0)
+        /* специальный случай: курсор прикреплён к родительской транзакции, но соответствующий dbi-дескриптор ещё
+         * не использовался во вложенной транзакции, т.е. курсор ещё не импортирован в дочернюю транзакцию и не имеет
+         * связанного сохранённого состояния (поэтому mc→backup равен nullptr). */
+        rc = MDBX_EINVAL;
+    return LOG_IFERR(rc);
+  }
+
+  if (unlikely(!mc->txn || mc->txn->signature != txn_signature)) {
+    ERROR("Wrong cursor's transaction %p 0x%x", __Wpedantic_format_voidptr(mc->txn), mc->txn ? mc->txn->signature : 0);
+    return LOG_IFERR(MDBX_PROBLEM);
+  }
+
+  if (mc->next != mc) {
+    const size_t dbi = cursor_dbi(mc);
+    cASSERT(mc, dbi < mc->txn->n_dbi);
+    cASSERT(mc, &mc->txn->env->kvs[dbi].clc == mc->clc);
+    if (dbi < mc->txn->n_dbi) {
+      MDBX_cursor **prev = &mc->txn->cursors[dbi];
+      while (/* *prev && */ *prev != mc) {
+        ENSURE(mc->txn->env, (*prev)->signature == cur_signature_live || (*prev)->signature == cur_signature_wait4eot);
+        prev = &(*prev)->next;
+      }
+      cASSERT(mc, *prev == mc);
+      *prev = mc->next;
+    }
+    mc->next = mc;
+  }
+  cursor_drown((cursor_couple_t *)mc);
+  mc->signature = cur_signature_ready4dispose;
+  return MDBX_SUCCESS;
+}
+
+int mdbx_cursor_open(MDBX_txn *txn, MDBX_dbi dbi, MDBX_cursor **ret) {
+  if (unlikely(!ret))
+    return LOG_IFERR(MDBX_EINVAL);
+  *ret = nullptr;
+
+  MDBX_cursor *const mc = mdbx_cursor_create(nullptr);
+  if (unlikely(!mc))
+    return LOG_IFERR(MDBX_ENOMEM);
+
+  int rc = mdbx_cursor_bind(txn, mc, dbi);
+  if (unlikely(rc != MDBX_SUCCESS)) {
+    mdbx_cursor_close(mc);
+    return LOG_IFERR(rc);
+  }
+
+  *ret = mc;
+  return MDBX_SUCCESS;
+}
+
+void mdbx_cursor_close(MDBX_cursor *cursor) {
+  if (likely(cursor)) {
+    int err = mdbx_cursor_close2(cursor);
+    if (unlikely(err != MDBX_SUCCESS))
+      mdbx_panic("%s:%d error %d (%s) while closing cursor", __func__, __LINE__, err, mdbx_liberr2str(err));
+  }
+}
+
+int mdbx_cursor_close2(MDBX_cursor *mc) {
+  if (unlikely(!mc))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  if (mc->signature == cur_signature_ready4dispose) {
+    if (unlikely(mc->txn || mc->backup))
+      return LOG_IFERR(MDBX_PANIC);
+    cursor_drown((cursor_couple_t *)mc);
+    mc->signature = 0;
+    osal_free(mc);
+    return MDBX_SUCCESS;
+  }
+
+  if (unlikely(mc->signature != cur_signature_live))
+    return LOG_IFERR(MDBX_EBADSIGN);
+
+  MDBX_txn *const txn = mc->txn;
+  int rc = check_txn(txn, MDBX_TXN_FINISHED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (mc->backup) {
+    /* Cursor closed before nested txn ends */
+    cursor_reset((cursor_couple_t *)mc);
+    mc->signature = cur_signature_wait4eot;
+    return MDBX_SUCCESS;
+  }
+
+  if (mc->next != mc) {
+    const size_t dbi = cursor_dbi(mc);
+    cASSERT(mc, dbi < mc->txn->n_dbi);
+    cASSERT(mc, &mc->txn->env->kvs[dbi].clc == mc->clc);
+    if (likely(dbi < txn->n_dbi)) {
+      MDBX_cursor **prev = &txn->cursors[dbi];
+      while (/* *prev && */ *prev != mc) {
+        ENSURE(txn->env, (*prev)->signature == cur_signature_live || (*prev)->signature == cur_signature_wait4eot);
+        prev = &(*prev)->next;
+      }
+      tASSERT(txn, *prev == mc);
+      *prev = mc->next;
+    }
+    mc->next = mc;
+  }
+  cursor_drown((cursor_couple_t *)mc);
+  mc->signature = 0;
+  osal_free(mc);
+  return MDBX_SUCCESS;
+}
+
+int mdbx_cursor_copy(const MDBX_cursor *src, MDBX_cursor *dest) {
+  int rc = cursor_check(src, MDBX_TXN_FINISHED | MDBX_TXN_HAS_CHILD);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  rc = mdbx_cursor_bind(src->txn, dest, cursor_dbi(src));
+  if (unlikely(rc != MDBX_SUCCESS))
+    return rc;
+
+  assert(dest->tree == src->tree);
+  assert(cursor_dbi(dest) == cursor_dbi(src));
+again:
+  assert(dest->clc == src->clc);
+  assert(dest->txn == src->txn);
+  dest->top_and_flags = src->top_and_flags;
+  for (intptr_t i = 0; i <= src->top; ++i) {
+    dest->ki[i] = src->ki[i];
+    dest->pg[i] = src->pg[i];
+  }
+
+  if (src->subcur) {
+    dest->subcur->nested_tree = src->subcur->nested_tree;
+    src = &src->subcur->cursor;
+    dest = &dest->subcur->cursor;
+    goto again;
+  }
+
+  return MDBX_SUCCESS;
+}
+
+int mdbx_txn_release_all_cursors_ex(const MDBX_txn *txn, bool unbind, size_t *count) {
+  int rc = check_txn(txn, MDBX_TXN_FINISHED | MDBX_TXN_HAS_CHILD);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  size_t n = 0;
+  do {
+    TXN_FOREACH_DBI_FROM(txn, i, MAIN_DBI) {
+      MDBX_cursor *mc = txn->cursors[i], *next = nullptr;
+      if (mc) {
+        txn->cursors[i] = nullptr;
+        do {
+          next = mc->next;
+          if (mc->signature == cur_signature_live) {
+            mc->signature = cur_signature_wait4eot;
+            cursor_drown((cursor_couple_t *)mc);
+          } else
+            ENSURE(nullptr, mc->signature == cur_signature_wait4eot);
+          if (mc->backup) {
+            MDBX_cursor *bk = mc->backup;
+            mc->next = bk->next;
+            mc->backup = bk->backup;
+            bk->backup = nullptr;
+            bk->signature = 0;
+            osal_free(bk);
+          } else {
+            mc->signature = cur_signature_ready4dispose;
+            mc->next = mc;
+            ++n;
+            if (!unbind) {
+              mc->signature = 0;
+              osal_free(mc);
+            }
+          }
+        } while ((mc = next) != nullptr);
+      }
+    }
+    txn = txn->parent;
+  } while (txn);
+
+  if (count)
+    *count = n;
+  return MDBX_SUCCESS;
+}
+
+int mdbx_cursor_compare(const MDBX_cursor *l, const MDBX_cursor *r, bool ignore_multival) {
+  const int incomparable = INT16_MAX + 1;
+
+  if (unlikely(!l))
+    return r ? -incomparable * 9 : 0;
+  else if (unlikely(!r))
+    return incomparable * 9;
+
+  if (unlikely(cursor_check_pure(l) != MDBX_SUCCESS))
+    return (cursor_check_pure(r) == MDBX_SUCCESS) ? -incomparable * 8 : 0;
+  if (unlikely(cursor_check_pure(r) != MDBX_SUCCESS))
+    return (cursor_check_pure(l) == MDBX_SUCCESS) ? incomparable * 8 : 0;
+
+  if (unlikely(l->clc != r->clc)) {
+    if (l->txn->env != r->txn->env)
+      return (l->txn->env > r->txn->env) ? incomparable * 7 : -incomparable * 7;
+    if (l->txn->txnid != r->txn->txnid)
+      return (l->txn->txnid > r->txn->txnid) ? incomparable * 6 : -incomparable * 6;
+    return (l->clc > r->clc) ? incomparable * 5 : -incomparable * 5;
+  }
+  assert(cursor_dbi(l) == cursor_dbi(r));
+
+  int diff = is_pointed(l) - is_pointed(r);
+  if (unlikely(diff))
+    return (diff > 0) ? incomparable * 4 : -incomparable * 4;
+  if (unlikely(!is_pointed(l)))
+    return 0;
+
+  intptr_t detent = (l->top <= r->top) ? l->top : r->top;
+  for (intptr_t i = 0; i <= detent; ++i) {
+    diff = l->ki[i] - r->ki[i];
+    if (diff)
+      return diff;
+  }
+  if (unlikely(l->top != r->top))
+    return (l->top > r->top) ? incomparable * 3 : -incomparable * 3;
+
+  assert((l->subcur != nullptr) == (r->subcur != nullptr));
+  if (unlikely((l->subcur != nullptr) != (r->subcur != nullptr)))
+    return l->subcur ? incomparable * 2 : -incomparable * 2;
+  if (ignore_multival || !l->subcur)
+    return 0;
+
+#if MDBX_DEBUG
+  if (is_pointed(&l->subcur->cursor)) {
+    const page_t *mp = l->pg[l->top];
+    const node_t *node = page_node(mp, l->ki[l->top]);
+    assert(node_flags(node) & N_DUP);
+  }
+  if (is_pointed(&r->subcur->cursor)) {
+    const page_t *mp = r->pg[r->top];
+    const node_t *node = page_node(mp, r->ki[r->top]);
+    assert(node_flags(node) & N_DUP);
+  }
+#endif /* MDBX_DEBUG */
+
+  l = &l->subcur->cursor;
+  r = &r->subcur->cursor;
+  diff = is_pointed(l) - is_pointed(r);
+  if (unlikely(diff))
+    return (diff > 0) ? incomparable * 2 : -incomparable * 2;
+  if (unlikely(!is_pointed(l)))
+    return 0;
+
+  detent = (l->top <= r->top) ? l->top : r->top;
+  for (intptr_t i = 0; i <= detent; ++i) {
+    diff = l->ki[i] - r->ki[i];
+    if (diff)
+      return diff;
+  }
+  if (unlikely(l->top != r->top))
+    return (l->top > r->top) ? incomparable : -incomparable;
+
+  return (l->flags & z_eof_hard) - (r->flags & z_eof_hard);
+}
+
+int mdbx_cursor_count_ex(const MDBX_cursor *mc, size_t *count, MDBX_stat *ns, size_t bytes) {
+  int rc = cursor_check_ro(mc);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (ns) {
+    const size_t size_before_modtxnid = offsetof(MDBX_stat, ms_mod_txnid);
+    if (unlikely(bytes != sizeof(MDBX_stat)) && bytes != size_before_modtxnid)
+      return LOG_IFERR(MDBX_EINVAL);
+    memset(ns, 0, sizeof(*ns));
+  }
+
+  size_t nvals = 0;
+  if (is_filled(mc)) {
+    nvals = 1;
+    if (!inner_hollow(mc)) {
+      const page_t *mp = mc->pg[mc->top];
+      const node_t *node = page_node(mp, mc->ki[mc->top]);
+      cASSERT(mc, node_flags(node) & N_DUP);
+      const tree_t *nt = &mc->subcur->nested_tree;
+      nvals = unlikely(nt->items > PTRDIFF_MAX) ? PTRDIFF_MAX : (size_t)nt->items;
+      if (ns) {
+        ns->ms_psize = (unsigned)node_ds(node);
+        if (node_flags(node) & N_TREE) {
+          ns->ms_psize = mc->txn->env->ps;
+          ns->ms_depth = nt->height;
+          ns->ms_branch_pages = nt->branch_pages;
+        }
+        cASSERT(mc, nt->large_pages == 0);
+        ns->ms_leaf_pages = nt->leaf_pages;
+        ns->ms_entries = nt->items;
+        if (likely(bytes >= offsetof(MDBX_stat, ms_mod_txnid) + sizeof(ns->ms_mod_txnid)))
+          ns->ms_mod_txnid = nt->mod_txnid;
+      }
+    }
+  }
+
+  if (likely(count))
+    *count = nvals;
+
+  return MDBX_SUCCESS;
+}
+
+int mdbx_cursor_count(const MDBX_cursor *mc, size_t *count) {
+  if (unlikely(count == nullptr))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  return mdbx_cursor_count_ex(mc, count, nullptr, 0);
+}
+
+int mdbx_cursor_on_first(const MDBX_cursor *mc) {
+  int rc = cursor_check_pure(mc);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  for (intptr_t i = 0; i <= mc->top; ++i) {
+    if (mc->ki[i])
+      return MDBX_RESULT_FALSE;
+  }
+
+  return MDBX_RESULT_TRUE;
+}
+
+int mdbx_cursor_on_first_dup(const MDBX_cursor *mc) {
+  int rc = cursor_check_pure(mc);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (is_filled(mc) && mc->subcur) {
+    mc = &mc->subcur->cursor;
+    for (intptr_t i = 0; i <= mc->top; ++i) {
+      if (mc->ki[i])
+        return MDBX_RESULT_FALSE;
+    }
+  }
+
+  return MDBX_RESULT_TRUE;
+}
+
+int mdbx_cursor_on_last(const MDBX_cursor *mc) {
+  int rc = cursor_check_pure(mc);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  for (intptr_t i = 0; i <= mc->top; ++i) {
+    size_t nkeys = page_numkeys(mc->pg[i]);
+    if (mc->ki[i] < nkeys - 1)
+      return MDBX_RESULT_FALSE;
+  }
+
+  return MDBX_RESULT_TRUE;
+}
+
+int mdbx_cursor_on_last_dup(const MDBX_cursor *mc) {
+  int rc = cursor_check_pure(mc);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (is_filled(mc) && mc->subcur) {
+    mc = &mc->subcur->cursor;
+    for (intptr_t i = 0; i <= mc->top; ++i) {
+      size_t nkeys = page_numkeys(mc->pg[i]);
+      if (mc->ki[i] < nkeys - 1)
+        return MDBX_RESULT_FALSE;
+    }
+  }
+
+  return MDBX_RESULT_TRUE;
+}
+
+int mdbx_cursor_eof(const MDBX_cursor *mc) {
+  int rc = cursor_check_pure(mc);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  return is_eof(mc) ? MDBX_RESULT_TRUE : MDBX_RESULT_FALSE;
+}
+
+int mdbx_cursor_get(MDBX_cursor *mc, MDBX_val *key, MDBX_val *data, MDBX_cursor_op op) {
+  int rc = cursor_check_ro(mc);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  return LOG_IFERR(cursor_ops(mc, key, data, op));
+}
+
+__hot static int scan_confinue(MDBX_cursor *mc, MDBX_predicate_func *predicate, void *context, void *arg, MDBX_val *key,
+                               MDBX_val *value, MDBX_cursor_op turn_op) {
+  int rc;
+  switch (turn_op) {
+  case MDBX_NEXT:
+  case MDBX_NEXT_NODUP:
+    for (;;) {
+      rc = predicate(context, key, value, arg);
+      if (rc != MDBX_RESULT_FALSE)
+        return rc;
+      rc = outer_next(mc, key, value, turn_op);
+      if (unlikely(rc != MDBX_SUCCESS))
+        return (rc == MDBX_NOTFOUND) ? MDBX_RESULT_FALSE : rc;
+    }
+
+  case MDBX_PREV:
+  case MDBX_PREV_NODUP:
+    for (;;) {
+      rc = predicate(context, key, value, arg);
+      if (rc != MDBX_RESULT_FALSE)
+        return rc;
+      rc = outer_prev(mc, key, value, turn_op);
+      if (unlikely(rc != MDBX_SUCCESS))
+        return (rc == MDBX_NOTFOUND) ? MDBX_RESULT_FALSE : rc;
+    }
+
+  case MDBX_NEXT_DUP:
+    if (mc->subcur)
+      for (;;) {
+        rc = predicate(context, key, value, arg);
+        if (rc != MDBX_RESULT_FALSE)
+          return rc;
+        rc = inner_next(&mc->subcur->cursor, value);
+        if (unlikely(rc != MDBX_SUCCESS))
+          return (rc == MDBX_NOTFOUND) ? MDBX_RESULT_FALSE : rc;
+      }
+    return MDBX_NOTFOUND;
+
+  case MDBX_PREV_DUP:
+    if (mc->subcur)
+      for (;;) {
+        rc = predicate(context, key, value, arg);
+        if (rc != MDBX_RESULT_FALSE)
+          return rc;
+        rc = inner_prev(&mc->subcur->cursor, value);
+        if (unlikely(rc != MDBX_SUCCESS))
+          return (rc == MDBX_NOTFOUND) ? MDBX_RESULT_FALSE : rc;
+      }
+    return MDBX_NOTFOUND;
+
+  default:
+    for (;;) {
+      rc = predicate(context, key, value, arg);
+      if (rc != MDBX_RESULT_FALSE)
+        return rc;
+      rc = cursor_ops(mc, key, value, turn_op);
+      if (unlikely(rc != MDBX_SUCCESS))
+        return (rc == MDBX_NOTFOUND) ? MDBX_RESULT_FALSE : rc;
+    }
+  }
+}
+
+int mdbx_cursor_scan(MDBX_cursor *mc, MDBX_predicate_func *predicate, void *context, MDBX_cursor_op start_op,
+                     MDBX_cursor_op turn_op, void *arg) {
+  if (unlikely(!predicate))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  const unsigned valid_start_mask = 1 << MDBX_FIRST | 1 << MDBX_FIRST_DUP | 1 << MDBX_LAST | 1 << MDBX_LAST_DUP |
+                                    1 << MDBX_GET_CURRENT | 1 << MDBX_GET_MULTIPLE;
+  if (unlikely(start_op > 30 || ((1 << start_op) & valid_start_mask) == 0))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  const unsigned valid_turn_mask = 1 << MDBX_NEXT | 1 << MDBX_NEXT_DUP | 1 << MDBX_NEXT_NODUP | 1 << MDBX_PREV |
+                                   1 << MDBX_PREV_DUP | 1 << MDBX_PREV_NODUP | 1 << MDBX_NEXT_MULTIPLE |
+                                   1 << MDBX_PREV_MULTIPLE;
+  if (unlikely(turn_op > 30 || ((1 << turn_op) & valid_turn_mask) == 0))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  MDBX_val key = {nullptr, 0}, value = {nullptr, 0};
+  int rc = mdbx_cursor_get(mc, &key, &value, start_op);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+  return LOG_IFERR(scan_confinue(mc, predicate, context, arg, &key, &value, turn_op));
+}
+
+int mdbx_cursor_scan_from(MDBX_cursor *mc, MDBX_predicate_func *predicate, void *context, MDBX_cursor_op from_op,
+                          MDBX_val *key, MDBX_val *value, MDBX_cursor_op turn_op, void *arg) {
+  if (unlikely(!predicate || !key))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  const unsigned valid_start_mask = 1 << MDBX_GET_BOTH | 1 << MDBX_GET_BOTH_RANGE | 1 << MDBX_SET_KEY |
+                                    1 << MDBX_GET_MULTIPLE | 1 << MDBX_SET_LOWERBOUND | 1 << MDBX_SET_UPPERBOUND;
+  if (unlikely(from_op < MDBX_TO_KEY_LESSER_THAN && ((1 << from_op) & valid_start_mask) == 0))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  const unsigned valid_turn_mask = 1 << MDBX_NEXT | 1 << MDBX_NEXT_DUP | 1 << MDBX_NEXT_NODUP | 1 << MDBX_PREV |
+                                   1 << MDBX_PREV_DUP | 1 << MDBX_PREV_NODUP | 1 << MDBX_NEXT_MULTIPLE |
+                                   1 << MDBX_PREV_MULTIPLE;
+  if (unlikely(turn_op > 30 || ((1 << turn_op) & valid_turn_mask) == 0))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = mdbx_cursor_get(mc, key, value, from_op);
+  if (unlikely(MDBX_IS_ERROR(rc)))
+    return LOG_IFERR(rc);
+
+  cASSERT(mc, key != nullptr);
+  MDBX_val stub;
+  if (!value) {
+    value = &stub;
+    rc = cursor_ops(mc, key, value, MDBX_GET_CURRENT);
+    if (unlikely(rc != MDBX_SUCCESS))
+      return LOG_IFERR(rc);
+  }
+  return LOG_IFERR(scan_confinue(mc, predicate, context, arg, key, value, turn_op));
+}
+
+int mdbx_cursor_get_batch(MDBX_cursor *mc, size_t *count, MDBX_val *pairs, size_t limit, MDBX_cursor_op op) {
+  if (unlikely(!count))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  *count = 0;
+  if (unlikely(limit < 4 || limit > INTPTR_MAX - 2))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = cursor_check_ro(mc);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(mc->subcur))
+    return LOG_IFERR(MDBX_INCOMPATIBLE) /* must be a non-dupsort table */;
+
+  switch (op) {
+  case MDBX_NEXT:
+    if (unlikely(is_eof(mc)))
+      return LOG_IFERR(is_pointed(mc) ? MDBX_NOTFOUND : MDBX_ENODATA);
+    break;
+
+  case MDBX_FIRST:
+    if (!is_filled(mc)) {
+      rc = outer_first(mc, nullptr, nullptr);
+      if (unlikely(rc != MDBX_SUCCESS))
+        return LOG_IFERR(rc);
+    }
+    break;
+
+  default:
+    DEBUG("unhandled/unimplemented cursor operation %u", op);
+    return LOG_IFERR(MDBX_EINVAL);
+  }
+
+  const page_t *mp = mc->pg[mc->top];
+  size_t nkeys = page_numkeys(mp);
+  size_t ki = mc->ki[mc->top];
+  size_t n = 0;
+  while (n + 2 <= limit) {
+    cASSERT(mc, ki < nkeys);
+    if (unlikely(ki >= nkeys))
+      goto sibling;
+
+    const node_t *leaf = page_node(mp, ki);
+    pairs[n] = get_key(leaf);
+    rc = node_read(mc, leaf, &pairs[n + 1], mp);
+    if (unlikely(rc != MDBX_SUCCESS))
+      goto bailout;
+
+    n += 2;
+    if (++ki == nkeys) {
+    sibling:
+      rc = cursor_sibling_right(mc);
+      if (rc != MDBX_SUCCESS) {
+        if (rc == MDBX_NOTFOUND)
+          rc = MDBX_RESULT_TRUE;
+        goto bailout;
+      }
+
+      mp = mc->pg[mc->top];
+      DEBUG("next page is %" PRIaPGNO ", key index %u", mp->pgno, mc->ki[mc->top]);
+      if (!MDBX_DISABLE_VALIDATION && unlikely(!check_leaf_type(mc, mp))) {
+        ERROR("unexpected leaf-page #%" PRIaPGNO " type 0x%x seen by cursor", mp->pgno, mp->flags);
+        rc = MDBX_CORRUPTED;
+        goto bailout;
+      }
+      nkeys = page_numkeys(mp);
+      ki = 0;
+    }
+  }
+  mc->ki[mc->top] = (indx_t)ki;
+
+bailout:
+  *count = n;
+  return LOG_IFERR(rc);
+}
+
+/*----------------------------------------------------------------------------*/
+
+int mdbx_cursor_set_userctx(MDBX_cursor *mc, void *ctx) {
+  int rc = cursor_check(mc, 0);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cursor_couple_t *couple = container_of(mc, cursor_couple_t, outer);
+  couple->userctx = ctx;
+  return MDBX_SUCCESS;
+}
+
+void *mdbx_cursor_get_userctx(const MDBX_cursor *mc) {
+  if (unlikely(!mc))
+    return nullptr;
+
+  if (unlikely(mc->signature != cur_signature_ready4dispose && mc->signature != cur_signature_live))
+    return nullptr;
+
+  cursor_couple_t *couple = container_of(mc, cursor_couple_t, outer);
+  return couple->userctx;
+}
+
+MDBX_txn *mdbx_cursor_txn(const MDBX_cursor *mc) {
+  if (unlikely(!mc || mc->signature != cur_signature_live))
+    return nullptr;
+  MDBX_txn *txn = mc->txn;
+  if (unlikely(!txn || txn->signature != txn_signature || (txn->flags & MDBX_TXN_FINISHED)))
+    return nullptr;
+  return (txn->flags & MDBX_TXN_HAS_CHILD) ? txn->env->txn : txn;
+}
+
+MDBX_dbi mdbx_cursor_dbi(const MDBX_cursor *mc) {
+  if (unlikely(!mc || mc->signature != cur_signature_live))
+    return UINT_MAX;
+  return cursor_dbi(mc);
+}
+
+/*----------------------------------------------------------------------------*/
+
+int mdbx_cursor_put(MDBX_cursor *mc, const MDBX_val *key, MDBX_val *data, MDBX_put_flags_t flags) {
+  if (unlikely(key == nullptr || data == nullptr))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = cursor_check_rw(mc);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(flags & MDBX_MULTIPLE)) {
+    rc = cursor_check_multiple(mc, key, data, flags);
+    if (unlikely(rc != MDBX_SUCCESS))
+      return LOG_IFERR(rc);
+  }
+
+  if (flags & MDBX_RESERVE) {
+    if (unlikely(mc->tree->flags & (MDBX_DUPSORT | MDBX_REVERSEDUP | MDBX_INTEGERDUP | MDBX_DUPFIXED)))
+      return LOG_IFERR(MDBX_INCOMPATIBLE);
+    data->iov_base = nullptr;
+  }
+
+  return LOG_IFERR(cursor_put_checklen(mc, key, data, flags));
+}
+
+int mdbx_cursor_del(MDBX_cursor *mc, MDBX_put_flags_t flags) {
+  int rc = cursor_check_rw(mc);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  return LOG_IFERR(cursor_del(mc, flags));
+}
+
+__cold int mdbx_cursor_ignord(MDBX_cursor *mc) {
+  int rc = cursor_check(mc, 0);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  mc->checking |= z_ignord;
+  if (mc->subcur)
+    mc->subcur->cursor.checking |= z_ignord;
+
+  return MDBX_SUCCESS;
+}

src/api-dbi.c

@@ -0,0 +1,315 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+int mdbx_dbi_open2(MDBX_txn *txn, const MDBX_val *name, MDBX_db_flags_t flags, MDBX_dbi *dbi) {
+  return LOG_IFERR(dbi_open(txn, name, flags, dbi, nullptr, nullptr));
+}
+
+int mdbx_dbi_open_ex2(MDBX_txn *txn, const MDBX_val *name, MDBX_db_flags_t flags, MDBX_dbi *dbi, MDBX_cmp_func *keycmp,
+                      MDBX_cmp_func *datacmp) {
+  return LOG_IFERR(dbi_open(txn, name, flags, dbi, keycmp, datacmp));
+}
+
+static int dbi_open_cstr(MDBX_txn *txn, const char *name_cstr, MDBX_db_flags_t flags, MDBX_dbi *dbi,
+                         MDBX_cmp_func *keycmp, MDBX_cmp_func *datacmp) {
+  MDBX_val thunk, *name;
+  if (name_cstr == MDBX_CHK_MAIN || name_cstr == MDBX_CHK_GC || name_cstr == MDBX_CHK_META)
+    name = (void *)name_cstr;
+  else {
+    thunk.iov_len = strlen(name_cstr);
+    thunk.iov_base = (void *)name_cstr;
+    name = &thunk;
+  }
+  return dbi_open(txn, name, flags, dbi, keycmp, datacmp);
+}
+
+int mdbx_dbi_open(MDBX_txn *txn, const char *name, MDBX_db_flags_t flags, MDBX_dbi *dbi) {
+  return LOG_IFERR(dbi_open_cstr(txn, name, flags, dbi, nullptr, nullptr));
+}
+
+int mdbx_dbi_open_ex(MDBX_txn *txn, const char *name, MDBX_db_flags_t flags, MDBX_dbi *dbi, MDBX_cmp_func *keycmp,
+                     MDBX_cmp_func *datacmp) {
+  return LOG_IFERR(dbi_open_cstr(txn, name, flags, dbi, keycmp, datacmp));
+}
+
+__cold int mdbx_drop(MDBX_txn *txn, MDBX_dbi dbi, bool del) {
+  int rc = check_txn_rw(txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cursor_couple_t cx;
+  rc = cursor_init(&cx.outer, txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (txn->dbs[dbi].height) {
+    cx.outer.next = txn->cursors[dbi];
+    txn->cursors[dbi] = &cx.outer;
+    rc = tree_drop(&cx.outer, dbi == MAIN_DBI || (cx.outer.tree->flags & MDBX_DUPSORT));
+    txn->cursors[dbi] = cx.outer.next;
+    if (unlikely(rc != MDBX_SUCCESS))
+      return LOG_IFERR(rc);
+  }
+
+  /* Invalidate the dropped DB's cursors */
+  for (MDBX_cursor *mc = txn->cursors[dbi]; mc; mc = mc->next)
+    be_poor(mc);
+
+  if (!del || dbi < CORE_DBS) {
+    /* reset the DB record, mark it dirty */
+    txn->dbi_state[dbi] |= DBI_DIRTY;
+    txn->dbs[dbi].height = 0;
+    txn->dbs[dbi].branch_pages = 0;
+    txn->dbs[dbi].leaf_pages = 0;
+    txn->dbs[dbi].large_pages = 0;
+    txn->dbs[dbi].items = 0;
+    txn->dbs[dbi].root = P_INVALID;
+    txn->dbs[dbi].sequence = 0;
+    /* txn->dbs[dbi].mod_txnid = txn->txnid; */
+    txn->flags |= MDBX_TXN_DIRTY;
+    return MDBX_SUCCESS;
+  }
+
+  MDBX_env *const env = txn->env;
+  MDBX_val name = env->kvs[dbi].name;
+  rc = cursor_init(&cx.outer, txn, MAIN_DBI);
+  if (likely(rc == MDBX_SUCCESS)) {
+    rc = cursor_seek(&cx.outer, &name, nullptr, MDBX_SET).err;
+    if (likely(rc == MDBX_SUCCESS)) {
+      cx.outer.next = txn->cursors[MAIN_DBI];
+      txn->cursors[MAIN_DBI] = &cx.outer;
+      rc = cursor_del(&cx.outer, N_TREE);
+      txn->cursors[MAIN_DBI] = cx.outer.next;
+      if (likely(rc == MDBX_SUCCESS)) {
+        tASSERT(txn, txn->dbi_state[MAIN_DBI] & DBI_DIRTY);
+        tASSERT(txn, txn->flags & MDBX_TXN_DIRTY);
+        txn->dbi_state[dbi] = DBI_LINDO | DBI_OLDEN;
+        rc = osal_fastmutex_acquire(&env->dbi_lock);
+        if (likely(rc == MDBX_SUCCESS))
+          return LOG_IFERR(dbi_close_release(env, dbi));
+      }
+    }
+  }
+
+  txn->flags |= MDBX_TXN_ERROR;
+  return LOG_IFERR(rc);
+}
+
+__cold int mdbx_dbi_rename(MDBX_txn *txn, MDBX_dbi dbi, const char *name_cstr) {
+  MDBX_val thunk, *name;
+  if (name_cstr == MDBX_CHK_MAIN || name_cstr == MDBX_CHK_GC || name_cstr == MDBX_CHK_META)
+    name = (void *)name_cstr;
+  else {
+    thunk.iov_len = strlen(name_cstr);
+    thunk.iov_base = (void *)name_cstr;
+    name = &thunk;
+  }
+  return mdbx_dbi_rename2(txn, dbi, name);
+}
+
+__cold int mdbx_dbi_rename2(MDBX_txn *txn, MDBX_dbi dbi, const MDBX_val *new_name) {
+  int rc = check_txn_rw(txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(new_name == MDBX_CHK_MAIN || new_name->iov_base == MDBX_CHK_MAIN || new_name == MDBX_CHK_GC ||
+               new_name->iov_base == MDBX_CHK_GC || new_name == MDBX_CHK_META || new_name->iov_base == MDBX_CHK_META))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  if (unlikely(dbi < CORE_DBS))
+    return LOG_IFERR(MDBX_EINVAL);
+  rc = dbi_check(txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  rc = osal_fastmutex_acquire(&txn->env->dbi_lock);
+  if (likely(rc == MDBX_SUCCESS)) {
+    struct dbi_rename_result pair = dbi_rename_locked(txn, dbi, *new_name);
+    if (pair.defer)
+      pair.defer->next = nullptr;
+    dbi_defer_release(txn->env, pair.defer);
+    rc = pair.err;
+  }
+  return LOG_IFERR(rc);
+}
+
+int mdbx_dbi_close(MDBX_env *env, MDBX_dbi dbi) {
+  int rc = check_env(env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(dbi < CORE_DBS))
+    return (dbi == MAIN_DBI) ? MDBX_SUCCESS : LOG_IFERR(MDBX_BAD_DBI);
+
+  if (unlikely(dbi >= env->max_dbi))
+    return LOG_IFERR(MDBX_BAD_DBI);
+
+  rc = osal_fastmutex_acquire(&env->dbi_lock);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(dbi >= env->n_dbi)) {
+    rc = MDBX_BAD_DBI;
+  bailout:
+    osal_fastmutex_release(&env->dbi_lock);
+    return LOG_IFERR(rc);
+  }
+
+  while (env->basal_txn && (env->dbs_flags[dbi] & DB_VALID) && (env->basal_txn->flags & MDBX_TXN_FINISHED) == 0) {
+    /* LY: Опасный код, так как env->txn может быть изменено в другом потоке.
+     * К сожалению тут нет надежного решения и может быть падение при неверном
+     * использовании API (вызове mdbx_dbi_close конкурентно с завершением
+     * пишущей транзакции).
+     *
+     * Для минимизации вероятности падения сначала проверяем dbi-флаги
+     * в basal_txn, а уже после в env->txn. Таким образом, падение может быть
+     * только при коллизии с завершением вложенной транзакции.
+     *
+     * Альтернативно можно попробовать выполнять обновление/put записи в
+     * mainDb соответствующей таблице закрываемого хендла. Семантически это
+     * верный путь, но проблема в текущем API, в котором исторически dbi-хендл
+     * живет и закрывается вне транзакции. Причем проблема не только в том,
+     * что нет указателя на текущую пишущую транзакцию, а в том что
+     * пользователь точно не ожидает что закрытие хендла приведет к
+     * скрытой/непрозрачной активности внутри транзакции потенциально
+     * выполняемой в другом потоке. Другими словами, проблема может быть
+     * только при неверном использовании API и если пользователь это
+     * допускает, то точно не будет ожидать скрытых действий внутри
+     * транзакции, и поэтому этот путь потенциально более опасен. */
+    const MDBX_txn *const hazard = env->txn;
+    osal_compiler_barrier();
+    if ((dbi_state(env->basal_txn, dbi) & (DBI_LINDO | DBI_DIRTY | DBI_CREAT)) > DBI_LINDO) {
+      rc = MDBX_DANGLING_DBI;
+      goto bailout;
+    }
+    osal_memory_barrier();
+    if (unlikely(hazard != env->txn))
+      continue;
+    if (hazard != env->basal_txn && hazard && (hazard->flags & MDBX_TXN_FINISHED) == 0 &&
+        hazard->signature == txn_signature &&
+        (dbi_state(hazard, dbi) & (DBI_LINDO | DBI_DIRTY | DBI_CREAT)) > DBI_LINDO) {
+      rc = MDBX_DANGLING_DBI;
+      goto bailout;
+    }
+    osal_compiler_barrier();
+    if (likely(hazard == env->txn))
+      break;
+  }
+  rc = dbi_close_release(env, dbi);
+  return LOG_IFERR(rc);
+}
+
+int mdbx_dbi_flags_ex(const MDBX_txn *txn, MDBX_dbi dbi, unsigned *flags, unsigned *state) {
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED - MDBX_TXN_ERROR - MDBX_TXN_PARKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  rc = dbi_check(txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(!flags || !state))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  *flags = txn->dbs[dbi].flags & DB_PERSISTENT_FLAGS;
+  *state = txn->dbi_state[dbi] & (DBI_FRESH | DBI_CREAT | DBI_DIRTY | DBI_STALE);
+  return MDBX_SUCCESS;
+}
+
+static void stat_get(const tree_t *db, MDBX_stat *st, size_t bytes) {
+  st->ms_depth = db->height;
+  st->ms_branch_pages = db->branch_pages;
+  st->ms_leaf_pages = db->leaf_pages;
+  st->ms_overflow_pages = db->large_pages;
+  st->ms_entries = db->items;
+  if (likely(bytes >= offsetof(MDBX_stat, ms_mod_txnid) + sizeof(st->ms_mod_txnid)))
+    st->ms_mod_txnid = db->mod_txnid;
+}
+
+__cold int mdbx_dbi_stat(const MDBX_txn *txn, MDBX_dbi dbi, MDBX_stat *dest, size_t bytes) {
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  rc = dbi_check(txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(txn->flags & MDBX_TXN_BLOCKED))
+    return LOG_IFERR(MDBX_BAD_TXN);
+
+  if (unlikely(txn->dbi_state[dbi] & DBI_STALE)) {
+    rc = tbl_fetch((MDBX_txn *)txn, dbi);
+    if (unlikely(rc != MDBX_SUCCESS))
+      return LOG_IFERR(rc);
+  }
+
+  if (unlikely(!dest))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  const size_t size_before_modtxnid = offsetof(MDBX_stat, ms_mod_txnid);
+  if (unlikely(bytes != sizeof(MDBX_stat)) && bytes != size_before_modtxnid)
+    return LOG_IFERR(MDBX_EINVAL);
+
+  dest->ms_psize = txn->env->ps;
+  stat_get(&txn->dbs[dbi], dest, bytes);
+  return MDBX_SUCCESS;
+}
+
+__cold int mdbx_enumerate_tables(const MDBX_txn *txn, MDBX_table_enum_func *func, void *ctx) {
+  if (unlikely(!func))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cursor_couple_t cx;
+  rc = cursor_init(&cx.outer, txn, MAIN_DBI);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cx.outer.next = txn->cursors[MAIN_DBI];
+  txn->cursors[MAIN_DBI] = &cx.outer;
+  for (rc = outer_first(&cx.outer, nullptr, nullptr); rc == MDBX_SUCCESS;
+       rc = outer_next(&cx.outer, nullptr, nullptr, MDBX_NEXT_NODUP)) {
+    node_t *node = page_node(cx.outer.pg[cx.outer.top], cx.outer.ki[cx.outer.top]);
+    if (node_flags(node) != N_TREE)
+      continue;
+    if (unlikely(node_ds(node) != sizeof(tree_t))) {
+      ERROR("%s/%d: %s %u", "MDBX_CORRUPTED", MDBX_CORRUPTED, "invalid dupsort sub-tree node size",
+            (unsigned)node_ds(node));
+      rc = MDBX_CORRUPTED;
+      break;
+    }
+
+    tree_t reside;
+    const tree_t *tree = memcpy(&reside, node_data(node), sizeof(reside));
+    const MDBX_val name = {node_key(node), node_ks(node)};
+    const MDBX_env *const env = txn->env;
+    MDBX_dbi dbi = 0;
+    for (size_t i = CORE_DBS; i < env->n_dbi; ++i) {
+      if (i >= txn->n_dbi || !(env->dbs_flags[i] & DB_VALID))
+        continue;
+      if (env->kvs[MAIN_DBI].clc.k.cmp(&name, &env->kvs[i].name))
+        continue;
+
+      tree = dbi_dig(txn, i, &reside);
+      dbi = (MDBX_dbi)i;
+      break;
+    }
+
+    MDBX_stat stat;
+    stat_get(tree, &stat, sizeof(stat));
+    rc = func(ctx, txn, &name, tree->flags, &stat, dbi);
+    if (rc != MDBX_SUCCESS)
+      goto bailout;
+  }
+  rc = (rc == MDBX_NOTFOUND) ? MDBX_SUCCESS : rc;
+
+bailout:
+  txn->cursors[MAIN_DBI] = cx.outer.next;
+  return LOG_IFERR(rc);
+}

src/api-extra.c

@@ -0,0 +1,165 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+/*------------------------------------------------------------------------------
+ * Readers API */
+
+__cold int mdbx_reader_list(const MDBX_env *env, MDBX_reader_list_func *func, void *ctx) {
+  int rc = check_env(env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(!func))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  rc = MDBX_RESULT_TRUE;
+  int serial = 0;
+  lck_t *const lck = env->lck_mmap.lck;
+  if (likely(lck)) {
+    const size_t snap_nreaders = atomic_load32(&lck->rdt_length, mo_AcquireRelease);
+    for (size_t i = 0; i < snap_nreaders; i++) {
+      const reader_slot_t *r = lck->rdt + i;
+    retry_reader:;
+      const uint32_t pid = atomic_load32(&r->pid, mo_AcquireRelease);
+      if (!pid)
+        continue;
+      txnid_t txnid = safe64_read(&r->txnid);
+      const uint64_t tid = atomic_load64(&r->tid, mo_Relaxed);
+      const pgno_t pages_used = atomic_load32(&r->snapshot_pages_used, mo_Relaxed);
+      const uint64_t reader_pages_retired = atomic_load64(&r->snapshot_pages_retired, mo_Relaxed);
+      if (unlikely(txnid != safe64_read(&r->txnid) || pid != atomic_load32(&r->pid, mo_AcquireRelease) ||
+                   tid != atomic_load64(&r->tid, mo_Relaxed) ||
+                   pages_used != atomic_load32(&r->snapshot_pages_used, mo_Relaxed) ||
+                   reader_pages_retired != atomic_load64(&r->snapshot_pages_retired, mo_Relaxed)))
+        goto retry_reader;
+
+      eASSERT(env, txnid > 0);
+      if (txnid >= SAFE64_INVALID_THRESHOLD)
+        txnid = 0;
+
+      size_t bytes_used = 0;
+      size_t bytes_retained = 0;
+      uint64_t lag = 0;
+      if (txnid) {
+        troika_t troika = meta_tap(env);
+      retry_header:;
+        const meta_ptr_t head = meta_recent(env, &troika);
+        const uint64_t head_pages_retired = unaligned_peek_u64_volatile(4, head.ptr_v->pages_retired);
+        if (unlikely(meta_should_retry(env, &troika) ||
+                     head_pages_retired != unaligned_peek_u64_volatile(4, head.ptr_v->pages_retired)))
+          goto retry_header;
+
+        lag = (head.txnid - txnid) / xMDBX_TXNID_STEP;
+        bytes_used = pgno2bytes(env, pages_used);
+        bytes_retained = (head_pages_retired > reader_pages_retired)
+                             ? pgno2bytes(env, (pgno_t)(head_pages_retired - reader_pages_retired))
+                             : 0;
+      }
+      rc = func(ctx, ++serial, (unsigned)i, pid, (mdbx_tid_t)((intptr_t)tid), txnid, lag, bytes_used, bytes_retained);
+      if (unlikely(rc != MDBX_SUCCESS))
+        break;
+    }
+  }
+
+  return LOG_IFERR(rc);
+}
+
+__cold int mdbx_reader_check(MDBX_env *env, int *dead) {
+  if (dead)
+    *dead = 0;
+  return LOG_IFERR(mvcc_cleanup_dead(env, false, dead));
+}
+
+__cold int mdbx_thread_register(const MDBX_env *env) {
+  int rc = check_env(env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(!env->lck_mmap.lck))
+    return LOG_IFERR((env->flags & MDBX_EXCLUSIVE) ? MDBX_EINVAL : MDBX_EPERM);
+
+  if (unlikely((env->flags & ENV_TXKEY) == 0)) {
+    eASSERT(env, env->flags & MDBX_NOSTICKYTHREADS);
+    return LOG_IFERR(MDBX_EINVAL) /* MDBX_NOSTICKYTHREADS mode */;
+  }
+
+  eASSERT(env, (env->flags & (MDBX_NOSTICKYTHREADS | ENV_TXKEY)) == ENV_TXKEY);
+  reader_slot_t *r = thread_rthc_get(env->me_txkey);
+  if (unlikely(r != nullptr)) {
+    eASSERT(env, r->pid.weak == env->pid);
+    eASSERT(env, r->tid.weak == osal_thread_self());
+    if (unlikely(r->pid.weak != env->pid))
+      return LOG_IFERR(MDBX_BAD_RSLOT);
+    return MDBX_RESULT_TRUE /* already registered */;
+  }
+
+  return LOG_IFERR(mvcc_bind_slot((MDBX_env *)env).err);
+}
+
+__cold int mdbx_thread_unregister(const MDBX_env *env) {
+  int rc = check_env(env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(!env->lck_mmap.lck))
+    return MDBX_RESULT_TRUE;
+
+  if (unlikely((env->flags & ENV_TXKEY) == 0)) {
+    eASSERT(env, env->flags & MDBX_NOSTICKYTHREADS);
+    return MDBX_RESULT_TRUE /* MDBX_NOSTICKYTHREADS mode */;
+  }
+
+  eASSERT(env, (env->flags & (MDBX_NOSTICKYTHREADS | ENV_TXKEY)) == ENV_TXKEY);
+  reader_slot_t *r = thread_rthc_get(env->me_txkey);
+  if (unlikely(r == nullptr))
+    return MDBX_RESULT_TRUE /* not registered */;
+
+  eASSERT(env, r->pid.weak == env->pid);
+  if (unlikely(r->pid.weak != env->pid || r->tid.weak != osal_thread_self()))
+    return LOG_IFERR(MDBX_BAD_RSLOT);
+
+  eASSERT(env, r->txnid.weak >= SAFE64_INVALID_THRESHOLD);
+  if (unlikely(r->txnid.weak < SAFE64_INVALID_THRESHOLD))
+    return LOG_IFERR(MDBX_BUSY) /* transaction is still active */;
+
+  atomic_store32(&r->pid, 0, mo_Relaxed);
+  atomic_store32(&env->lck->rdt_refresh_flag, true, mo_AcquireRelease);
+  thread_rthc_set(env->me_txkey, nullptr);
+  return MDBX_SUCCESS;
+}
+
+/*------------------------------------------------------------------------------
+ * Locking API */
+
+int mdbx_txn_lock(MDBX_env *env, bool dont_wait) {
+  int rc = check_env(env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(env->flags & MDBX_RDONLY))
+    return LOG_IFERR(MDBX_EACCESS);
+  if (dont_wait && unlikely(env->basal_txn->owner || (env->basal_txn->flags & MDBX_TXN_FINISHED) == 0))
+    return LOG_IFERR(MDBX_BUSY);
+
+  return LOG_IFERR(lck_txn_lock(env, dont_wait));
+}
+
+int mdbx_txn_unlock(MDBX_env *env) {
+  int rc = check_env(env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(env->flags & MDBX_RDONLY))
+    return LOG_IFERR(MDBX_EACCESS);
+#if MDBX_TXN_CHECKOWNER
+  if (unlikely(env->basal_txn->owner != osal_thread_self()))
+    return LOG_IFERR(MDBX_THREAD_MISMATCH);
+#endif /* MDBX_TXN_CHECKOWNER */
+  if (unlikely((env->basal_txn->flags & MDBX_TXN_FINISHED) == 0))
+    return LOG_IFERR(MDBX_BUSY);
+
+  lck_txn_unlock(env);
+  return MDBX_SUCCESS;
+}

src/api-key-transform.c

@@ -0,0 +1,197 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+static inline double key2double(const int64_t key) {
+  union {
+    uint64_t u;
+    double f;
+  } casting;
+
+  casting.u = (key < 0) ? key + UINT64_C(0x8000000000000000) : UINT64_C(0xffffFFFFffffFFFF) - key;
+  return casting.f;
+}
+
+static inline uint64_t double2key(const double *const ptr) {
+  STATIC_ASSERT(sizeof(double) == sizeof(int64_t));
+  const int64_t i = *(const int64_t *)ptr;
+  const uint64_t u = (i < 0) ? UINT64_C(0xffffFFFFffffFFFF) - i : i + UINT64_C(0x8000000000000000);
+  if (ASSERT_ENABLED()) {
+    const double f = key2double(u);
+    assert(memcmp(&f, ptr, sizeof(double)) == 0);
+  }
+  return u;
+}
+
+static inline float key2float(const int32_t key) {
+  union {
+    uint32_t u;
+    float f;
+  } casting;
+
+  casting.u = (key < 0) ? key + UINT32_C(0x80000000) : UINT32_C(0xffffFFFF) - key;
+  return casting.f;
+}
+
+static inline uint32_t float2key(const float *const ptr) {
+  STATIC_ASSERT(sizeof(float) == sizeof(int32_t));
+  const int32_t i = *(const int32_t *)ptr;
+  const uint32_t u = (i < 0) ? UINT32_C(0xffffFFFF) - i : i + UINT32_C(0x80000000);
+  if (ASSERT_ENABLED()) {
+    const float f = key2float(u);
+    assert(memcmp(&f, ptr, sizeof(float)) == 0);
+  }
+  return u;
+}
+
+uint64_t mdbx_key_from_double(const double ieee754_64bit) { return double2key(&ieee754_64bit); }
+
+uint64_t mdbx_key_from_ptrdouble(const double *const ieee754_64bit) { return double2key(ieee754_64bit); }
+
+uint32_t mdbx_key_from_float(const float ieee754_32bit) { return float2key(&ieee754_32bit); }
+
+uint32_t mdbx_key_from_ptrfloat(const float *const ieee754_32bit) { return float2key(ieee754_32bit); }
+
+#define IEEE754_DOUBLE_MANTISSA_SIZE 52
+#define IEEE754_DOUBLE_EXPONENTA_BIAS 0x3FF
+#define IEEE754_DOUBLE_EXPONENTA_MAX 0x7FF
+#define IEEE754_DOUBLE_IMPLICIT_LEAD UINT64_C(0x0010000000000000)
+#define IEEE754_DOUBLE_MANTISSA_MASK UINT64_C(0x000FFFFFFFFFFFFF)
+#define IEEE754_DOUBLE_MANTISSA_AMAX UINT64_C(0x001FFFFFFFFFFFFF)
+
+static inline int clz64(uint64_t value) {
+#if __GNUC_PREREQ(4, 1) || __has_builtin(__builtin_clzl)
+  if (sizeof(value) == sizeof(int))
+    return __builtin_clz(value);
+  if (sizeof(value) == sizeof(long))
+    return __builtin_clzl(value);
+#if (defined(__SIZEOF_LONG_LONG__) && __SIZEOF_LONG_LONG__ == 8) || __has_builtin(__builtin_clzll)
+  return __builtin_clzll(value);
+#endif /* have(long long) && long long == uint64_t */
+#endif /* GNU C */
+
+#if defined(_MSC_VER)
+  unsigned long index;
+#if defined(_M_AMD64) || defined(_M_ARM64) || defined(_M_X64)
+  _BitScanReverse64(&index, value);
+  return 63 - index;
+#else
+  if (value > UINT32_MAX) {
+    _BitScanReverse(&index, (uint32_t)(value >> 32));
+    return 31 - index;
+  }
+  _BitScanReverse(&index, (uint32_t)value);
+  return 63 - index;
+#endif
+#endif /* MSVC */
+
+  value |= value >> 1;
+  value |= value >> 2;
+  value |= value >> 4;
+  value |= value >> 8;
+  value |= value >> 16;
+  value |= value >> 32;
+  static const uint8_t debruijn_clz64[64] = {63, 16, 62, 7,  15, 36, 61, 3,  6,  14, 22, 26, 35, 47, 60, 2,
+                                             9,  5,  28, 11, 13, 21, 42, 19, 25, 31, 34, 40, 46, 52, 59, 1,
+                                             17, 8,  37, 4,  23, 27, 48, 10, 29, 12, 43, 20, 32, 41, 53, 18,
+                                             38, 24, 49, 30, 44, 33, 54, 39, 50, 45, 55, 51, 56, 57, 58, 0};
+  return debruijn_clz64[value * UINT64_C(0x03F79D71B4CB0A89) >> 58];
+}
+
+static inline uint64_t round_mantissa(const uint64_t u64, int shift) {
+  assert(shift < 0 && u64 > 0);
+  shift = -shift;
+  const unsigned half = 1 << (shift - 1);
+  const unsigned lsb = 1 & (unsigned)(u64 >> shift);
+  const unsigned tie2even = 1 ^ lsb;
+  return (u64 + half - tie2even) >> shift;
+}
+
+uint64_t mdbx_key_from_jsonInteger(const int64_t json_integer) {
+  const uint64_t bias = UINT64_C(0x8000000000000000);
+  if (json_integer > 0) {
+    const uint64_t u64 = json_integer;
+    int shift = clz64(u64) - (64 - IEEE754_DOUBLE_MANTISSA_SIZE - 1);
+    uint64_t mantissa = u64 << shift;
+    if (unlikely(shift < 0)) {
+      mantissa = round_mantissa(u64, shift);
+      if (mantissa > IEEE754_DOUBLE_MANTISSA_AMAX)
+        mantissa = round_mantissa(u64, --shift);
+    }
+
+    assert(mantissa >= IEEE754_DOUBLE_IMPLICIT_LEAD && mantissa <= IEEE754_DOUBLE_MANTISSA_AMAX);
+    const uint64_t exponent = (uint64_t)IEEE754_DOUBLE_EXPONENTA_BIAS + IEEE754_DOUBLE_MANTISSA_SIZE - shift;
+    assert(exponent > 0 && exponent <= IEEE754_DOUBLE_EXPONENTA_MAX);
+    const uint64_t key = bias + (exponent << IEEE754_DOUBLE_MANTISSA_SIZE) + (mantissa - IEEE754_DOUBLE_IMPLICIT_LEAD);
+#if !defined(_MSC_VER) || defined(_DEBUG) /* Workaround for MSVC error LNK2019: unresolved external                    \
+                                             symbol __except1 referenced in function __ftol3_except */
+    assert(key == mdbx_key_from_double((double)json_integer));
+#endif /* Workaround for MSVC */
+    return key;
+  }
+
+  if (json_integer < 0) {
+    const uint64_t u64 = -json_integer;
+    int shift = clz64(u64) - (64 - IEEE754_DOUBLE_MANTISSA_SIZE - 1);
+    uint64_t mantissa = u64 << shift;
+    if (unlikely(shift < 0)) {
+      mantissa = round_mantissa(u64, shift);
+      if (mantissa > IEEE754_DOUBLE_MANTISSA_AMAX)
+        mantissa = round_mantissa(u64, --shift);
+    }
+
+    assert(mantissa >= IEEE754_DOUBLE_IMPLICIT_LEAD && mantissa <= IEEE754_DOUBLE_MANTISSA_AMAX);
+    const uint64_t exponent = (uint64_t)IEEE754_DOUBLE_EXPONENTA_BIAS + IEEE754_DOUBLE_MANTISSA_SIZE - shift;
+    assert(exponent > 0 && exponent <= IEEE754_DOUBLE_EXPONENTA_MAX);
+    const uint64_t key =
+        bias - 1 - (exponent << IEEE754_DOUBLE_MANTISSA_SIZE) - (mantissa - IEEE754_DOUBLE_IMPLICIT_LEAD);
+#if !defined(_MSC_VER) || defined(_DEBUG) /* Workaround for MSVC error LNK2019: unresolved external                    \
+                                             symbol __except1 referenced in function __ftol3_except */
+    assert(key == mdbx_key_from_double((double)json_integer));
+#endif /* Workaround for MSVC */
+    return key;
+  }
+
+  return bias;
+}
+
+int64_t mdbx_jsonInteger_from_key(const MDBX_val v) {
+  assert(v.iov_len == 8);
+  const uint64_t key = unaligned_peek_u64(2, v.iov_base);
+  const uint64_t bias = UINT64_C(0x8000000000000000);
+  const uint64_t covalent = (key > bias) ? key - bias : bias - key - 1;
+  const int shift = IEEE754_DOUBLE_EXPONENTA_BIAS + 63 -
+                    (IEEE754_DOUBLE_EXPONENTA_MAX & (int)(covalent >> IEEE754_DOUBLE_MANTISSA_SIZE));
+  if (unlikely(shift < 1))
+    return (key < bias) ? INT64_MIN : INT64_MAX;
+  if (unlikely(shift > 63))
+    return 0;
+
+  const uint64_t unscaled = ((covalent & IEEE754_DOUBLE_MANTISSA_MASK) << (63 - IEEE754_DOUBLE_MANTISSA_SIZE)) + bias;
+  const int64_t absolute = unscaled >> shift;
+  const int64_t value = (key < bias) ? -absolute : absolute;
+  assert(key == mdbx_key_from_jsonInteger(value) ||
+         (mdbx_key_from_jsonInteger(value - 1) < key && key < mdbx_key_from_jsonInteger(value + 1)));
+  return value;
+}
+
+double mdbx_double_from_key(const MDBX_val v) {
+  assert(v.iov_len == 8);
+  return key2double(unaligned_peek_u64(2, v.iov_base));
+}
+
+float mdbx_float_from_key(const MDBX_val v) {
+  assert(v.iov_len == 4);
+  return key2float(unaligned_peek_u32(2, v.iov_base));
+}
+
+int32_t mdbx_int32_from_key(const MDBX_val v) {
+  assert(v.iov_len == 4);
+  return (int32_t)(unaligned_peek_u32(2, v.iov_base) - UINT32_C(0x80000000));
+}
+
+int64_t mdbx_int64_from_key(const MDBX_val v) {
+  assert(v.iov_len == 8);
+  return (int64_t)(unaligned_peek_u64(2, v.iov_base) - UINT64_C(0x8000000000000000));
+}

src/api-misc.c

@@ -0,0 +1,286 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+__cold int mdbx_is_readahead_reasonable(size_t volume, intptr_t redundancy) {
+  if (volume <= 1024 * 1024 * 4ul)
+    return MDBX_RESULT_TRUE;
+
+  intptr_t pagesize, total_ram_pages;
+  int err = mdbx_get_sysraminfo(&pagesize, &total_ram_pages, nullptr);
+  if (unlikely(err != MDBX_SUCCESS))
+    return LOG_IFERR(err);
+
+  const int log2page = log2n_powerof2(pagesize);
+  const intptr_t volume_pages = (volume + pagesize - 1) >> log2page;
+  const intptr_t redundancy_pages = (redundancy < 0) ? -(intptr_t)((-redundancy + pagesize - 1) >> log2page)
+                                                     : (intptr_t)(redundancy + pagesize - 1) >> log2page;
+  if (volume_pages >= total_ram_pages || volume_pages + redundancy_pages >= total_ram_pages)
+    return MDBX_RESULT_FALSE;
+
+  intptr_t avail_ram_pages;
+  err = mdbx_get_sysraminfo(nullptr, nullptr, &avail_ram_pages);
+  if (unlikely(err != MDBX_SUCCESS))
+    return LOG_IFERR(err);
+
+  return (volume_pages + redundancy_pages >= avail_ram_pages) ? MDBX_RESULT_FALSE : MDBX_RESULT_TRUE;
+}
+
+int mdbx_dbi_sequence(MDBX_txn *txn, MDBX_dbi dbi, uint64_t *result, uint64_t increment) {
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  rc = dbi_check(txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(txn->dbi_state[dbi] & DBI_STALE)) {
+    rc = tbl_fetch(txn, dbi);
+    if (unlikely(rc != MDBX_SUCCESS))
+      return LOG_IFERR(rc);
+  }
+
+  tree_t *dbs = &txn->dbs[dbi];
+  if (likely(result))
+    *result = dbs->sequence;
+
+  if (likely(increment > 0)) {
+    if (unlikely(dbi == FREE_DBI || (txn->flags & MDBX_TXN_RDONLY) != 0))
+      return MDBX_EACCESS;
+
+    uint64_t new = dbs->sequence + increment;
+    if (unlikely(new < increment))
+      return MDBX_RESULT_TRUE;
+
+    tASSERT(txn, new > dbs->sequence);
+    if ((txn->dbi_state[dbi] & DBI_DIRTY) == 0) {
+      txn->flags |= MDBX_TXN_DIRTY;
+      txn->dbi_state[dbi] |= DBI_DIRTY;
+      if (unlikely(dbi == MAIN_DBI) && txn->dbs[MAIN_DBI].root != P_INVALID) {
+        /* LY: Временная подпорка для coherency_check(), которую в перспективе
+         * следует заменить вместе с переделкой установки mod_txnid.
+         *
+         * Суть проблемы:
+         *  - coherency_check() в качестве одного из критериев "когерентности"
+         *    проверяет условие meta.maindb.mod_txnid == maindb.root->txnid;
+         *  - при обновлении maindb.sequence высталяется DBI_DIRTY, что приведет
+         *    к обновлению meta.maindb.mod_txnid = current_txnid;
+         *  - однако, если в само дерево maindb обновление не вносились и оно
+         *    не пустое, то корневая страницы останеться с прежним txnid и из-за
+         *    этого ложно сработает coherency_check().
+         *
+         * Временное (текущее) решение: Принудительно обновляем корневую
+         * страницу в описанной выше ситуации. Это устраняет проблему, но и
+         * не создает рисков регресса.
+         *
+         * FIXME: Итоговое решение, которое предстоит реализовать:
+         *  - изменить семантику установки/обновления mod_txnid, привязав его
+         *    строго к изменению b-tree, но не атрибутов;
+         *  - обновлять mod_txnid при фиксации вложенных транзакций;
+         *  - для dbi-хендлов пользовательских table (видимо) можно оставить
+         *    DBI_DIRTY в качестве признака необходимости обновления записи
+         *    table в MainDB, при этом взводить DBI_DIRTY вместе с обновлением
+         *    mod_txnid, в том числе при обновлении sequence.
+         *  - для MAIN_DBI при обновлении sequence не следует взводить DBI_DIRTY
+         *    и/или обновлять mod_txnid, а только взводить MDBX_TXN_DIRTY.
+         *  - альтернативно, можно перераспределить флажки-признаки dbi_state,
+         *    чтобы различать состояние dirty-tree и dirty-attributes. */
+        cursor_couple_t cx;
+        rc = cursor_init(&cx.outer, txn, MAIN_DBI);
+        if (unlikely(rc != MDBX_SUCCESS))
+          return LOG_IFERR(rc);
+        rc = tree_search(&cx.outer, nullptr, Z_MODIFY | Z_ROOTONLY);
+        if (unlikely(rc != MDBX_SUCCESS))
+          return LOG_IFERR(rc);
+      }
+    }
+    dbs->sequence = new;
+  }
+
+  return MDBX_SUCCESS;
+}
+
+int mdbx_cmp(const MDBX_txn *txn, MDBX_dbi dbi, const MDBX_val *a, const MDBX_val *b) {
+  eASSERT(nullptr, txn->signature == txn_signature);
+  tASSERT(txn, (dbi_state(txn, dbi) & DBI_VALID) && !dbi_changed(txn, dbi));
+  tASSERT(txn, dbi < txn->env->n_dbi && (txn->env->dbs_flags[dbi] & DB_VALID) != 0);
+  return txn->env->kvs[dbi].clc.k.cmp(a, b);
+}
+
+int mdbx_dcmp(const MDBX_txn *txn, MDBX_dbi dbi, const MDBX_val *a, const MDBX_val *b) {
+  eASSERT(nullptr, txn->signature == txn_signature);
+  tASSERT(txn, (dbi_state(txn, dbi) & DBI_VALID) && !dbi_changed(txn, dbi));
+  tASSERT(txn, dbi < txn->env->n_dbi && (txn->env->dbs_flags[dbi] & DB_VALID));
+  return txn->env->kvs[dbi].clc.v.cmp(a, b);
+}
+
+__cold MDBX_cmp_func *mdbx_get_keycmp(MDBX_db_flags_t flags) { return builtin_keycmp(flags); }
+
+__cold MDBX_cmp_func *mdbx_get_datacmp(MDBX_db_flags_t flags) { return builtin_datacmp(flags); }
+
+/*----------------------------------------------------------------------------*/
+
+__cold const char *mdbx_liberr2str(int errnum) {
+  /* Table of descriptions for MDBX errors */
+  static const char *const tbl[] = {
+      "MDBX_KEYEXIST: Key/data pair already exists",
+      "MDBX_NOTFOUND: No matching key/data pair found",
+      "MDBX_PAGE_NOTFOUND: Requested page not found",
+      "MDBX_CORRUPTED: Database is corrupted",
+      "MDBX_PANIC: Environment had fatal error",
+      "MDBX_VERSION_MISMATCH: DB version mismatch libmdbx",
+      "MDBX_INVALID: File is not an MDBX file",
+      "MDBX_MAP_FULL: Environment mapsize limit reached",
+      "MDBX_DBS_FULL: Too many DBI-handles (maxdbs reached)",
+      "MDBX_READERS_FULL: Too many readers (maxreaders reached)",
+      nullptr /* MDBX_TLS_FULL (-30789): unused in MDBX */,
+      "MDBX_TXN_FULL: Transaction has too many dirty pages,"
+      " i.e transaction is too big",
+      "MDBX_CURSOR_FULL: Cursor stack limit reachedn - this usually indicates"
+      " corruption, i.e branch-pages loop",
+      "MDBX_PAGE_FULL: Internal error - Page has no more space",
+      "MDBX_UNABLE_EXTEND_MAPSIZE: Database engine was unable to extend"
+      " mapping, e.g. since address space is unavailable or busy,"
+      " or Operation system not supported such operations",
+      "MDBX_INCOMPATIBLE: Environment or database is not compatible"
+      " with the requested operation or the specified flags",
+      "MDBX_BAD_RSLOT: Invalid reuse of reader locktable slot,"
+      " e.g. read-transaction already run for current thread",
+      "MDBX_BAD_TXN: Transaction is not valid for requested operation,"
+      " e.g. had errored and be must aborted, has a child, or is invalid",
+      "MDBX_BAD_VALSIZE: Invalid size or alignment of key or data"
+      " for target database, either invalid table name",
+      "MDBX_BAD_DBI: The specified DBI-handle is invalid"
+      " or changed by another thread/transaction",
+      "MDBX_PROBLEM: Unexpected internal error, transaction should be aborted",
+      "MDBX_BUSY: Another write transaction is running,"
+      " or environment is already used while opening with MDBX_EXCLUSIVE flag",
+  };
+
+  if (errnum >= MDBX_KEYEXIST && errnum <= MDBX_BUSY) {
+    int i = errnum - MDBX_KEYEXIST;
+    return tbl[i];
+  }
+
+  switch (errnum) {
+  case MDBX_SUCCESS:
+    return "MDBX_SUCCESS: Successful";
+  case MDBX_EMULTIVAL:
+    return "MDBX_EMULTIVAL: The specified key has"
+           " more than one associated value";
+  case MDBX_EBADSIGN:
+    return "MDBX_EBADSIGN: Wrong signature of a runtime object(s),"
+           " e.g. memory corruption or double-free";
+  case MDBX_WANNA_RECOVERY:
+    return "MDBX_WANNA_RECOVERY: Database should be recovered,"
+           " but this could NOT be done automatically for now"
+           " since it opened in read-only mode";
+  case MDBX_EKEYMISMATCH:
+    return "MDBX_EKEYMISMATCH: The given key value is mismatched to the"
+           " current cursor position";
+  case MDBX_TOO_LARGE:
+    return "MDBX_TOO_LARGE: Database is too large for current system,"
+           " e.g. could NOT be mapped into RAM";
+  case MDBX_THREAD_MISMATCH:
+    return "MDBX_THREAD_MISMATCH: A thread has attempted to use a not"
+           " owned object, e.g. a transaction that started by another thread";
+  case MDBX_TXN_OVERLAPPING:
+    return "MDBX_TXN_OVERLAPPING: Overlapping read and write transactions for"
+           " the current thread";
+  case MDBX_DUPLICATED_CLK:
+    return "MDBX_DUPLICATED_CLK: Alternative/Duplicate LCK-file is exists,"
+           " please keep one and remove unused other";
+  case MDBX_DANGLING_DBI:
+    return "MDBX_DANGLING_DBI: Some cursors and/or other resources should be"
+           " closed before table or corresponding DBI-handle could be (re)used";
+  case MDBX_OUSTED:
+    return "MDBX_OUSTED: The parked read transaction was outed for the sake"
+           " of recycling old MVCC snapshots";
+  case MDBX_MVCC_RETARDED:
+    return "MDBX_MVCC_RETARDED: MVCC snapshot used by parked transaction was bygone";
+  default:
+    return nullptr;
+  }
+}
+
+__cold const char *mdbx_strerror_r(int errnum, char *buf, size_t buflen) {
+  const char *msg = mdbx_liberr2str(errnum);
+  if (!msg && buflen > 0 && buflen < INT_MAX) {
+#if defined(_WIN32) || defined(_WIN64)
+    DWORD size = FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_IGNORE_INSERTS, nullptr, errnum,
+                                MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT), buf, (DWORD)buflen, nullptr);
+    while (size && buf[size - 1] <= ' ')
+      --size;
+    buf[size] = 0;
+    return size ? buf : "FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM) failed";
+#elif defined(_GNU_SOURCE) && defined(__GLIBC__)
+    /* GNU-specific */
+    if (errnum > 0)
+      msg = strerror_r(errnum, buf, buflen);
+#elif (_POSIX_C_SOURCE >= 200112L || _XOPEN_SOURCE >= 600)
+    /* XSI-compliant */
+    if (errnum > 0 && strerror_r(errnum, buf, buflen) == 0)
+      msg = buf;
+#else
+    if (errnum > 0) {
+      msg = strerror(errnum);
+      if (msg) {
+        strncpy(buf, msg, buflen);
+        msg = buf;
+      }
+    }
+#endif
+    if (!msg) {
+      (void)snprintf(buf, buflen, "error %d", errnum);
+      msg = buf;
+    }
+    buf[buflen - 1] = '\0';
+  }
+  return msg;
+}
+
+__cold const char *mdbx_strerror(int errnum) {
+#if defined(_WIN32) || defined(_WIN64)
+  static char buf[1024];
+  return mdbx_strerror_r(errnum, buf, sizeof(buf));
+#else
+  const char *msg = mdbx_liberr2str(errnum);
+  if (!msg) {
+    if (errnum > 0)
+      msg = strerror(errnum);
+    if (!msg) {
+      static char buf[32];
+      (void)snprintf(buf, sizeof(buf) - 1, "error %d", errnum);
+      msg = buf;
+    }
+  }
+  return msg;
+#endif
+}
+
+#if defined(_WIN32) || defined(_WIN64) /* Bit of madness for Windows */
+const char *mdbx_strerror_r_ANSI2OEM(int errnum, char *buf, size_t buflen) {
+  const char *msg = mdbx_liberr2str(errnum);
+  if (!msg && buflen > 0 && buflen < INT_MAX) {
+    DWORD size = FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM | FORMAT_MESSAGE_IGNORE_INSERTS, nullptr, errnum,
+                                MAKELANGID(LANG_NEUTRAL, SUBLANG_DEFAULT), buf, (DWORD)buflen, nullptr);
+    while (size && buf[size - 1] <= ' ')
+      --size;
+    buf[size] = 0;
+    if (!size)
+      msg = "FormatMessageA(FORMAT_MESSAGE_FROM_SYSTEM) failed";
+    else if (!CharToOemBuffA(buf, buf, size))
+      msg = "CharToOemBuffA() failed";
+    else
+      msg = buf;
+  }
+  return msg;
+}
+
+const char *mdbx_strerror_ANSI2OEM(int errnum) {
+  static char buf[1024];
+  return mdbx_strerror_r_ANSI2OEM(errnum, buf, sizeof(buf));
+}
+#endif /* Bit of madness for Windows */

src/api-opts.c

@@ -0,0 +1,578 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+static pgno_t env_max_pgno(const MDBX_env *env) {
+  return env->ps ? bytes2pgno(env, env->geo_in_bytes.upper ? env->geo_in_bytes.upper : MAX_MAPSIZE) : PAGELIST_LIMIT;
+}
+
+__cold pgno_t default_dp_limit(const MDBX_env *env) {
+  /* auto-setup dp_limit by "The42" ;-) */
+  intptr_t total_ram_pages, avail_ram_pages;
+  int err = mdbx_get_sysraminfo(nullptr, &total_ram_pages, &avail_ram_pages);
+  pgno_t dp_limit = 1024;
+  if (unlikely(err != MDBX_SUCCESS))
+    ERROR("mdbx_get_sysraminfo(), rc %d", err);
+  else {
+    size_t estimate = (size_t)(total_ram_pages + avail_ram_pages) / 42;
+    if (env->ps) {
+      if (env->ps > globals.sys_pagesize)
+        estimate /= env->ps / globals.sys_pagesize;
+      else if (env->ps < globals.sys_pagesize)
+        estimate *= globals.sys_pagesize / env->ps;
+    }
+    dp_limit = (pgno_t)estimate;
+  }
+
+  dp_limit = (dp_limit < PAGELIST_LIMIT) ? dp_limit : PAGELIST_LIMIT;
+  const pgno_t max_pgno = env_max_pgno(env);
+  if (dp_limit > max_pgno - NUM_METAS)
+    dp_limit = max_pgno - NUM_METAS;
+  dp_limit = (dp_limit > CURSOR_STACK_SIZE * 4) ? dp_limit : CURSOR_STACK_SIZE * 4;
+  return dp_limit;
+}
+
+__cold static pgno_t default_rp_augment_limit(const MDBX_env *env) {
+  const size_t timeframe = /* 16 секунд */ 16 << 16;
+  const size_t remain_1sec =
+      (env->options.gc_time_limit < timeframe) ? timeframe - (size_t)env->options.gc_time_limit : 0;
+  const size_t minimum = (env->maxgc_large1page * 2 > MDBX_PNL_INITIAL) ? env->maxgc_large1page * 2 : MDBX_PNL_INITIAL;
+  const size_t one_third = env->geo_in_bytes.now / 3 >> env->ps2ln;
+  const size_t augment_limit =
+      (one_third > minimum) ? minimum + (one_third - minimum) / timeframe * remain_1sec : minimum;
+  eASSERT(env, augment_limit < PAGELIST_LIMIT);
+  return pnl_bytes2size(pnl_size2bytes(augment_limit));
+}
+
+static bool default_prefault_write(const MDBX_env *env) {
+  return !MDBX_MMAP_INCOHERENT_FILE_WRITE && !env->incore &&
+         (env->flags & (MDBX_WRITEMAP | MDBX_RDONLY)) == MDBX_WRITEMAP;
+}
+
+static bool default_prefer_waf_insteadof_balance(const MDBX_env *env) {
+  (void)env;
+  return false;
+}
+
+static uint16_t default_subpage_limit(const MDBX_env *env) {
+  (void)env;
+  return 65535 /* 100% */;
+}
+
+static uint16_t default_subpage_room_threshold(const MDBX_env *env) {
+  (void)env;
+  return 0 /* 0% */;
+}
+
+static uint16_t default_subpage_reserve_prereq(const MDBX_env *env) {
+  (void)env;
+  return 27525 /* 42% */;
+}
+
+static uint16_t default_subpage_reserve_limit(const MDBX_env *env) {
+  (void)env;
+  return 2753 /* 4.2% */;
+}
+
+static uint16_t default_merge_threshold_16dot16_percent(const MDBX_env *env) {
+  (void)env;
+  return 65536 / 4 /* 25% */;
+}
+
+static pgno_t default_dp_reserve_limit(const MDBX_env *env) {
+  (void)env;
+  return MDBX_PNL_INITIAL;
+}
+
+static pgno_t default_dp_initial(const MDBX_env *env) {
+  (void)env;
+  return MDBX_PNL_INITIAL;
+}
+
+static uint8_t default_spill_max_denominator(const MDBX_env *env) {
+  (void)env;
+  return 8;
+}
+
+static uint8_t default_spill_min_denominator(const MDBX_env *env) {
+  (void)env;
+  return 8;
+}
+
+static uint8_t default_spill_parent4child_denominator(const MDBX_env *env) {
+  (void)env;
+  return 0;
+}
+
+static uint8_t default_dp_loose_limit(const MDBX_env *env) {
+  (void)env;
+  return 64;
+}
+
+void env_options_init(MDBX_env *env) {
+  env->options.rp_augment_limit = default_rp_augment_limit(env);
+  env->options.dp_reserve_limit = default_dp_reserve_limit(env);
+  env->options.dp_initial = default_dp_initial(env);
+  env->options.dp_limit = default_dp_limit(env);
+  env->options.spill_max_denominator = default_spill_max_denominator(env);
+  env->options.spill_min_denominator = default_spill_min_denominator(env);
+  env->options.spill_parent4child_denominator = default_spill_parent4child_denominator(env);
+  env->options.dp_loose_limit = default_dp_loose_limit(env);
+  env->options.merge_threshold_16dot16_percent = default_merge_threshold_16dot16_percent(env);
+  if (default_prefer_waf_insteadof_balance(env))
+    env->options.prefer_waf_insteadof_balance = true;
+
+#if !(defined(_WIN32) || defined(_WIN64))
+  env->options.writethrough_threshold =
+#if defined(__linux__) || defined(__gnu_linux__)
+      globals.running_on_WSL1 ? MAX_PAGENO :
+#endif /* Linux */
+                              MDBX_WRITETHROUGH_THRESHOLD_DEFAULT;
+#endif /* Windows */
+
+  env->options.subpage.limit = default_subpage_limit(env);
+  env->options.subpage.room_threshold = default_subpage_room_threshold(env);
+  env->options.subpage.reserve_prereq = default_subpage_reserve_prereq(env);
+  env->options.subpage.reserve_limit = default_subpage_reserve_limit(env);
+}
+
+void env_options_adjust_dp_limit(MDBX_env *env) {
+  if (!env->options.flags.non_auto.dp_limit)
+    env->options.dp_limit = default_dp_limit(env);
+  else {
+    const pgno_t max_pgno = env_max_pgno(env);
+    if (env->options.dp_limit > max_pgno - NUM_METAS)
+      env->options.dp_limit = max_pgno - NUM_METAS;
+    if (env->options.dp_limit < CURSOR_STACK_SIZE * 4)
+      env->options.dp_limit = CURSOR_STACK_SIZE * 4;
+  }
+#ifdef MDBX_DEBUG_DPL_LIMIT
+  env->options.dp_limit = MDBX_DEBUG_DPL_LIMIT;
+#endif /* MDBX_DEBUG_DPL_LIMIT */
+  if (env->options.dp_initial > env->options.dp_limit && env->options.dp_initial > default_dp_initial(env))
+    env->options.dp_initial = env->options.dp_limit;
+  env->options.need_dp_limit_adjust = false;
+}
+
+void env_options_adjust_defaults(MDBX_env *env) {
+  if (!env->options.flags.non_auto.rp_augment_limit)
+    env->options.rp_augment_limit = default_rp_augment_limit(env);
+  if (!env->options.flags.non_auto.prefault_write)
+    env->options.prefault_write = default_prefault_write(env);
+
+  env->options.need_dp_limit_adjust = true;
+  if (!env->txn)
+    env_options_adjust_dp_limit(env);
+
+  const size_t basis = env->geo_in_bytes.now;
+  /* TODO: use options? */
+  const unsigned factor = 9;
+  size_t threshold = (basis < ((size_t)65536 << factor))  ? 65536        /* minimal threshold */
+                     : (basis > (MEGABYTE * 4 << factor)) ? MEGABYTE * 4 /* maximal threshold */
+                                                          : basis >> factor;
+  threshold =
+      (threshold < env->geo_in_bytes.shrink || !env->geo_in_bytes.shrink) ? threshold : env->geo_in_bytes.shrink;
+  env->madv_threshold = bytes2pgno(env, bytes_align2os_bytes(env, threshold));
+}
+
+//------------------------------------------------------------------------------
+
+__cold int mdbx_env_set_option(MDBX_env *env, const MDBX_option_t option, uint64_t value) {
+  int err = check_env(env, false);
+  if (unlikely(err != MDBX_SUCCESS))
+    return LOG_IFERR(err);
+
+  const bool lock_needed = ((env->flags & ENV_ACTIVE) && env->basal_txn && !env_owned_wrtxn(env));
+  bool should_unlock = false;
+  switch (option) {
+  case MDBX_opt_sync_bytes:
+    if (value == /* default */ UINT64_MAX)
+      value = MAX_WRITE;
+    if (unlikely(env->flags & MDBX_RDONLY))
+      return LOG_IFERR(MDBX_EACCESS);
+    if (unlikely(!(env->flags & ENV_ACTIVE)))
+      return LOG_IFERR(MDBX_EPERM);
+    if (unlikely(value > SIZE_MAX - 65536))
+      return LOG_IFERR(MDBX_EINVAL);
+    value = bytes2pgno(env, (size_t)value + env->ps - 1);
+    if ((uint32_t)value != atomic_load32(&env->lck->autosync_threshold, mo_AcquireRelease) &&
+        atomic_store32(&env->lck->autosync_threshold, (uint32_t)value, mo_Relaxed)
+        /* Дергаем sync(force=off) только если задано новое не-нулевое значение
+         * и мы вне транзакции */
+        && lock_needed) {
+      err = env_sync(env, false, false);
+      if (err == /* нечего сбрасывать на диск */ MDBX_RESULT_TRUE)
+        err = MDBX_SUCCESS;
+    }
+    break;
+
+  case MDBX_opt_sync_period:
+    if (value == /* default */ UINT64_MAX)
+      value = 2780315 /* 42.42424 секунды */;
+    if (unlikely(env->flags & MDBX_RDONLY))
+      return LOG_IFERR(MDBX_EACCESS);
+    if (unlikely(!(env->flags & ENV_ACTIVE)))
+      return LOG_IFERR(MDBX_EPERM);
+    if (unlikely(value > UINT32_MAX))
+      return LOG_IFERR(MDBX_EINVAL);
+    value = osal_16dot16_to_monotime((uint32_t)value);
+    if (value != atomic_load64(&env->lck->autosync_period, mo_AcquireRelease) &&
+        atomic_store64(&env->lck->autosync_period, value, mo_Relaxed)
+        /* Дергаем sync(force=off) только если задано новое не-нулевое значение
+         * и мы вне транзакции */
+        && lock_needed) {
+      err = env_sync(env, false, false);
+      if (err == /* нечего сбрасывать на диск */ MDBX_RESULT_TRUE)
+        err = MDBX_SUCCESS;
+    }
+    break;
+
+  case MDBX_opt_max_db:
+    if (value == /* default */ UINT64_MAX)
+      value = 42;
+    if (unlikely(value > MDBX_MAX_DBI))
+      return LOG_IFERR(MDBX_EINVAL);
+    if (unlikely(env->dxb_mmap.base))
+      return LOG_IFERR(MDBX_EPERM);
+    env->max_dbi = (unsigned)value + CORE_DBS;
+    break;
+
+  case MDBX_opt_max_readers:
+    if (value == /* default */ UINT64_MAX)
+      value = MDBX_READERS_LIMIT;
+    if (unlikely(value < 1 || value > MDBX_READERS_LIMIT))
+      return LOG_IFERR(MDBX_EINVAL);
+    if (unlikely(env->dxb_mmap.base))
+      return LOG_IFERR(MDBX_EPERM);
+    env->max_readers = (unsigned)value;
+    break;
+
+  case MDBX_opt_dp_reserve_limit:
+    if (value == /* default */ UINT64_MAX)
+      value = default_dp_reserve_limit(env);
+    if (unlikely(value > INT_MAX))
+      return LOG_IFERR(MDBX_EINVAL);
+    if (env->options.dp_reserve_limit != (unsigned)value) {
+      if (lock_needed) {
+        err = lck_txn_lock(env, false);
+        if (unlikely(err != MDBX_SUCCESS))
+          return LOG_IFERR(err);
+        should_unlock = true;
+      }
+      env->options.dp_reserve_limit = (unsigned)value;
+      while (env->shadow_reserve_len > env->options.dp_reserve_limit) {
+        eASSERT(env, env->shadow_reserve != nullptr);
+        page_t *dp = env->shadow_reserve;
+        MDBX_ASAN_UNPOISON_MEMORY_REGION(dp, env->ps);
+        VALGRIND_MAKE_MEM_DEFINED(&page_next(dp), sizeof(page_t *));
+        env->shadow_reserve = page_next(dp);
+        void *const ptr = ptr_disp(dp, -(ptrdiff_t)sizeof(size_t));
+        osal_free(ptr);
+        env->shadow_reserve_len -= 1;
+      }
+    }
+    break;
+
+  case MDBX_opt_rp_augment_limit:
+    if (value == /* default */ UINT64_MAX) {
+      env->options.flags.non_auto.rp_augment_limit = 0;
+      env->options.rp_augment_limit = default_rp_augment_limit(env);
+    } else if (unlikely(value > PAGELIST_LIMIT))
+      return LOG_IFERR(MDBX_EINVAL);
+    else {
+      env->options.flags.non_auto.rp_augment_limit = 1;
+      env->options.rp_augment_limit = (unsigned)value;
+    }
+    break;
+
+  case MDBX_opt_gc_time_limit:
+    if (value == /* default */ UINT64_MAX)
+      value = 0;
+    if (unlikely(value > UINT32_MAX))
+      return LOG_IFERR(MDBX_EINVAL);
+    if (unlikely(env->flags & MDBX_RDONLY))
+      return LOG_IFERR(MDBX_EACCESS);
+    value = osal_16dot16_to_monotime((uint32_t)value);
+    if (value != env->options.gc_time_limit) {
+      if (env->txn && lock_needed)
+        return LOG_IFERR(MDBX_EPERM);
+      env->options.gc_time_limit = value;
+      if (!env->options.flags.non_auto.rp_augment_limit)
+        env->options.rp_augment_limit = default_rp_augment_limit(env);
+    }
+    break;
+
+  case MDBX_opt_txn_dp_limit:
+  case MDBX_opt_txn_dp_initial:
+    if (value != /* default */ UINT64_MAX && unlikely(value > PAGELIST_LIMIT || value < CURSOR_STACK_SIZE * 4))
+      return LOG_IFERR(MDBX_EINVAL);
+    if (unlikely(env->flags & MDBX_RDONLY))
+      return LOG_IFERR(MDBX_EACCESS);
+    if (lock_needed) {
+      err = lck_txn_lock(env, false);
+      if (unlikely(err != MDBX_SUCCESS))
+        return LOG_IFERR(err);
+      should_unlock = true;
+    }
+    if (env->txn)
+      err = MDBX_EPERM /* unable change during transaction */;
+    else {
+      const pgno_t max_pgno = env_max_pgno(env);
+      if (option == MDBX_opt_txn_dp_initial) {
+        if (value == /* default */ UINT64_MAX)
+          env->options.dp_initial = default_dp_initial(env);
+        else {
+          env->options.dp_initial = (pgno_t)value;
+          if (env->options.dp_initial > max_pgno)
+            env->options.dp_initial = (max_pgno > CURSOR_STACK_SIZE * 4) ? max_pgno : CURSOR_STACK_SIZE * 4;
+        }
+      }
+      if (option == MDBX_opt_txn_dp_limit) {
+        if (value == /* default */ UINT64_MAX) {
+          env->options.flags.non_auto.dp_limit = 0;
+        } else {
+          env->options.flags.non_auto.dp_limit = 1;
+          env->options.dp_limit = (pgno_t)value;
+        }
+        env_options_adjust_dp_limit(env);
+      }
+    }
+    break;
+
+  case MDBX_opt_spill_max_denominator:
+    if (value == /* default */ UINT64_MAX)
+      value = default_spill_max_denominator(env);
+    if (unlikely(value > 255))
+      return LOG_IFERR(MDBX_EINVAL);
+    env->options.spill_max_denominator = (uint8_t)value;
+    break;
+  case MDBX_opt_spill_min_denominator:
+    if (value == /* default */ UINT64_MAX)
+      value = default_spill_min_denominator(env);
+    if (unlikely(value > 255))
+      return LOG_IFERR(MDBX_EINVAL);
+    env->options.spill_min_denominator = (uint8_t)value;
+    break;
+  case MDBX_opt_spill_parent4child_denominator:
+    if (value == /* default */ UINT64_MAX)
+      value = default_spill_parent4child_denominator(env);
+    if (unlikely(value > 255))
+      return LOG_IFERR(MDBX_EINVAL);
+    env->options.spill_parent4child_denominator = (uint8_t)value;
+    break;
+
+  case MDBX_opt_loose_limit:
+    if (value == /* default */ UINT64_MAX)
+      value = default_dp_loose_limit(env);
+    if (unlikely(value > 255))
+      return LOG_IFERR(MDBX_EINVAL);
+    env->options.dp_loose_limit = (uint8_t)value;
+    break;
+
+  case MDBX_opt_merge_threshold_16dot16_percent:
+    if (value == /* default */ UINT64_MAX)
+      value = default_merge_threshold_16dot16_percent(env);
+    if (unlikely(value < 8192 || value > 32768))
+      return LOG_IFERR(MDBX_EINVAL);
+    env->options.merge_threshold_16dot16_percent = (unsigned)value;
+    recalculate_merge_thresholds(env);
+    break;
+
+  case MDBX_opt_writethrough_threshold:
+#if defined(_WIN32) || defined(_WIN64)
+    /* позволяем "установить" значение по-умолчанию и совпадающее
+     * с поведением соответствующим текущей установке MDBX_NOMETASYNC */
+    if (value == /* default */ UINT64_MAX && value != ((env->flags & MDBX_NOMETASYNC) ? 0 : UINT_MAX))
+      err = MDBX_EINVAL;
+#else
+    if (value == /* default */ UINT64_MAX)
+      value = MDBX_WRITETHROUGH_THRESHOLD_DEFAULT;
+    if (value != (unsigned)value)
+      err = MDBX_EINVAL;
+    else
+      env->options.writethrough_threshold = (unsigned)value;
+#endif
+    break;
+
+  case MDBX_opt_prefault_write_enable:
+    if (value == /* default */ UINT64_MAX) {
+      env->options.prefault_write = default_prefault_write(env);
+      env->options.flags.non_auto.prefault_write = false;
+    } else if (value > 1)
+      err = MDBX_EINVAL;
+    else {
+      env->options.prefault_write = value != 0;
+      env->options.flags.non_auto.prefault_write = true;
+    }
+    break;
+
+  case MDBX_opt_prefer_waf_insteadof_balance:
+    if (value == /* default */ UINT64_MAX)
+      env->options.prefer_waf_insteadof_balance = default_prefer_waf_insteadof_balance(env);
+    else if (value > 1)
+      err = MDBX_EINVAL;
+    else
+      env->options.prefer_waf_insteadof_balance = value != 0;
+    break;
+
+  case MDBX_opt_subpage_limit:
+    if (value == /* default */ UINT64_MAX) {
+      env->options.subpage.limit = default_subpage_limit(env);
+      recalculate_subpage_thresholds(env);
+    } else if (value > 65535)
+      err = MDBX_EINVAL;
+    else {
+      env->options.subpage.limit = (uint16_t)value;
+      recalculate_subpage_thresholds(env);
+    }
+    break;
+
+  case MDBX_opt_subpage_room_threshold:
+    if (value == /* default */ UINT64_MAX) {
+      env->options.subpage.room_threshold = default_subpage_room_threshold(env);
+      recalculate_subpage_thresholds(env);
+    } else if (value > 65535)
+      err = MDBX_EINVAL;
+    else {
+      env->options.subpage.room_threshold = (uint16_t)value;
+      recalculate_subpage_thresholds(env);
+    }
+    break;
+
+  case MDBX_opt_subpage_reserve_prereq:
+    if (value == /* default */ UINT64_MAX) {
+      env->options.subpage.reserve_prereq = default_subpage_reserve_prereq(env);
+      recalculate_subpage_thresholds(env);
+    } else if (value > 65535)
+      err = MDBX_EINVAL;
+    else {
+      env->options.subpage.reserve_prereq = (uint16_t)value;
+      recalculate_subpage_thresholds(env);
+    }
+    break;
+
+  case MDBX_opt_subpage_reserve_limit:
+    if (value == /* default */ UINT64_MAX) {
+      env->options.subpage.reserve_limit = default_subpage_reserve_limit(env);
+      recalculate_subpage_thresholds(env);
+    } else if (value > 65535)
+      err = MDBX_EINVAL;
+    else {
+      env->options.subpage.reserve_limit = (uint16_t)value;
+      recalculate_subpage_thresholds(env);
+    }
+    break;
+
+  default:
+    return LOG_IFERR(MDBX_EINVAL);
+  }
+
+  if (should_unlock)
+    lck_txn_unlock(env);
+  return LOG_IFERR(err);
+}
+
+__cold int mdbx_env_get_option(const MDBX_env *env, const MDBX_option_t option, uint64_t *pvalue) {
+  int err = check_env(env, false);
+  if (unlikely(err != MDBX_SUCCESS))
+    return LOG_IFERR(err);
+  if (unlikely(!pvalue))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  switch (option) {
+  case MDBX_opt_sync_bytes:
+    if (unlikely(!(env->flags & ENV_ACTIVE)))
+      return LOG_IFERR(MDBX_EPERM);
+    *pvalue = pgno2bytes(env, atomic_load32(&env->lck->autosync_threshold, mo_Relaxed));
+    break;
+
+  case MDBX_opt_sync_period:
+    if (unlikely(!(env->flags & ENV_ACTIVE)))
+      return LOG_IFERR(MDBX_EPERM);
+    *pvalue = osal_monotime_to_16dot16(atomic_load64(&env->lck->autosync_period, mo_Relaxed));
+    break;
+
+  case MDBX_opt_max_db:
+    *pvalue = env->max_dbi - CORE_DBS;
+    break;
+
+  case MDBX_opt_max_readers:
+    *pvalue = env->max_readers;
+    break;
+
+  case MDBX_opt_dp_reserve_limit:
+    *pvalue = env->options.dp_reserve_limit;
+    break;
+
+  case MDBX_opt_rp_augment_limit:
+    *pvalue = env->options.rp_augment_limit;
+    break;
+
+  case MDBX_opt_gc_time_limit:
+    *pvalue = osal_monotime_to_16dot16(env->options.gc_time_limit);
+    break;
+
+  case MDBX_opt_txn_dp_limit:
+    *pvalue = env->options.dp_limit;
+    break;
+  case MDBX_opt_txn_dp_initial:
+    *pvalue = env->options.dp_initial;
+    break;
+
+  case MDBX_opt_spill_max_denominator:
+    *pvalue = env->options.spill_max_denominator;
+    break;
+  case MDBX_opt_spill_min_denominator:
+    *pvalue = env->options.spill_min_denominator;
+    break;
+  case MDBX_opt_spill_parent4child_denominator:
+    *pvalue = env->options.spill_parent4child_denominator;
+    break;
+
+  case MDBX_opt_loose_limit:
+    *pvalue = env->options.dp_loose_limit;
+    break;
+
+  case MDBX_opt_merge_threshold_16dot16_percent:
+    *pvalue = env->options.merge_threshold_16dot16_percent;
+    break;
+
+  case MDBX_opt_writethrough_threshold:
+#if defined(_WIN32) || defined(_WIN64)
+    *pvalue = (env->flags & MDBX_NOMETASYNC) ? 0 : INT_MAX;
+#else
+    *pvalue = env->options.writethrough_threshold;
+#endif
+    break;
+
+  case MDBX_opt_prefault_write_enable:
+    *pvalue = env->options.prefault_write;
+    break;
+
+  case MDBX_opt_prefer_waf_insteadof_balance:
+    *pvalue = env->options.prefer_waf_insteadof_balance;
+    break;
+
+  case MDBX_opt_subpage_limit:
+    *pvalue = env->options.subpage.limit;
+    break;
+
+  case MDBX_opt_subpage_room_threshold:
+    *pvalue = env->options.subpage.room_threshold;
+    break;
+
+  case MDBX_opt_subpage_reserve_prereq:
+    *pvalue = env->options.subpage.reserve_prereq;
+    break;
+
+  case MDBX_opt_subpage_reserve_limit:
+    *pvalue = env->options.subpage.reserve_limit;
+    break;
+
+  default:
+    return LOG_IFERR(MDBX_EINVAL);
+  }
+
+  return MDBX_SUCCESS;
+}

src/api-range-estimate.c

@@ -0,0 +1,365 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+typedef struct diff_result {
+  ptrdiff_t diff;
+  intptr_t level;
+  ptrdiff_t root_nkeys;
+} diff_t;
+
+/* calculates: r = x - y */
+__hot static int cursor_diff(const MDBX_cursor *const __restrict x, const MDBX_cursor *const __restrict y,
+                             diff_t *const __restrict r) {
+  r->diff = 0;
+  r->level = 0;
+  r->root_nkeys = 0;
+
+  int rc = check_txn(x->txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return rc;
+
+  if (unlikely(x->txn != y->txn))
+    return MDBX_BAD_TXN;
+
+  if (unlikely(y->dbi_state != x->dbi_state))
+    return MDBX_EINVAL;
+
+  const intptr_t depth = (x->top < y->top) ? x->top : y->top;
+  if (unlikely(depth < 0))
+    return MDBX_ENODATA;
+
+  r->root_nkeys = page_numkeys(x->pg[0]);
+  intptr_t nkeys = r->root_nkeys;
+  for (;;) {
+    if (unlikely(y->pg[r->level] != x->pg[r->level])) {
+      ERROR("Mismatch cursors's pages at %zu level", r->level);
+      return MDBX_PROBLEM;
+    }
+    r->diff = x->ki[r->level] - y->ki[r->level];
+    if (r->diff)
+      break;
+    r->level += 1;
+    if (r->level > depth) {
+      r->diff = CMP2INT(x->flags & z_eof_hard, y->flags & z_eof_hard);
+      return MDBX_SUCCESS;
+    }
+    nkeys = page_numkeys(x->pg[r->level]);
+  }
+
+  while (unlikely(r->diff == 1) && likely(r->level < depth)) {
+    r->level += 1;
+    /*   DB'PAGEs: 0------------------>MAX
+     *
+     *    CURSORs:       y < x
+     *  STACK[i ]:         |
+     *  STACK[+1]:  ...y++N|0++x...
+     */
+    nkeys = page_numkeys(y->pg[r->level]);
+    r->diff = (nkeys - y->ki[r->level]) + x->ki[r->level];
+    assert(r->diff > 0);
+  }
+
+  while (unlikely(r->diff == -1) && likely(r->level < depth)) {
+    r->level += 1;
+    /*   DB'PAGEs: 0------------------>MAX
+     *
+     *    CURSORs:       x < y
+     *  STACK[i ]:         |
+     *  STACK[+1]:  ...x--N|0--y...
+     */
+    nkeys = page_numkeys(x->pg[r->level]);
+    r->diff = -(nkeys - x->ki[r->level]) - y->ki[r->level];
+    assert(r->diff < 0);
+  }
+
+  return MDBX_SUCCESS;
+}
+
+__hot static ptrdiff_t estimate(const tree_t *tree, diff_t *const __restrict dr) {
+  /*        root: branch-page    => scale = leaf-factor * branch-factor^(N-1)
+   *     level-1: branch-page(s) => scale = leaf-factor * branch-factor^2
+   *     level-2: branch-page(s) => scale = leaf-factor * branch-factor
+   *     level-N: branch-page(s) => scale = leaf-factor
+   *  leaf-level: leaf-page(s)   => scale = 1
+   */
+  ptrdiff_t btree_power = (ptrdiff_t)tree->height - 2 - (ptrdiff_t)dr->level;
+  if (btree_power < 0)
+    return dr->diff;
+
+  ptrdiff_t estimated = (ptrdiff_t)tree->items * dr->diff / (ptrdiff_t)tree->leaf_pages;
+  if (btree_power == 0)
+    return estimated;
+
+  if (tree->height < 4) {
+    assert(dr->level == 0 && btree_power == 1);
+    return (ptrdiff_t)tree->items * dr->diff / (ptrdiff_t)dr->root_nkeys;
+  }
+
+  /* average_branchpage_fillfactor = total(branch_entries) / branch_pages
+     total(branch_entries) = leaf_pages + branch_pages - 1 (root page) */
+  const size_t log2_fixedpoint = sizeof(size_t) - 1;
+  const size_t half = UINT64_C(1) << (log2_fixedpoint - 1);
+  const size_t factor = ((tree->leaf_pages + tree->branch_pages - 1) << log2_fixedpoint) / tree->branch_pages;
+  while (1) {
+    switch ((size_t)btree_power) {
+    default: {
+      const size_t square = (factor * factor + half) >> log2_fixedpoint;
+      const size_t quad = (square * square + half) >> log2_fixedpoint;
+      do {
+        estimated = estimated * quad + half;
+        estimated >>= log2_fixedpoint;
+        btree_power -= 4;
+      } while (btree_power >= 4);
+      continue;
+    }
+    case 3:
+      estimated = estimated * factor + half;
+      estimated >>= log2_fixedpoint;
+      __fallthrough /* fall through */;
+    case 2:
+      estimated = estimated * factor + half;
+      estimated >>= log2_fixedpoint;
+      __fallthrough /* fall through */;
+    case 1:
+      estimated = estimated * factor + half;
+      estimated >>= log2_fixedpoint;
+      __fallthrough /* fall through */;
+    case 0:
+      if (unlikely(estimated > (ptrdiff_t)tree->items))
+        return (ptrdiff_t)tree->items;
+      if (unlikely(estimated < -(ptrdiff_t)tree->items))
+        return -(ptrdiff_t)tree->items;
+      return estimated;
+    }
+  }
+}
+
+/*------------------------------------------------------------------------------
+ * Range-Estimation API */
+
+__hot int mdbx_estimate_distance(const MDBX_cursor *first, const MDBX_cursor *last, ptrdiff_t *distance_items) {
+  if (unlikely(!distance_items))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = cursor_check_pure(first);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  rc = cursor_check_pure(last);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  *distance_items = 0;
+  diff_t dr;
+  rc = cursor_diff(last, first, &dr);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cASSERT(first, dr.diff || inner_pointed(first) == inner_pointed(last));
+  if (unlikely(dr.diff == 0) && inner_pointed(first)) {
+    first = &first->subcur->cursor;
+    last = &last->subcur->cursor;
+    rc = cursor_diff(first, last, &dr);
+    if (unlikely(rc != MDBX_SUCCESS))
+      return LOG_IFERR(rc);
+  }
+
+  if (likely(dr.diff != 0))
+    *distance_items = estimate(first->tree, &dr);
+
+  return MDBX_SUCCESS;
+}
+
+__hot int mdbx_estimate_move(const MDBX_cursor *cursor, MDBX_val *key, MDBX_val *data, MDBX_cursor_op move_op,
+                             ptrdiff_t *distance_items) {
+  if (unlikely(!distance_items || move_op == MDBX_GET_CURRENT || move_op == MDBX_GET_MULTIPLE))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = cursor_check_ro(cursor);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(!is_pointed(cursor)))
+    return LOG_IFERR(MDBX_ENODATA);
+
+  cursor_couple_t next;
+  rc = cursor_init(&next.outer, cursor->txn, cursor_dbi(cursor));
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cursor_cpstk(cursor, &next.outer);
+  if (cursor->tree->flags & MDBX_DUPSORT) {
+    subcur_t *mx = &container_of(cursor, cursor_couple_t, outer)->inner;
+    cursor_cpstk(&mx->cursor, &next.inner.cursor);
+  }
+
+  MDBX_val stub_data;
+  if (data == nullptr) {
+    const unsigned mask = 1 << MDBX_GET_BOTH | 1 << MDBX_GET_BOTH_RANGE | 1 << MDBX_SET_KEY;
+    if (unlikely(mask & (1 << move_op)))
+      return LOG_IFERR(MDBX_EINVAL);
+    stub_data.iov_base = nullptr;
+    stub_data.iov_len = 0;
+    data = &stub_data;
+  }
+
+  MDBX_val stub_key;
+  if (key == nullptr) {
+    const unsigned mask =
+        1 << MDBX_GET_BOTH | 1 << MDBX_GET_BOTH_RANGE | 1 << MDBX_SET_KEY | 1 << MDBX_SET | 1 << MDBX_SET_RANGE;
+    if (unlikely(mask & (1 << move_op)))
+      return LOG_IFERR(MDBX_EINVAL);
+    stub_key.iov_base = nullptr;
+    stub_key.iov_len = 0;
+    key = &stub_key;
+  }
+
+  next.outer.signature = cur_signature_live;
+  rc = cursor_ops(&next.outer, key, data, move_op);
+  if (unlikely(rc != MDBX_SUCCESS && (rc != MDBX_NOTFOUND || !is_pointed(&next.outer))))
+    return LOG_IFERR(rc);
+
+  if (move_op == MDBX_LAST) {
+    next.outer.flags |= z_eof_hard;
+    next.inner.cursor.flags |= z_eof_hard;
+  }
+  return mdbx_estimate_distance(cursor, &next.outer, distance_items);
+}
+
+__hot int mdbx_estimate_range(const MDBX_txn *txn, MDBX_dbi dbi, const MDBX_val *begin_key, const MDBX_val *begin_data,
+                              const MDBX_val *end_key, const MDBX_val *end_data, ptrdiff_t *size_items) {
+  if (unlikely(!size_items))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  if (unlikely(begin_data && (begin_key == nullptr || begin_key == MDBX_EPSILON)))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  if (unlikely(end_data && (end_key == nullptr || end_key == MDBX_EPSILON)))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  if (unlikely(begin_key == MDBX_EPSILON && end_key == MDBX_EPSILON))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cursor_couple_t begin;
+  /* LY: first, initialize cursor to refresh a DB in case it have DB_STALE */
+  rc = cursor_init(&begin.outer, txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(begin.outer.tree->items == 0)) {
+    *size_items = 0;
+    return MDBX_SUCCESS;
+  }
+
+  if (!begin_key) {
+    if (unlikely(!end_key)) {
+      /* LY: FIRST..LAST case */
+      *size_items = (ptrdiff_t)begin.outer.tree->items;
+      return MDBX_SUCCESS;
+    }
+    rc = outer_first(&begin.outer, nullptr, nullptr);
+    if (unlikely(end_key == MDBX_EPSILON)) {
+      /* LY: FIRST..+epsilon case */
+      return LOG_IFERR((rc == MDBX_SUCCESS) ? mdbx_cursor_count(&begin.outer, (size_t *)size_items) : rc);
+    }
+  } else {
+    if (unlikely(begin_key == MDBX_EPSILON)) {
+      if (end_key == nullptr) {
+        /* LY: -epsilon..LAST case */
+        rc = outer_last(&begin.outer, nullptr, nullptr);
+        return LOG_IFERR((rc == MDBX_SUCCESS) ? mdbx_cursor_count(&begin.outer, (size_t *)size_items) : rc);
+      }
+      /* LY: -epsilon..value case */
+      assert(end_key != MDBX_EPSILON);
+      begin_key = end_key;
+    } else if (unlikely(end_key == MDBX_EPSILON)) {
+      /* LY: value..+epsilon case */
+      assert(begin_key != MDBX_EPSILON);
+      end_key = begin_key;
+    }
+    if (end_key && !begin_data && !end_data &&
+        (begin_key == end_key || begin.outer.clc->k.cmp(begin_key, end_key) == 0)) {
+      /* LY: single key case */
+      rc = cursor_seek(&begin.outer, (MDBX_val *)begin_key, nullptr, MDBX_SET).err;
+      if (unlikely(rc != MDBX_SUCCESS)) {
+        *size_items = 0;
+        return LOG_IFERR((rc == MDBX_NOTFOUND) ? MDBX_SUCCESS : rc);
+      }
+      *size_items = 1;
+      if (inner_pointed(&begin.outer))
+        *size_items = (sizeof(*size_items) >= sizeof(begin.inner.nested_tree.items) ||
+                       begin.inner.nested_tree.items <= PTRDIFF_MAX)
+                          ? (size_t)begin.inner.nested_tree.items
+                          : PTRDIFF_MAX;
+
+      return MDBX_SUCCESS;
+    } else {
+      MDBX_val proxy_key = *begin_key;
+      MDBX_val proxy_data = {nullptr, 0};
+      if (begin_data)
+        proxy_data = *begin_data;
+      rc = LOG_IFERR(cursor_seek(&begin.outer, &proxy_key, &proxy_data, MDBX_SET_LOWERBOUND).err);
+    }
+  }
+
+  if (unlikely(rc != MDBX_SUCCESS)) {
+    if (rc != MDBX_NOTFOUND || !is_pointed(&begin.outer))
+      return LOG_IFERR(rc);
+  }
+
+  cursor_couple_t end;
+  rc = cursor_init(&end.outer, txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+  if (!end_key) {
+    rc = outer_last(&end.outer, nullptr, nullptr);
+    end.outer.flags |= z_eof_hard;
+    end.inner.cursor.flags |= z_eof_hard;
+  } else {
+    MDBX_val proxy_key = *end_key;
+    MDBX_val proxy_data = {nullptr, 0};
+    if (end_data)
+      proxy_data = *end_data;
+    rc = cursor_seek(&end.outer, &proxy_key, &proxy_data, MDBX_SET_LOWERBOUND).err;
+  }
+  if (unlikely(rc != MDBX_SUCCESS)) {
+    if (rc != MDBX_NOTFOUND || !is_pointed(&end.outer))
+      return LOG_IFERR(rc);
+  }
+
+  rc = mdbx_estimate_distance(&begin.outer, &end.outer, size_items);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+  assert(*size_items >= -(ptrdiff_t)begin.outer.tree->items && *size_items <= (ptrdiff_t)begin.outer.tree->items);
+
+#if 0 /* LY: Was decided to returns as-is (i.e. negative) the estimation                                               \
+       * results for an inverted ranges. */
+
+  /* Commit 8ddfd1f34ad7cf7a3c4aa75d2e248ca7e639ed63
+     Change-Id: If59eccf7311123ab6384c4b93f9b1fed5a0a10d1 */
+
+  if (*size_items < 0) {
+    /* LY: inverted range case */
+    *size_items += (ptrdiff_t)begin.outer.tree->items;
+  } else if (*size_items == 0 && begin_key && end_key) {
+    int cmp = begin.outer.kvx->cmp(&origin_begin_key, &origin_end_key);
+    if (cmp == 0 && cursor_pointed(begin.inner.cursor.flags) &&
+        begin_data && end_data)
+      cmp = begin.outer.kvx->v.cmp(&origin_begin_data, &origin_end_data);
+    if (cmp > 0) {
+      /* LY: inverted range case with empty scope */
+      *size_items = (ptrdiff_t)begin.outer.tree->items;
+    }
+  }
+  assert(*size_items >= 0 &&
+         *size_items <= (ptrdiff_t)begin.outer.tree->items);
+#endif
+
+  return MDBX_SUCCESS;
+}

src/api-txn-data.c

@@ -0,0 +1,454 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+__cold int mdbx_dbi_dupsort_depthmask(const MDBX_txn *txn, MDBX_dbi dbi, uint32_t *mask) {
+  if (unlikely(!mask))
+    return LOG_IFERR(MDBX_EINVAL);
+  *mask = 0;
+
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cursor_couple_t cx;
+  rc = cursor_init(&cx.outer, txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if ((cx.outer.tree->flags & MDBX_DUPSORT) == 0)
+    return MDBX_RESULT_TRUE;
+
+  MDBX_val key, data;
+  rc = outer_first(&cx.outer, &key, &data);
+  while (rc == MDBX_SUCCESS) {
+    const node_t *node = page_node(cx.outer.pg[cx.outer.top], cx.outer.ki[cx.outer.top]);
+    const tree_t *db = node_data(node);
+    const unsigned flags = node_flags(node);
+    switch (flags) {
+    case N_BIG:
+    case 0:
+      /* single-value entry, deep = 0 */
+      *mask |= 1 << 0;
+      break;
+    case N_DUP:
+      /* single sub-page, deep = 1 */
+      *mask |= 1 << 1;
+      break;
+    case N_DUP | N_TREE:
+      /* sub-tree */
+      *mask |= 1 << UNALIGNED_PEEK_16(db, tree_t, height);
+      break;
+    default:
+      ERROR("%s/%d: %s %u", "MDBX_CORRUPTED", MDBX_CORRUPTED, "invalid node-size", flags);
+      return LOG_IFERR(MDBX_CORRUPTED);
+    }
+    rc = outer_next(&cx.outer, &key, &data, MDBX_NEXT_NODUP);
+  }
+
+  return LOG_IFERR((rc == MDBX_NOTFOUND) ? MDBX_SUCCESS : rc);
+}
+
+int mdbx_canary_get(const MDBX_txn *txn, MDBX_canary *canary) {
+  if (unlikely(canary == nullptr))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED - MDBX_TXN_PARKED);
+  if (unlikely(rc != MDBX_SUCCESS)) {
+    memset(canary, 0, sizeof(*canary));
+    return LOG_IFERR(rc);
+  }
+
+  *canary = txn->canary;
+  return MDBX_SUCCESS;
+}
+
+int mdbx_get(const MDBX_txn *txn, MDBX_dbi dbi, const MDBX_val *key, MDBX_val *data) {
+  DKBUF_DEBUG;
+  DEBUG("===> get db %u key [%s]", dbi, DKEY_DEBUG(key));
+
+  if (unlikely(!key || !data))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cursor_couple_t cx;
+  rc = cursor_init(&cx.outer, txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  return LOG_IFERR(cursor_seek(&cx.outer, (MDBX_val *)key, data, MDBX_SET).err);
+}
+
+int mdbx_get_equal_or_great(const MDBX_txn *txn, MDBX_dbi dbi, MDBX_val *key, MDBX_val *data) {
+  if (unlikely(!key || !data))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cursor_couple_t cx;
+  rc = cursor_init(&cx.outer, txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  return LOG_IFERR(cursor_ops(&cx.outer, key, data, MDBX_SET_LOWERBOUND));
+}
+
+int mdbx_get_ex(const MDBX_txn *txn, MDBX_dbi dbi, MDBX_val *key, MDBX_val *data, size_t *values_count) {
+  DKBUF_DEBUG;
+  DEBUG("===> get db %u key [%s]", dbi, DKEY_DEBUG(key));
+
+  if (unlikely(!key || !data))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cursor_couple_t cx;
+  rc = cursor_init(&cx.outer, txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  rc = cursor_seek(&cx.outer, key, data, MDBX_SET_KEY).err;
+  if (unlikely(rc != MDBX_SUCCESS)) {
+    if (values_count)
+      *values_count = 0;
+    return LOG_IFERR(rc);
+  }
+
+  if (values_count) {
+    *values_count = 1;
+    if (inner_pointed(&cx.outer))
+      *values_count =
+          (sizeof(*values_count) >= sizeof(cx.inner.nested_tree.items) || cx.inner.nested_tree.items <= PTRDIFF_MAX)
+              ? (size_t)cx.inner.nested_tree.items
+              : PTRDIFF_MAX;
+  }
+  return MDBX_SUCCESS;
+}
+
+/*----------------------------------------------------------------------------*/
+
+int mdbx_canary_put(MDBX_txn *txn, const MDBX_canary *canary) {
+  int rc = check_txn_rw(txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (likely(canary)) {
+    if (txn->canary.x == canary->x && txn->canary.y == canary->y && txn->canary.z == canary->z)
+      return MDBX_SUCCESS;
+    txn->canary.x = canary->x;
+    txn->canary.y = canary->y;
+    txn->canary.z = canary->z;
+  }
+  txn->canary.v = txn->txnid;
+  txn->flags |= MDBX_TXN_DIRTY;
+
+  return MDBX_SUCCESS;
+}
+
+/* Функция сообщает находится ли указанный адрес в "грязной" странице у
+ * заданной пишущей транзакции. В конечном счете это позволяет избавиться от
+ * лишнего копирования данных из НЕ-грязных страниц.
+ *
+ * "Грязные" страницы - это те, которые уже были изменены в ходе пишущей
+ * транзакции. Соответственно, какие-либо дальнейшие изменения могут привести
+ * к перезаписи таких страниц. Поэтому все функции, выполняющие изменения, в
+ * качестве аргументов НЕ должны получать указатели на данные в таких
+ * страницах. В свою очередь "НЕ грязные" страницы перед модификацией будут
+ * скопированы.
+ *
+ * Другими словами, данные из "грязных" страниц должны быть либо скопированы
+ * перед передачей в качестве аргументов для дальнейших модификаций, либо
+ * отвергнуты на стадии проверки корректности аргументов.
+ *
+ * Таким образом, функция позволяет как избавится от лишнего копирования,
+ * так и выполнить более полную проверку аргументов.
+ *
+ * ВАЖНО: Передаваемый указатель должен указывать на начало данных. Только
+ * так гарантируется что актуальный заголовок страницы будет физически
+ * расположен в той-же странице памяти, в том числе для многостраничных
+ * P_LARGE страниц с длинными данными. */
+int mdbx_is_dirty(const MDBX_txn *txn, const void *ptr) {
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED - MDBX_TXN_PARKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  const MDBX_env *env = txn->env;
+  const ptrdiff_t offset = ptr_dist(ptr, env->dxb_mmap.base);
+  if (offset >= 0) {
+    const pgno_t pgno = bytes2pgno(env, offset);
+    if (likely(pgno < txn->geo.first_unallocated)) {
+      const page_t *page = pgno2page(env, pgno);
+      if (unlikely(page->pgno != pgno || (page->flags & P_ILL_BITS) != 0)) {
+        /* The ptr pointed into middle of a large page,
+         * not to the beginning of a data. */
+        return LOG_IFERR(MDBX_EINVAL);
+      }
+      return ((txn->flags & MDBX_TXN_RDONLY) || !is_modifable(txn, page)) ? MDBX_RESULT_FALSE : MDBX_RESULT_TRUE;
+    }
+    if ((size_t)offset < env->dxb_mmap.limit) {
+      /* Указатель адресует что-то в пределах mmap, но за границей
+       * распределенных страниц. Такое может случится если mdbx_is_dirty()
+       * вызывается после операции, в ходе которой грязная страница была
+       * возвращена в нераспределенное пространство. */
+      return (txn->flags & MDBX_TXN_RDONLY) ? LOG_IFERR(MDBX_EINVAL) : MDBX_RESULT_TRUE;
+    }
+  }
+
+  /* Страница вне используемого mmap-диапазона, т.е. либо в функцию был
+   * передан некорректный адрес, либо адрес в теневой странице, которая была
+   * выделена посредством malloc().
+   *
+   * Для режима MDBX_WRITE_MAP режима страница однозначно "не грязная",
+   * а для режимов без MDBX_WRITE_MAP однозначно "не чистая". */
+  return (txn->flags & (MDBX_WRITEMAP | MDBX_TXN_RDONLY)) ? LOG_IFERR(MDBX_EINVAL) : MDBX_RESULT_TRUE;
+}
+
+int mdbx_del(MDBX_txn *txn, MDBX_dbi dbi, const MDBX_val *key, const MDBX_val *data) {
+  if (unlikely(!key))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  if (unlikely(dbi <= FREE_DBI))
+    return LOG_IFERR(MDBX_BAD_DBI);
+
+  int rc = check_txn_rw(txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cursor_couple_t cx;
+  rc = cursor_init(&cx.outer, txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  MDBX_val proxy;
+  MDBX_cursor_op op = MDBX_SET;
+  unsigned flags = MDBX_ALLDUPS;
+  if (data) {
+    proxy = *data;
+    data = &proxy;
+    op = MDBX_GET_BOTH;
+    flags = 0;
+  }
+  rc = cursor_seek(&cx.outer, (MDBX_val *)key, (MDBX_val *)data, op).err;
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cx.outer.next = txn->cursors[dbi];
+  txn->cursors[dbi] = &cx.outer;
+  rc = cursor_del(&cx.outer, flags);
+  txn->cursors[dbi] = cx.outer.next;
+  return LOG_IFERR(rc);
+}
+
+int mdbx_put(MDBX_txn *txn, MDBX_dbi dbi, const MDBX_val *key, MDBX_val *data, MDBX_put_flags_t flags) {
+  if (unlikely(!key || !data))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  if (unlikely(dbi <= FREE_DBI))
+    return LOG_IFERR(MDBX_BAD_DBI);
+
+  if (unlikely(flags & ~(MDBX_NOOVERWRITE | MDBX_NODUPDATA | MDBX_ALLDUPS | MDBX_ALLDUPS | MDBX_RESERVE | MDBX_APPEND |
+                         MDBX_APPENDDUP | MDBX_CURRENT | MDBX_MULTIPLE)))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = check_txn_rw(txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cursor_couple_t cx;
+  rc = cursor_init(&cx.outer, txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(flags & MDBX_MULTIPLE)) {
+    rc = cursor_check_multiple(&cx.outer, key, data, flags);
+    if (unlikely(rc != MDBX_SUCCESS))
+      return LOG_IFERR(rc);
+  }
+
+  if (flags & MDBX_RESERVE) {
+    if (unlikely(cx.outer.tree->flags & (MDBX_DUPSORT | MDBX_REVERSEDUP | MDBX_INTEGERDUP | MDBX_DUPFIXED)))
+      return LOG_IFERR(MDBX_INCOMPATIBLE);
+    data->iov_base = nullptr;
+  }
+
+  cx.outer.next = txn->cursors[dbi];
+  txn->cursors[dbi] = &cx.outer;
+
+  /* LY: support for update (explicit overwrite) */
+  if (flags & MDBX_CURRENT) {
+    rc = cursor_seek(&cx.outer, (MDBX_val *)key, nullptr, MDBX_SET).err;
+    if (likely(rc == MDBX_SUCCESS) && (txn->dbs[dbi].flags & MDBX_DUPSORT) && (flags & MDBX_ALLDUPS) == 0) {
+      /* LY: allows update (explicit overwrite) only for unique keys */
+      node_t *node = page_node(cx.outer.pg[cx.outer.top], cx.outer.ki[cx.outer.top]);
+      if (node_flags(node) & N_DUP) {
+        tASSERT(txn, inner_pointed(&cx.outer) && cx.outer.subcur->nested_tree.items > 1);
+        rc = MDBX_EMULTIVAL;
+        if ((flags & MDBX_NOOVERWRITE) == 0) {
+          flags -= MDBX_CURRENT;
+          rc = cursor_del(&cx.outer, MDBX_ALLDUPS);
+        }
+      }
+    }
+  }
+
+  if (likely(rc == MDBX_SUCCESS))
+    rc = cursor_put_checklen(&cx.outer, key, data, flags);
+  txn->cursors[dbi] = cx.outer.next;
+
+  return LOG_IFERR(rc);
+}
+
+//------------------------------------------------------------------------------
+
+/* Позволяет обновить или удалить существующую запись с получением
+ * в old_data предыдущего значения данных. При этом если new_data равен
+ * нулю, то выполняется удаление, иначе обновление/вставка.
+ *
+ * Текущее значение может находиться в уже измененной (грязной) странице.
+ * В этом случае страница будет перезаписана при обновлении, а само старое
+ * значение утрачено. Поэтому исходно в old_data должен быть передан
+ * дополнительный буфер для копирования старого значения.
+ * Если переданный буфер слишком мал, то функция вернет -1, установив
+ * old_data->iov_len в соответствующее значение.
+ *
+ * Для не-уникальных ключей также возможен второй сценарий использования,
+ * когда посредством old_data из записей с одинаковым ключом для
+ * удаления/обновления выбирается конкретная. Для выбора этого сценария
+ * во flags следует одновременно указать MDBX_CURRENT и MDBX_NOOVERWRITE.
+ * Именно эта комбинация выбрана, так как она лишена смысла, и этим позволяет
+ * идентифицировать запрос такого сценария.
+ *
+ * Функция может быть замещена соответствующими операциями с курсорами
+ * после двух доработок (TODO):
+ *  - внешняя аллокация курсоров, в том числе на стеке (без malloc).
+ *  - получения dirty-статуса страницы по адресу (знать о MUTABLE/WRITEABLE).
+ */
+
+int mdbx_replace_ex(MDBX_txn *txn, MDBX_dbi dbi, const MDBX_val *key, MDBX_val *new_data, MDBX_val *old_data,
+                    MDBX_put_flags_t flags, MDBX_preserve_func preserver, void *preserver_context) {
+  if (unlikely(!key || !old_data || old_data == new_data))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  if (unlikely(old_data->iov_base == nullptr && old_data->iov_len))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  if (unlikely(new_data == nullptr && (flags & (MDBX_CURRENT | MDBX_RESERVE)) != MDBX_CURRENT))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  if (unlikely(dbi <= FREE_DBI))
+    return LOG_IFERR(MDBX_BAD_DBI);
+
+  if (unlikely(flags & ~(MDBX_NOOVERWRITE | MDBX_NODUPDATA | MDBX_ALLDUPS | MDBX_RESERVE | MDBX_APPEND |
+                         MDBX_APPENDDUP | MDBX_CURRENT)))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = check_txn_rw(txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  cursor_couple_t cx;
+  rc = cursor_init(&cx.outer, txn, dbi);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+  cx.outer.next = txn->cursors[dbi];
+  txn->cursors[dbi] = &cx.outer;
+
+  MDBX_val present_key = *key;
+  if (F_ISSET(flags, MDBX_CURRENT | MDBX_NOOVERWRITE)) {
+    /* в old_data значение для выбора конкретного дубликата */
+    if (unlikely(!(txn->dbs[dbi].flags & MDBX_DUPSORT))) {
+      rc = MDBX_EINVAL;
+      goto bailout;
+    }
+
+    /* убираем лишний бит, он был признаком запрошенного режима */
+    flags -= MDBX_NOOVERWRITE;
+
+    rc = cursor_seek(&cx.outer, &present_key, old_data, MDBX_GET_BOTH).err;
+    if (rc != MDBX_SUCCESS)
+      goto bailout;
+  } else {
+    /* в old_data буфер для сохранения предыдущего значения */
+    if (unlikely(new_data && old_data->iov_base == new_data->iov_base))
+      return LOG_IFERR(MDBX_EINVAL);
+    MDBX_val present_data;
+    rc = cursor_seek(&cx.outer, &present_key, &present_data, MDBX_SET_KEY).err;
+    if (unlikely(rc != MDBX_SUCCESS)) {
+      old_data->iov_base = nullptr;
+      old_data->iov_len = 0;
+      if (rc != MDBX_NOTFOUND || (flags & MDBX_CURRENT))
+        goto bailout;
+    } else if (flags & MDBX_NOOVERWRITE) {
+      rc = MDBX_KEYEXIST;
+      *old_data = present_data;
+      goto bailout;
+    } else {
+      page_t *page = cx.outer.pg[cx.outer.top];
+      if (txn->dbs[dbi].flags & MDBX_DUPSORT) {
+        if (flags & MDBX_CURRENT) {
+          /* disallow update/delete for multi-values */
+          node_t *node = page_node(page, cx.outer.ki[cx.outer.top]);
+          if (node_flags(node) & N_DUP) {
+            tASSERT(txn, inner_pointed(&cx.outer) && cx.outer.subcur->nested_tree.items > 1);
+            if (cx.outer.subcur->nested_tree.items > 1) {
+              rc = MDBX_EMULTIVAL;
+              goto bailout;
+            }
+          }
+          /* В LMDB флажок MDBX_CURRENT здесь приведет
+           * к замене данных без учета MDBX_DUPSORT сортировки,
+           * но здесь это в любом случае допустимо, так как мы
+           * проверили что для ключа есть только одно значение. */
+        }
+      }
+
+      if (is_modifable(txn, page)) {
+        if (new_data && eq_fast(&present_data, new_data)) {
+          /* если данные совпадают, то ничего делать не надо */
+          *old_data = *new_data;
+          goto bailout;
+        }
+        rc = preserver ? preserver(preserver_context, old_data, present_data.iov_base, present_data.iov_len)
+                       : MDBX_SUCCESS;
+        if (unlikely(rc != MDBX_SUCCESS))
+          goto bailout;
+      } else {
+        *old_data = present_data;
+      }
+      flags |= MDBX_CURRENT;
+    }
+  }
+
+  if (likely(new_data))
+    rc = cursor_put_checklen(&cx.outer, key, new_data, flags);
+  else
+    rc = cursor_del(&cx.outer, flags & MDBX_ALLDUPS);
+
+bailout:
+  txn->cursors[dbi] = cx.outer.next;
+  return LOG_IFERR(rc);
+}
+
+static int default_value_preserver(void *context, MDBX_val *target, const void *src, size_t bytes) {
+  (void)context;
+  if (unlikely(target->iov_len < bytes)) {
+    target->iov_base = nullptr;
+    target->iov_len = bytes;
+    return MDBX_RESULT_TRUE;
+  }
+  memcpy(target->iov_base, src, target->iov_len = bytes);
+  return MDBX_SUCCESS;
+}
+
+int mdbx_replace(MDBX_txn *txn, MDBX_dbi dbi, const MDBX_val *key, MDBX_val *new_data, MDBX_val *old_data,
+                 MDBX_put_flags_t flags) {
+  return mdbx_replace_ex(txn, dbi, key, new_data, old_data, flags, default_value_preserver, nullptr);
+}

src/api-txn.c

@@ -0,0 +1,540 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+#ifdef __SANITIZE_THREAD__
+/* LY: avoid tsan-trap by txn, mm_last_pg and geo.first_unallocated */
+__attribute__((__no_sanitize_thread__, __noinline__))
+#endif
+int mdbx_txn_straggler(const MDBX_txn *txn, int *percent)
+{
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED - MDBX_TXN_PARKED);
+  if (likely(rc == MDBX_SUCCESS))
+    rc = check_env(txn->env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR((rc > 0) ? -rc : rc);
+
+  if (unlikely((txn->flags & MDBX_TXN_RDONLY) == 0)) {
+    if (percent)
+      *percent = (int)((txn->geo.first_unallocated * UINT64_C(100) + txn->geo.end_pgno / 2) / txn->geo.end_pgno);
+    return 0;
+  }
+
+  txnid_t lag;
+  troika_t troika = meta_tap(txn->env);
+  do {
+    const meta_ptr_t head = meta_recent(txn->env, &troika);
+    if (percent) {
+      const pgno_t maxpg = head.ptr_v->geometry.now;
+      *percent = (int)((head.ptr_v->geometry.first_unallocated * UINT64_C(100) + maxpg / 2) / maxpg);
+    }
+    lag = (head.txnid - txn->txnid) / xMDBX_TXNID_STEP;
+  } while (unlikely(meta_should_retry(txn->env, &troika)));
+
+  return (lag > INT_MAX) ? INT_MAX : (int)lag;
+}
+
+MDBX_env *mdbx_txn_env(const MDBX_txn *txn) {
+  if (unlikely(!txn || txn->signature != txn_signature || txn->env->signature.weak != env_signature))
+    return nullptr;
+  return txn->env;
+}
+
+uint64_t mdbx_txn_id(const MDBX_txn *txn) {
+  if (unlikely(!txn || txn->signature != txn_signature))
+    return 0;
+  return txn->txnid;
+}
+
+MDBX_txn_flags_t mdbx_txn_flags(const MDBX_txn *txn) {
+  STATIC_ASSERT(
+      (MDBX_TXN_INVALID & (MDBX_TXN_FINISHED | MDBX_TXN_ERROR | MDBX_TXN_DIRTY | MDBX_TXN_SPILLS | MDBX_TXN_HAS_CHILD |
+                           txn_gc_drained | txn_shrink_allowed | txn_rw_begin_flags | txn_ro_begin_flags)) == 0);
+  if (unlikely(!txn || txn->signature != txn_signature))
+    return MDBX_TXN_INVALID;
+  assert(0 == (int)(txn->flags & MDBX_TXN_INVALID));
+
+  MDBX_txn_flags_t flags = txn->flags;
+  if (F_ISSET(flags, MDBX_TXN_PARKED | MDBX_TXN_RDONLY) && txn->ro.slot &&
+      safe64_read(&txn->ro.slot->tid) == MDBX_TID_TXN_OUSTED)
+    flags |= MDBX_TXN_OUSTED;
+  return flags;
+}
+
+int mdbx_txn_reset(MDBX_txn *txn) {
+  int rc = check_txn(txn, 0);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  rc = check_env(txn->env, false);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  /* This call is only valid for read-only txns */
+  if (unlikely((txn->flags & MDBX_TXN_RDONLY) == 0))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  /* LY: don't close DBI-handles */
+  rc = txn_end(txn, TXN_END_RESET | TXN_END_UPDATE);
+  if (rc == MDBX_SUCCESS) {
+    tASSERT(txn, txn->signature == txn_signature);
+    tASSERT(txn, txn->owner == 0);
+  }
+  return LOG_IFERR(rc);
+}
+
+int mdbx_txn_break(MDBX_txn *txn) {
+  do {
+    int rc = check_txn(txn, 0);
+    if (unlikely(rc != MDBX_SUCCESS))
+      return LOG_IFERR(rc);
+    txn->flags |= MDBX_TXN_ERROR;
+    txn = txn->nested;
+  } while (txn);
+  return MDBX_SUCCESS;
+}
+
+int mdbx_txn_abort(MDBX_txn *txn) {
+  int rc = check_txn(txn, 0);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  rc = check_env(txn->env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+#if MDBX_TXN_CHECKOWNER
+  if ((txn->flags & (MDBX_TXN_RDONLY | MDBX_NOSTICKYTHREADS)) == MDBX_NOSTICKYTHREADS &&
+      unlikely(txn->owner != osal_thread_self())) {
+    mdbx_txn_break(txn);
+    return LOG_IFERR(MDBX_THREAD_MISMATCH);
+  }
+#endif /* MDBX_TXN_CHECKOWNER */
+
+  return LOG_IFERR(txn_abort(txn));
+}
+
+int mdbx_txn_park(MDBX_txn *txn, bool autounpark) {
+  STATIC_ASSERT(MDBX_TXN_BLOCKED > MDBX_TXN_ERROR);
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED - MDBX_TXN_ERROR);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  rc = check_env(txn->env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely((txn->flags & MDBX_TXN_RDONLY) == 0))
+    return LOG_IFERR(MDBX_TXN_INVALID);
+
+  if (unlikely((txn->flags & MDBX_TXN_ERROR))) {
+    rc = txn_end(txn, TXN_END_RESET | TXN_END_UPDATE);
+    return LOG_IFERR(rc ? rc : MDBX_OUSTED);
+  }
+
+  return LOG_IFERR(txn_ro_park(txn, autounpark));
+}
+
+int mdbx_txn_unpark(MDBX_txn *txn, bool restart_if_ousted) {
+  STATIC_ASSERT(MDBX_TXN_BLOCKED > MDBX_TXN_PARKED + MDBX_TXN_ERROR);
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED - MDBX_TXN_PARKED - MDBX_TXN_ERROR);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  rc = check_env(txn->env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(!F_ISSET(txn->flags, MDBX_TXN_RDONLY | MDBX_TXN_PARKED)))
+    return MDBX_SUCCESS;
+
+  rc = txn_ro_unpark(txn);
+  if (likely(rc != MDBX_OUSTED) || !restart_if_ousted)
+    return LOG_IFERR(rc);
+
+  tASSERT(txn, txn->flags & MDBX_TXN_FINISHED);
+  rc = txn_renew(txn, MDBX_TXN_RDONLY);
+  return (rc == MDBX_SUCCESS) ? MDBX_RESULT_TRUE : LOG_IFERR(rc);
+}
+
+int mdbx_txn_renew(MDBX_txn *txn) {
+  int rc = check_txn(txn, 0);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  rc = check_env(txn->env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely((txn->flags & MDBX_TXN_RDONLY) == 0))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  if (unlikely(txn->owner != 0 || !(txn->flags & MDBX_TXN_FINISHED))) {
+    rc = mdbx_txn_reset(txn);
+    if (unlikely(rc != MDBX_SUCCESS))
+      return rc;
+  }
+
+  rc = txn_renew(txn, MDBX_TXN_RDONLY);
+  if (rc == MDBX_SUCCESS) {
+    tASSERT(txn, txn->owner == (txn->flags & MDBX_NOSTICKYTHREADS) ? 0 : osal_thread_self());
+    DEBUG("renew txn %" PRIaTXN "%c %p on env %p, root page %" PRIaPGNO "/%" PRIaPGNO, txn->txnid,
+          (txn->flags & MDBX_TXN_RDONLY) ? 'r' : 'w', (void *)txn, (void *)txn->env, txn->dbs[MAIN_DBI].root,
+          txn->dbs[FREE_DBI].root);
+  }
+  return LOG_IFERR(rc);
+}
+
+int mdbx_txn_set_userctx(MDBX_txn *txn, void *ctx) {
+  int rc = check_txn(txn, 0);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  txn->userctx = ctx;
+  return MDBX_SUCCESS;
+}
+
+void *mdbx_txn_get_userctx(const MDBX_txn *txn) { return check_txn(txn, MDBX_TXN_FINISHED) ? nullptr : txn->userctx; }
+
+int mdbx_txn_begin_ex(MDBX_env *env, MDBX_txn *parent, MDBX_txn_flags_t flags, MDBX_txn **ret, void *context) {
+  if (unlikely(!ret))
+    return LOG_IFERR(MDBX_EINVAL);
+  *ret = nullptr;
+
+  if (unlikely((flags & ~txn_rw_begin_flags) && (parent || (flags & ~txn_ro_begin_flags))))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  int rc = check_env(env, true);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(env->flags & MDBX_RDONLY & ~flags)) /* write txn in RDONLY env */
+    return LOG_IFERR(MDBX_EACCESS);
+
+  /* Reuse preallocated write txn. However, do not touch it until
+   * txn_renew() succeeds, since it currently may be active. */
+  MDBX_txn *txn = nullptr;
+  if (parent) {
+    /* Nested transactions: Max 1 child, write txns only, no writemap */
+    rc = check_txn(parent, MDBX_TXN_BLOCKED - MDBX_TXN_PARKED);
+    if (unlikely(rc != MDBX_SUCCESS))
+      return LOG_IFERR(rc);
+
+    if (unlikely(parent->flags & (MDBX_TXN_RDONLY | MDBX_WRITEMAP))) {
+      rc = MDBX_BAD_TXN;
+      if ((parent->flags & MDBX_TXN_RDONLY) == 0) {
+        ERROR("%s mode is incompatible with nested transactions", "MDBX_WRITEMAP");
+        rc = MDBX_INCOMPATIBLE;
+      }
+      return LOG_IFERR(rc);
+    }
+    if (unlikely(parent->env != env))
+      return LOG_IFERR(MDBX_BAD_TXN);
+
+    flags |= parent->flags & (txn_rw_begin_flags | MDBX_TXN_SPILLS | MDBX_NOSTICKYTHREADS | MDBX_WRITEMAP);
+    rc = txn_nested_create(parent, flags);
+    txn = parent->nested;
+    if (unlikely(rc != MDBX_SUCCESS)) {
+      int err = txn_end(txn, TXN_END_FAIL_BEGIN_NESTED);
+      return err ? err : rc;
+    }
+    if (AUDIT_ENABLED() && ASSERT_ENABLED()) {
+      txn->signature = txn_signature;
+      tASSERT(txn, audit_ex(txn, 0, false) == 0);
+    }
+  } else {
+    txn = env->basal_txn;
+    if (flags & MDBX_TXN_RDONLY) {
+      txn = txn_alloc(flags, env);
+      if (unlikely(!txn))
+        return LOG_IFERR(MDBX_ENOMEM);
+    }
+    rc = txn_renew(txn, flags);
+    if (unlikely(rc != MDBX_SUCCESS)) {
+      if (txn != env->basal_txn)
+        osal_free(txn);
+      return LOG_IFERR(rc);
+    }
+  }
+
+  if (flags & (MDBX_TXN_RDONLY_PREPARE - MDBX_TXN_RDONLY))
+    eASSERT(env, txn->flags == (MDBX_TXN_RDONLY | MDBX_TXN_FINISHED));
+  else if (flags & MDBX_TXN_RDONLY)
+    eASSERT(env, (txn->flags & ~(MDBX_NOSTICKYTHREADS | MDBX_TXN_RDONLY | MDBX_WRITEMAP |
+                                 /* Win32: SRWL flag */ txn_shrink_allowed)) == 0);
+  else {
+    eASSERT(env, (txn->flags & ~(MDBX_NOSTICKYTHREADS | MDBX_WRITEMAP | txn_shrink_allowed | txn_may_have_cursors |
+                                 MDBX_NOMETASYNC | MDBX_SAFE_NOSYNC | MDBX_TXN_SPILLS)) == 0);
+    assert(!txn->wr.spilled.list && !txn->wr.spilled.least_removed);
+  }
+  txn->signature = txn_signature;
+  txn->userctx = context;
+  *ret = txn;
+  DEBUG("begin txn %" PRIaTXN "%c %p on env %p, root page %" PRIaPGNO "/%" PRIaPGNO, txn->txnid,
+        (flags & MDBX_TXN_RDONLY) ? 'r' : 'w', (void *)txn, (void *)env, txn->dbs[MAIN_DBI].root,
+        txn->dbs[FREE_DBI].root);
+  return MDBX_SUCCESS;
+}
+
+static void latency_gcprof(MDBX_commit_latency *latency, const MDBX_txn *txn) {
+  MDBX_env *const env = txn->env;
+  if (latency && likely(env->lck) && MDBX_ENABLE_PROFGC) {
+    pgop_stat_t *const ptr = &env->lck->pgops;
+    latency->gc_prof.work_counter = ptr->gc_prof.work.spe_counter;
+    latency->gc_prof.work_rtime_monotonic = osal_monotime_to_16dot16(ptr->gc_prof.work.rtime_monotonic);
+    latency->gc_prof.work_xtime_cpu = osal_monotime_to_16dot16(ptr->gc_prof.work.xtime_cpu);
+    latency->gc_prof.work_rsteps = ptr->gc_prof.work.rsteps;
+    latency->gc_prof.work_xpages = ptr->gc_prof.work.xpages;
+    latency->gc_prof.work_majflt = ptr->gc_prof.work.majflt;
+
+    latency->gc_prof.self_counter = ptr->gc_prof.self.spe_counter;
+    latency->gc_prof.self_rtime_monotonic = osal_monotime_to_16dot16(ptr->gc_prof.self.rtime_monotonic);
+    latency->gc_prof.self_xtime_cpu = osal_monotime_to_16dot16(ptr->gc_prof.self.xtime_cpu);
+    latency->gc_prof.self_rsteps = ptr->gc_prof.self.rsteps;
+    latency->gc_prof.self_xpages = ptr->gc_prof.self.xpages;
+    latency->gc_prof.self_majflt = ptr->gc_prof.self.majflt;
+
+    latency->gc_prof.wloops = ptr->gc_prof.wloops;
+    latency->gc_prof.coalescences = ptr->gc_prof.coalescences;
+    latency->gc_prof.wipes = ptr->gc_prof.wipes;
+    latency->gc_prof.flushes = ptr->gc_prof.flushes;
+    latency->gc_prof.kicks = ptr->gc_prof.kicks;
+
+    latency->gc_prof.pnl_merge_work.time = osal_monotime_to_16dot16(ptr->gc_prof.work.pnl_merge.time);
+    latency->gc_prof.pnl_merge_work.calls = ptr->gc_prof.work.pnl_merge.calls;
+    latency->gc_prof.pnl_merge_work.volume = ptr->gc_prof.work.pnl_merge.volume;
+    latency->gc_prof.pnl_merge_self.time = osal_monotime_to_16dot16(ptr->gc_prof.self.pnl_merge.time);
+    latency->gc_prof.pnl_merge_self.calls = ptr->gc_prof.self.pnl_merge.calls;
+    latency->gc_prof.pnl_merge_self.volume = ptr->gc_prof.self.pnl_merge.volume;
+
+    if (txn == env->basal_txn)
+      memset(&ptr->gc_prof, 0, sizeof(ptr->gc_prof));
+  }
+}
+
+static void latency_init(MDBX_commit_latency *latency, struct commit_timestamp *ts) {
+  ts->start = 0;
+  ts->gc_cpu = 0;
+  if (latency) {
+    ts->start = osal_monotime();
+    memset(latency, 0, sizeof(*latency));
+  }
+  ts->prep = ts->gc = ts->audit = ts->write = ts->sync = ts->start;
+}
+
+static void latency_done(MDBX_commit_latency *latency, struct commit_timestamp *ts) {
+  if (latency) {
+    latency->preparation = (ts->prep > ts->start) ? osal_monotime_to_16dot16(ts->prep - ts->start) : 0;
+    latency->gc_wallclock = (ts->gc > ts->prep) ? osal_monotime_to_16dot16(ts->gc - ts->prep) : 0;
+    latency->gc_cputime = ts->gc_cpu ? osal_monotime_to_16dot16(ts->gc_cpu) : 0;
+    latency->audit = (ts->audit > ts->gc) ? osal_monotime_to_16dot16(ts->audit - ts->gc) : 0;
+    latency->write = (ts->write > ts->audit) ? osal_monotime_to_16dot16(ts->write - ts->audit) : 0;
+    latency->sync = (ts->sync > ts->write) ? osal_monotime_to_16dot16(ts->sync - ts->write) : 0;
+    const uint64_t ts_end = osal_monotime();
+    latency->ending = (ts_end > ts->sync) ? osal_monotime_to_16dot16(ts_end - ts->sync) : 0;
+    latency->whole = osal_monotime_to_16dot16_noUnderflow(ts_end - ts->start);
+  }
+}
+
+int mdbx_txn_commit_ex(MDBX_txn *txn, MDBX_commit_latency *latency) {
+  STATIC_ASSERT(MDBX_TXN_FINISHED == MDBX_TXN_BLOCKED - MDBX_TXN_HAS_CHILD - MDBX_TXN_ERROR - MDBX_TXN_PARKED);
+
+  struct commit_timestamp ts;
+  latency_init(latency, &ts);
+
+  int rc = check_txn(txn, MDBX_TXN_FINISHED);
+  if (unlikely(rc != MDBX_SUCCESS)) {
+    if (rc == MDBX_BAD_TXN && F_ISSET(txn->flags, MDBX_TXN_FINISHED | MDBX_TXN_RDONLY)) {
+      rc = MDBX_RESULT_TRUE;
+      goto fail;
+    }
+    return LOG_IFERR(rc);
+  }
+
+  MDBX_env *const env = txn->env;
+  if (MDBX_ENV_CHECKPID && unlikely(env->pid != osal_getpid())) {
+    env->flags |= ENV_FATAL_ERROR;
+    rc = MDBX_PANIC;
+    return LOG_IFERR(rc);
+  }
+
+  if (txn->flags & MDBX_TXN_RDONLY) {
+    if (unlikely(txn->parent || (txn->flags & MDBX_TXN_HAS_CHILD) || txn == env->txn || txn == env->basal_txn)) {
+      ERROR("attempt to commit %s txn %p", "strange read-only", (void *)txn);
+      return MDBX_PROBLEM;
+    }
+    latency_gcprof(latency, txn);
+    rc = (txn->flags & MDBX_TXN_ERROR) ? MDBX_RESULT_TRUE : MDBX_SUCCESS;
+    txn_end(txn, TXN_END_PURE_COMMIT | TXN_END_UPDATE | TXN_END_SLOT | TXN_END_FREE);
+    goto done;
+  }
+
+#if MDBX_TXN_CHECKOWNER
+  if ((txn->flags & MDBX_NOSTICKYTHREADS) && txn == env->basal_txn && unlikely(txn->owner != osal_thread_self())) {
+    txn->flags |= MDBX_TXN_ERROR;
+    rc = MDBX_THREAD_MISMATCH;
+    return LOG_IFERR(rc);
+  }
+#endif /* MDBX_TXN_CHECKOWNER */
+
+  if (unlikely(txn->flags & MDBX_TXN_ERROR)) {
+    rc = MDBX_RESULT_TRUE;
+  fail:
+    latency_gcprof(latency, txn);
+    int err = txn_abort(txn);
+    if (unlikely(err != MDBX_SUCCESS))
+      rc = err;
+    goto done;
+  }
+
+  if (txn->nested) {
+    rc = mdbx_txn_commit_ex(txn->nested, nullptr);
+    tASSERT(txn, txn->nested == nullptr);
+    if (unlikely(rc != MDBX_SUCCESS))
+      goto fail;
+  }
+
+  if (unlikely(txn != env->txn)) {
+    ERROR("attempt to commit %s txn %p", "unknown", (void *)txn);
+    return MDBX_EINVAL;
+  }
+
+  if (txn->parent) {
+    if (unlikely(txn->parent->nested != txn || txn->parent->env != env)) {
+      ERROR("attempt to commit %s txn %p", "strange nested", (void *)txn);
+      return MDBX_PROBLEM;
+    }
+
+    latency_gcprof(latency, txn);
+    rc = txn_nested_join(txn, latency ? &ts : nullptr);
+    goto done;
+  }
+
+  rc = txn_basal_commit(txn, latency ? &ts : nullptr);
+  latency_gcprof(latency, txn);
+  int end = TXN_END_COMMITTED | TXN_END_UPDATE;
+  if (unlikely(rc != MDBX_SUCCESS)) {
+    end = TXN_END_ABORT;
+    if (rc == MDBX_RESULT_TRUE) {
+      end = TXN_END_PURE_COMMIT | TXN_END_UPDATE;
+      rc = MDBX_NOSUCCESS_PURE_COMMIT ? MDBX_RESULT_TRUE : MDBX_SUCCESS;
+    }
+  }
+  int err = txn_end(txn, end);
+  if (unlikely(err != MDBX_SUCCESS))
+    rc = err;
+
+done:
+  latency_done(latency, &ts);
+  return LOG_IFERR(rc);
+}
+
+int mdbx_txn_info(const MDBX_txn *txn, MDBX_txn_info *info, bool scan_rlt) {
+  int rc = check_txn(txn, MDBX_TXN_FINISHED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return LOG_IFERR(rc);
+
+  if (unlikely(!info))
+    return LOG_IFERR(MDBX_EINVAL);
+
+  MDBX_env *const env = txn->env;
+#if MDBX_ENV_CHECKPID
+  if (unlikely(env->pid != osal_getpid())) {
+    env->flags |= ENV_FATAL_ERROR;
+    return LOG_IFERR(MDBX_PANIC);
+  }
+#endif /* MDBX_ENV_CHECKPID */
+
+  info->txn_id = txn->txnid;
+  info->txn_space_used = pgno2bytes(env, txn->geo.first_unallocated);
+
+  if (txn->flags & MDBX_TXN_RDONLY) {
+    meta_ptr_t head;
+    uint64_t head_retired;
+    troika_t troika = meta_tap(env);
+    do {
+      /* fetch info from volatile head */
+      head = meta_recent(env, &troika);
+      head_retired = unaligned_peek_u64_volatile(4, head.ptr_v->pages_retired);
+      info->txn_space_limit_soft = pgno2bytes(env, head.ptr_v->geometry.now);
+      info->txn_space_limit_hard = pgno2bytes(env, head.ptr_v->geometry.upper);
+      info->txn_space_leftover = pgno2bytes(env, head.ptr_v->geometry.now - head.ptr_v->geometry.first_unallocated);
+    } while (unlikely(meta_should_retry(env, &troika)));
+
+    info->txn_reader_lag = head.txnid - info->txn_id;
+    info->txn_space_dirty = info->txn_space_retired = 0;
+    uint64_t reader_snapshot_pages_retired = 0;
+    if (txn->ro.slot &&
+        ((txn->flags & MDBX_TXN_PARKED) == 0 || safe64_read(&txn->ro.slot->tid) != MDBX_TID_TXN_OUSTED) &&
+        head_retired >
+            (reader_snapshot_pages_retired = atomic_load64(&txn->ro.slot->snapshot_pages_retired, mo_Relaxed))) {
+      info->txn_space_dirty = info->txn_space_retired =
+          pgno2bytes(env, (pgno_t)(head_retired - reader_snapshot_pages_retired));
+
+      size_t retired_next_reader = 0;
+      lck_t *const lck = env->lck_mmap.lck;
+      if (scan_rlt && info->txn_reader_lag > 1 && lck) {
+        /* find next more recent reader */
+        txnid_t next_reader = head.txnid;
+        const size_t snap_nreaders = atomic_load32(&lck->rdt_length, mo_AcquireRelease);
+        for (size_t i = 0; i < snap_nreaders; ++i) {
+        retry:
+          if (atomic_load32(&lck->rdt[i].pid, mo_AcquireRelease)) {
+            jitter4testing(true);
+            const uint64_t snap_tid = safe64_read(&lck->rdt[i].tid);
+            const txnid_t snap_txnid = safe64_read(&lck->rdt[i].txnid);
+            const uint64_t snap_retired = atomic_load64(&lck->rdt[i].snapshot_pages_retired, mo_AcquireRelease);
+            if (unlikely(snap_retired != atomic_load64(&lck->rdt[i].snapshot_pages_retired, mo_Relaxed)) ||
+                snap_txnid != safe64_read(&lck->rdt[i].txnid) || snap_tid != safe64_read(&lck->rdt[i].tid))
+              goto retry;
+            if (snap_txnid <= txn->txnid) {
+              retired_next_reader = 0;
+              break;
+            }
+            if (snap_txnid < next_reader && snap_tid >= MDBX_TID_TXN_OUSTED) {
+              next_reader = snap_txnid;
+              retired_next_reader = pgno2bytes(
+                  env, (pgno_t)(snap_retired - atomic_load64(&txn->ro.slot->snapshot_pages_retired, mo_Relaxed)));
+            }
+          }
+        }
+      }
+      info->txn_space_dirty = retired_next_reader;
+    }
+  } else {
+    info->txn_space_limit_soft = pgno2bytes(env, txn->geo.now);
+    info->txn_space_limit_hard = pgno2bytes(env, txn->geo.upper);
+    info->txn_space_retired =
+        pgno2bytes(env, txn->nested ? (size_t)txn->wr.retired_pages : MDBX_PNL_GETSIZE(txn->wr.retired_pages));
+    info->txn_space_leftover = pgno2bytes(env, txn->wr.dirtyroom);
+    info->txn_space_dirty =
+        pgno2bytes(env, txn->wr.dirtylist ? txn->wr.dirtylist->pages_including_loose
+                                          : (txn->wr.writemap_dirty_npages + txn->wr.writemap_spilled_npages));
+    info->txn_reader_lag = INT64_MAX;
+    lck_t *const lck = env->lck_mmap.lck;
+    if (scan_rlt && lck) {
+      txnid_t oldest_reading = txn->txnid;
+      const size_t snap_nreaders = atomic_load32(&lck->rdt_length, mo_AcquireRelease);
+      if (snap_nreaders) {
+        txn_gc_detent(txn);
+        oldest_reading = txn->env->gc.detent;
+        if (oldest_reading == txn->wr.troika.txnid[txn->wr.troika.recent]) {
+          /* Если самый старый используемый снимок является предыдущим, т. е. непосредственно предшествующим текущей
+           * транзакции, то просматриваем таблицу читателей чтобы выяснить действительно ли снимок используется
+           * читателями. */
+          oldest_reading = txn->txnid;
+          for (size_t i = 0; i < snap_nreaders; ++i) {
+            if (atomic_load32(&lck->rdt[i].pid, mo_Relaxed) && txn->env->gc.detent == safe64_read(&lck->rdt[i].txnid)) {
+              oldest_reading = txn->env->gc.detent;
+              break;
+            }
+          }
+        }
+      }
+      info->txn_reader_lag = txn->txnid - oldest_reading;
+    }
+  }
+
+  return MDBX_SUCCESS;
+}

src/atomics-ops.h

@@ -0,0 +1,364 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#pragma once
+
+#include "essentials.h"
+
+#ifndef __cplusplus
+
+#ifdef MDBX_HAVE_C11ATOMICS
+#define osal_memory_fence(order, write) atomic_thread_fence((write) ? mo_c11_store(order) : mo_c11_load(order))
+#else /* MDBX_HAVE_C11ATOMICS */
+#define osal_memory_fence(order, write)                                                                                \
+  do {                                                                                                                 \
+    osal_compiler_barrier();                                                                                           \
+    if (write && order > (MDBX_CPU_WRITEBACK_INCOHERENT ? mo_Relaxed : mo_AcquireRelease))                             \
+      osal_memory_barrier();                                                                                           \
+  } while (0)
+#endif /* MDBX_HAVE_C11ATOMICS */
+
+#if defined(MDBX_HAVE_C11ATOMICS) && defined(__LCC__)
+#define atomic_store32(p, value, order)                                                                                \
+  ({                                                                                                                   \
+    const uint32_t value_to_store = (value);                                                                           \
+    atomic_store_explicit(MDBX_c11a_rw(uint32_t, p), value_to_store, mo_c11_store(order));                             \
+    value_to_store;                                                                                                    \
+  })
+#define atomic_load32(p, order) atomic_load_explicit(MDBX_c11a_ro(uint32_t, p), mo_c11_load(order))
+#define atomic_store64(p, value, order)                                                                                \
+  ({                                                                                                                   \
+    const uint64_t value_to_store = (value);                                                                           \
+    atomic_store_explicit(MDBX_c11a_rw(uint64_t, p), value_to_store, mo_c11_store(order));                             \
+    value_to_store;                                                                                                    \
+  })
+#define atomic_load64(p, order) atomic_load_explicit(MDBX_c11a_ro(uint64_t, p), mo_c11_load(order))
+#endif /* LCC && MDBX_HAVE_C11ATOMICS */
+
+#ifndef atomic_store32
+MDBX_MAYBE_UNUSED static __always_inline uint32_t atomic_store32(mdbx_atomic_uint32_t *p, const uint32_t value,
+                                                                 enum mdbx_memory_order order) {
+  STATIC_ASSERT(sizeof(mdbx_atomic_uint32_t) == 4);
+#ifdef MDBX_HAVE_C11ATOMICS
+  assert(atomic_is_lock_free(MDBX_c11a_rw(uint32_t, p)));
+  atomic_store_explicit(MDBX_c11a_rw(uint32_t, p), value, mo_c11_store(order));
+#else  /* MDBX_HAVE_C11ATOMICS */
+  if (order != mo_Relaxed)
+    osal_compiler_barrier();
+  p->weak = value;
+  osal_memory_fence(order, true);
+#endif /* MDBX_HAVE_C11ATOMICS */
+  return value;
+}
+#endif /* atomic_store32 */
+
+#ifndef atomic_load32
+MDBX_MAYBE_UNUSED static __always_inline uint32_t atomic_load32(const volatile mdbx_atomic_uint32_t *p,
+                                                                enum mdbx_memory_order order) {
+  STATIC_ASSERT(sizeof(mdbx_atomic_uint32_t) == 4);
+#ifdef MDBX_HAVE_C11ATOMICS
+  assert(atomic_is_lock_free(MDBX_c11a_ro(uint32_t, p)));
+  return atomic_load_explicit(MDBX_c11a_ro(uint32_t, p), mo_c11_load(order));
+#else  /* MDBX_HAVE_C11ATOMICS */
+  osal_memory_fence(order, false);
+  const uint32_t value = p->weak;
+  if (order != mo_Relaxed)
+    osal_compiler_barrier();
+  return value;
+#endif /* MDBX_HAVE_C11ATOMICS */
+}
+#endif /* atomic_load32 */
+
+/*------------------------------------------------------------------------------
+ * safe read/write volatile 64-bit fields on 32-bit architectures. */
+
+/* LY: for testing non-atomic 64-bit txnid on 32-bit arches.
+ * #define xMDBX_TXNID_STEP (UINT32_MAX / 3) */
+#ifndef xMDBX_TXNID_STEP
+#if MDBX_64BIT_CAS
+#define xMDBX_TXNID_STEP 1u
+#else
+#define xMDBX_TXNID_STEP 2u
+#endif
+#endif /* xMDBX_TXNID_STEP */
+
+#ifndef atomic_store64
+MDBX_MAYBE_UNUSED static __always_inline uint64_t atomic_store64(mdbx_atomic_uint64_t *p, const uint64_t value,
+                                                                 enum mdbx_memory_order order) {
+  STATIC_ASSERT(sizeof(mdbx_atomic_uint64_t) == 8);
+#if MDBX_64BIT_ATOMIC
+#if __GNUC_PREREQ(11, 0)
+  STATIC_ASSERT(__alignof__(mdbx_atomic_uint64_t) >= sizeof(uint64_t));
+#endif /* GNU C >= 11 */
+#ifdef MDBX_HAVE_C11ATOMICS
+  assert(atomic_is_lock_free(MDBX_c11a_rw(uint64_t, p)));
+  atomic_store_explicit(MDBX_c11a_rw(uint64_t, p), value, mo_c11_store(order));
+#else  /* MDBX_HAVE_C11ATOMICS */
+  if (order != mo_Relaxed)
+    osal_compiler_barrier();
+  p->weak = value;
+  osal_memory_fence(order, true);
+#endif /* MDBX_HAVE_C11ATOMICS */
+#else  /* !MDBX_64BIT_ATOMIC */
+  osal_compiler_barrier();
+  atomic_store32(&p->low, (uint32_t)value, mo_Relaxed);
+  jitter4testing(true);
+  atomic_store32(&p->high, (uint32_t)(value >> 32), order);
+  jitter4testing(true);
+#endif /* !MDBX_64BIT_ATOMIC */
+  return value;
+}
+#endif /* atomic_store64 */
+
+#ifndef atomic_load64
+MDBX_MAYBE_UNUSED static
+#if MDBX_64BIT_ATOMIC
+    __always_inline
+#endif /* MDBX_64BIT_ATOMIC */
+        uint64_t
+        atomic_load64(const volatile mdbx_atomic_uint64_t *p, enum mdbx_memory_order order) {
+  STATIC_ASSERT(sizeof(mdbx_atomic_uint64_t) == 8);
+#if MDBX_64BIT_ATOMIC
+#ifdef MDBX_HAVE_C11ATOMICS
+  assert(atomic_is_lock_free(MDBX_c11a_ro(uint64_t, p)));
+  return atomic_load_explicit(MDBX_c11a_ro(uint64_t, p), mo_c11_load(order));
+#else  /* MDBX_HAVE_C11ATOMICS */
+  osal_memory_fence(order, false);
+  const uint64_t value = p->weak;
+  if (order != mo_Relaxed)
+    osal_compiler_barrier();
+  return value;
+#endif /* MDBX_HAVE_C11ATOMICS */
+#else  /* !MDBX_64BIT_ATOMIC */
+  osal_compiler_barrier();
+  uint64_t value = (uint64_t)atomic_load32(&p->high, order) << 32;
+  jitter4testing(true);
+  value |= atomic_load32(&p->low, (order == mo_Relaxed) ? mo_Relaxed : mo_AcquireRelease);
+  jitter4testing(true);
+  for (;;) {
+    osal_compiler_barrier();
+    uint64_t again = (uint64_t)atomic_load32(&p->high, order) << 32;
+    jitter4testing(true);
+    again |= atomic_load32(&p->low, (order == mo_Relaxed) ? mo_Relaxed : mo_AcquireRelease);
+    jitter4testing(true);
+    if (likely(value == again))
+      return value;
+    value = again;
+  }
+#endif /* !MDBX_64BIT_ATOMIC */
+}
+#endif /* atomic_load64 */
+
+MDBX_MAYBE_UNUSED static __always_inline void atomic_yield(void) {
+#if defined(_WIN32) || defined(_WIN64)
+  YieldProcessor();
+#elif defined(__ia32__) || defined(__e2k__)
+  __builtin_ia32_pause();
+#elif defined(__ia64__)
+#if defined(__HP_cc__) || defined(__HP_aCC__)
+  _Asm_hint(_HINT_PAUSE);
+#else
+  __asm__ __volatile__("hint @pause");
+#endif
+#elif defined(__aarch64__) || (defined(__ARM_ARCH) && __ARM_ARCH > 6) || defined(__ARM_ARCH_6K__)
+#ifdef __CC_ARM
+  __yield();
+#else
+  __asm__ __volatile__("yield");
+#endif
+#elif (defined(__mips64) || defined(__mips64__)) && defined(__mips_isa_rev) && __mips_isa_rev >= 2
+  __asm__ __volatile__("pause");
+#elif defined(__mips) || defined(__mips__) || defined(__mips64) || defined(__mips64__) || defined(_M_MRX000) ||        \
+    defined(_MIPS_) || defined(__MWERKS__) || defined(__sgi)
+  __asm__ __volatile__(".word 0x00000140");
+#elif defined(__linux__) || defined(__gnu_linux__) || defined(_UNIX03_SOURCE)
+  sched_yield();
+#elif (defined(_GNU_SOURCE) && __GLIBC_PREREQ(2, 1)) || defined(_OPEN_THREADS)
+  pthread_yield();
+#endif
+}
+
+#if MDBX_64BIT_CAS
+MDBX_MAYBE_UNUSED static __always_inline bool atomic_cas64(mdbx_atomic_uint64_t *p, uint64_t c, uint64_t v) {
+#ifdef MDBX_HAVE_C11ATOMICS
+  STATIC_ASSERT(sizeof(long long) >= sizeof(uint64_t));
+  assert(atomic_is_lock_free(MDBX_c11a_rw(uint64_t, p)));
+  return atomic_compare_exchange_strong(MDBX_c11a_rw(uint64_t, p), &c, v);
+#elif defined(__GNUC__) || defined(__clang__)
+  return __sync_bool_compare_and_swap(&p->weak, c, v);
+#elif defined(_MSC_VER)
+  return c == (uint64_t)_InterlockedCompareExchange64((volatile __int64 *)&p->weak, v, c);
+#elif defined(__APPLE__)
+  return OSAtomicCompareAndSwap64Barrier(c, v, &p->weak);
+#else
+#error FIXME: Unsupported compiler
+#endif
+}
+#endif /* MDBX_64BIT_CAS */
+
+MDBX_MAYBE_UNUSED static __always_inline bool atomic_cas32(mdbx_atomic_uint32_t *p, uint32_t c, uint32_t v) {
+#ifdef MDBX_HAVE_C11ATOMICS
+  STATIC_ASSERT(sizeof(int) >= sizeof(uint32_t));
+  assert(atomic_is_lock_free(MDBX_c11a_rw(uint32_t, p)));
+  return atomic_compare_exchange_strong(MDBX_c11a_rw(uint32_t, p), &c, v);
+#elif defined(__GNUC__) || defined(__clang__)
+  return __sync_bool_compare_and_swap(&p->weak, c, v);
+#elif defined(_MSC_VER)
+  STATIC_ASSERT(sizeof(volatile long) == sizeof(volatile uint32_t));
+  return c == (uint32_t)_InterlockedCompareExchange((volatile long *)&p->weak, v, c);
+#elif defined(__APPLE__)
+  return OSAtomicCompareAndSwap32Barrier(c, v, &p->weak);
+#else
+#error FIXME: Unsupported compiler
+#endif
+}
+
+MDBX_MAYBE_UNUSED static __always_inline uint32_t atomic_add32(mdbx_atomic_uint32_t *p, uint32_t v) {
+#ifdef MDBX_HAVE_C11ATOMICS
+  STATIC_ASSERT(sizeof(int) >= sizeof(uint32_t));
+  assert(atomic_is_lock_free(MDBX_c11a_rw(uint32_t, p)));
+  return atomic_fetch_add(MDBX_c11a_rw(uint32_t, p), v);
+#elif defined(__GNUC__) || defined(__clang__)
+  return __sync_fetch_and_add(&p->weak, v);
+#elif defined(_MSC_VER)
+  STATIC_ASSERT(sizeof(volatile long) == sizeof(volatile uint32_t));
+  return (uint32_t)_InterlockedExchangeAdd((volatile long *)&p->weak, v);
+#elif defined(__APPLE__)
+  return OSAtomicAdd32Barrier(v, &p->weak);
+#else
+#error FIXME: Unsupported compiler
+#endif
+}
+
+#define atomic_sub32(p, v) atomic_add32(p, 0 - (v))
+
+MDBX_MAYBE_UNUSED static __always_inline uint64_t safe64_txnid_next(uint64_t txnid) {
+  txnid += xMDBX_TXNID_STEP;
+#if !MDBX_64BIT_CAS
+  /* avoid overflow of low-part in safe64_reset() */
+  txnid += (UINT32_MAX == (uint32_t)txnid);
+#endif
+  return txnid;
+}
+
+/* Atomically make target value >= SAFE64_INVALID_THRESHOLD */
+MDBX_MAYBE_UNUSED static __always_inline void safe64_reset(mdbx_atomic_uint64_t *p, bool single_writer) {
+  if (single_writer) {
+#if MDBX_64BIT_ATOMIC && MDBX_WORDBITS >= 64
+    atomic_store64(p, UINT64_MAX, mo_AcquireRelease);
+#else
+    atomic_store32(&p->high, UINT32_MAX, mo_AcquireRelease);
+#endif /* MDBX_64BIT_ATOMIC && MDBX_WORDBITS >= 64 */
+  } else {
+#if MDBX_64BIT_CAS && MDBX_64BIT_ATOMIC
+    /* atomically make value >= SAFE64_INVALID_THRESHOLD by 64-bit operation */
+    atomic_store64(p, UINT64_MAX, mo_AcquireRelease);
+#elif MDBX_64BIT_CAS
+    /* atomically make value >= SAFE64_INVALID_THRESHOLD by 32-bit operation */
+    atomic_store32(&p->high, UINT32_MAX, mo_AcquireRelease);
+#else
+    /* it is safe to increment low-part to avoid ABA, since xMDBX_TXNID_STEP > 1
+     * and overflow was preserved in safe64_txnid_next() */
+    STATIC_ASSERT(xMDBX_TXNID_STEP > 1);
+    atomic_add32(&p->low, 1) /* avoid ABA in safe64_reset_compare() */;
+    atomic_store32(&p->high, UINT32_MAX, mo_AcquireRelease);
+    atomic_add32(&p->low, 1) /* avoid ABA in safe64_reset_compare() */;
+#endif /* MDBX_64BIT_CAS && MDBX_64BIT_ATOMIC */
+  }
+  assert(p->weak >= SAFE64_INVALID_THRESHOLD);
+  jitter4testing(true);
+}
+
+MDBX_MAYBE_UNUSED static __always_inline bool safe64_reset_compare(mdbx_atomic_uint64_t *p, uint64_t compare) {
+  /* LY: This function is used to reset `txnid` from hsr-handler in case
+   *     the asynchronously cancellation of read transaction. Therefore,
+   *     there may be a collision between the cleanup performed here and
+   *     asynchronous termination and restarting of the read transaction
+   *     in another process/thread. In general we MUST NOT reset the `txnid`
+   *     if a new transaction was started (i.e. if `txnid` was changed). */
+#if MDBX_64BIT_CAS
+  bool rc = atomic_cas64(p, compare, UINT64_MAX);
+#else
+  /* LY: There is no gold ratio here since shared mutex is too costly,
+   *     in such way we must acquire/release it for every update of txnid,
+   *     i.e. twice for each read transaction). */
+  bool rc = false;
+  if (likely(atomic_load32(&p->low, mo_AcquireRelease) == (uint32_t)compare &&
+             atomic_cas32(&p->high, (uint32_t)(compare >> 32), UINT32_MAX))) {
+    if (unlikely(atomic_load32(&p->low, mo_AcquireRelease) != (uint32_t)compare))
+      atomic_cas32(&p->high, UINT32_MAX, (uint32_t)(compare >> 32));
+    else
+      rc = true;
+  }
+#endif /* MDBX_64BIT_CAS */
+  jitter4testing(true);
+  return rc;
+}
+
+MDBX_MAYBE_UNUSED static __always_inline void safe64_write(mdbx_atomic_uint64_t *p, const uint64_t v) {
+  assert(p->weak >= SAFE64_INVALID_THRESHOLD);
+#if MDBX_64BIT_ATOMIC && MDBX_64BIT_CAS
+  atomic_store64(p, v, mo_AcquireRelease);
+#else  /* MDBX_64BIT_ATOMIC */
+  osal_compiler_barrier();
+  /* update low-part but still value >= SAFE64_INVALID_THRESHOLD */
+  atomic_store32(&p->low, (uint32_t)v, mo_Relaxed);
+  assert(p->weak >= SAFE64_INVALID_THRESHOLD);
+  jitter4testing(true);
+  /* update high-part from SAFE64_INVALID_THRESHOLD to actual value */
+  atomic_store32(&p->high, (uint32_t)(v >> 32), mo_AcquireRelease);
+#endif /* MDBX_64BIT_ATOMIC */
+  assert(p->weak == v);
+  jitter4testing(true);
+}
+
+MDBX_MAYBE_UNUSED static __always_inline uint64_t safe64_read(const mdbx_atomic_uint64_t *p) {
+  jitter4testing(true);
+  uint64_t v;
+  do
+    v = atomic_load64(p, mo_AcquireRelease);
+  while (!MDBX_64BIT_ATOMIC && unlikely(v != p->weak));
+  return v;
+}
+
+#if 0 /* unused for now */
+MDBX_MAYBE_UNUSED static __always_inline bool safe64_is_valid(uint64_t v) {
+#if MDBX_WORDBITS >= 64
+  return v < SAFE64_INVALID_THRESHOLD;
+#else
+  return (v >> 32) != UINT32_MAX;
+#endif /* MDBX_WORDBITS */
+}
+
+MDBX_MAYBE_UNUSED static __always_inline bool
+ safe64_is_valid_ptr(const mdbx_atomic_uint64_t *p) {
+#if MDBX_64BIT_ATOMIC
+  return atomic_load64(p, mo_AcquireRelease) < SAFE64_INVALID_THRESHOLD;
+#else
+  return atomic_load32(&p->high, mo_AcquireRelease) != UINT32_MAX;
+#endif /* MDBX_64BIT_ATOMIC */
+}
+#endif /* unused for now */
+
+/* non-atomic write with safety for reading a half-updated value */
+MDBX_MAYBE_UNUSED static __always_inline void safe64_update(mdbx_atomic_uint64_t *p, const uint64_t v) {
+#if MDBX_64BIT_ATOMIC
+  atomic_store64(p, v, mo_Relaxed);
+#else
+  safe64_reset(p, true);
+  safe64_write(p, v);
+#endif /* MDBX_64BIT_ATOMIC */
+}
+
+/* non-atomic increment with safety for reading a half-updated value */
+MDBX_MAYBE_UNUSED static
+#if MDBX_64BIT_ATOMIC
+    __always_inline
+#endif /* MDBX_64BIT_ATOMIC */
+    void
+    safe64_inc(mdbx_atomic_uint64_t *p, const uint64_t v) {
+  assert(v > 0);
+  safe64_update(p, safe64_read(p) + v);
+}
+
+#endif /* !__cplusplus */

src/atomics-types.h

@@ -0,0 +1,97 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#pragma once
+
+#include "essentials.h"
+
+#ifndef MDBX_64BIT_ATOMIC
+#error "The MDBX_64BIT_ATOMIC must be defined before"
+#endif /* MDBX_64BIT_ATOMIC */
+
+#ifndef MDBX_64BIT_CAS
+#error "The MDBX_64BIT_CAS must be defined before"
+#endif /* MDBX_64BIT_CAS */
+
+#if defined(__cplusplus) && !defined(__STDC_NO_ATOMICS__) && __has_include(<cstdatomic>)
+#include <cstdatomic>
+#define MDBX_HAVE_C11ATOMICS
+#elif !defined(__cplusplus) && (__STDC_VERSION__ >= 201112L || __has_extension(c_atomic)) &&                           \
+    !defined(__STDC_NO_ATOMICS__) &&                                                                                   \
+    (__GNUC_PREREQ(4, 9) || __CLANG_PREREQ(3, 8) || !(defined(__GNUC__) || defined(__clang__)))
+#include <stdatomic.h>
+#define MDBX_HAVE_C11ATOMICS
+#elif defined(__GNUC__) || defined(__clang__)
+#elif defined(_MSC_VER)
+#pragma warning(disable : 4163) /* 'xyz': not available as an intrinsic */
+#pragma warning(disable : 4133) /* 'function': incompatible types - from                                               \
+                                   'size_t' to 'LONGLONG' */
+#pragma warning(disable : 4244) /* 'return': conversion from 'LONGLONG' to                                             \
+                                   'std::size_t', possible loss of data */
+#pragma warning(disable : 4267) /* 'function': conversion from 'size_t' to                                             \
+                                   'long', possible loss of data */
+#pragma intrinsic(_InterlockedExchangeAdd, _InterlockedCompareExchange)
+#pragma intrinsic(_InterlockedExchangeAdd64, _InterlockedCompareExchange64)
+#elif defined(__APPLE__)
+#include <libkern/OSAtomic.h>
+#else
+#error FIXME atomic-ops
+#endif
+
+typedef enum mdbx_memory_order {
+  mo_Relaxed,
+  mo_AcquireRelease
+  /* , mo_SequentialConsistency */
+} mdbx_memory_order_t;
+
+typedef union {
+  volatile uint32_t weak;
+#ifdef MDBX_HAVE_C11ATOMICS
+  volatile _Atomic uint32_t c11a;
+#endif /* MDBX_HAVE_C11ATOMICS */
+} mdbx_atomic_uint32_t;
+
+typedef union {
+  volatile uint64_t weak;
+#if defined(MDBX_HAVE_C11ATOMICS) && (MDBX_64BIT_CAS || MDBX_64BIT_ATOMIC)
+  volatile _Atomic uint64_t c11a;
+#endif
+#if !defined(MDBX_HAVE_C11ATOMICS) || !MDBX_64BIT_CAS || !MDBX_64BIT_ATOMIC
+  __anonymous_struct_extension__ struct {
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+    mdbx_atomic_uint32_t low, high;
+#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+    mdbx_atomic_uint32_t high, low;
+#else
+#error "FIXME: Unsupported byte order"
+#endif /* __BYTE_ORDER__ */
+  };
+#endif
+} mdbx_atomic_uint64_t;
+
+#ifdef MDBX_HAVE_C11ATOMICS
+
+/* Crutches for C11 atomic compiler's bugs */
+#if defined(__e2k__) && defined(__LCC__) && __LCC__ < /* FIXME */ 127
+#define MDBX_c11a_ro(type, ptr) (&(ptr)->weak)
+#define MDBX_c11a_rw(type, ptr) (&(ptr)->weak)
+#elif defined(__clang__) && __clang__ < 8
+#define MDBX_c11a_ro(type, ptr) ((volatile _Atomic(type) *)&(ptr)->c11a)
+#define MDBX_c11a_rw(type, ptr) (&(ptr)->c11a)
+#else
+#define MDBX_c11a_ro(type, ptr) (&(ptr)->c11a)
+#define MDBX_c11a_rw(type, ptr) (&(ptr)->c11a)
+#endif /* Crutches for C11 atomic compiler's bugs */
+
+#define mo_c11_store(fence)                                                                                            \
+  (((fence) == mo_Relaxed)          ? memory_order_relaxed                                                             \
+   : ((fence) == mo_AcquireRelease) ? memory_order_release                                                             \
+                                    : memory_order_seq_cst)
+#define mo_c11_load(fence)                                                                                             \
+  (((fence) == mo_Relaxed)          ? memory_order_relaxed                                                             \
+   : ((fence) == mo_AcquireRelease) ? memory_order_acquire                                                             \
+                                    : memory_order_seq_cst)
+
+#endif /* MDBX_HAVE_C11ATOMICS */
+
+#define SAFE64_INVALID_THRESHOLD UINT64_C(0xffffFFFF00000000)

src/audit.c

@@ -0,0 +1,107 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+struct audit_ctx {
+  size_t used;
+  uint8_t *const done_bitmap;
+};
+
+static int audit_dbi(void *ctx, const MDBX_txn *txn, const MDBX_val *name, MDBX_db_flags_t flags,
+                     const struct MDBX_stat *stat, MDBX_dbi dbi) {
+  struct audit_ctx *audit_ctx = ctx;
+  (void)name;
+  (void)txn;
+  (void)flags;
+  audit_ctx->used += (size_t)stat->ms_branch_pages + (size_t)stat->ms_leaf_pages + (size_t)stat->ms_overflow_pages;
+  if (dbi)
+    audit_ctx->done_bitmap[dbi / CHAR_BIT] |= 1 << dbi % CHAR_BIT;
+  return MDBX_SUCCESS;
+}
+
+static size_t audit_db_used(const tree_t *db) {
+  return db ? (size_t)db->branch_pages + (size_t)db->leaf_pages + (size_t)db->large_pages : 0;
+}
+
+__cold static int audit_ex_locked(MDBX_txn *txn, const size_t retired_stored, const bool dont_filter_gc) {
+  const MDBX_env *const env = txn->env;
+  tASSERT(txn, (txn->flags & MDBX_TXN_RDONLY) == 0);
+  const size_t pending = txn->wr.loose_count + MDBX_PNL_GETSIZE(txn->wr.repnl) +
+                         (MDBX_PNL_GETSIZE(txn->wr.retired_pages) - retired_stored);
+
+  cursor_couple_t cx;
+  int rc = cursor_init(&cx.outer, txn, FREE_DBI);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return rc;
+
+  size_t gc = 0;
+  MDBX_val key, data;
+  rc = outer_first(&cx.outer, &key, &data);
+  while (rc == MDBX_SUCCESS) {
+    if (unlikely(key.iov_len != sizeof(txnid_t))) {
+      ERROR("%s/%d: %s %u", "MDBX_CORRUPTED", MDBX_CORRUPTED, "invalid GC-key size", (unsigned)key.iov_len);
+      return MDBX_CORRUPTED;
+    }
+    const txnid_t id = unaligned_peek_u64(4, key.iov_base);
+    const size_t len = *(pgno_t *)data.iov_base;
+    const bool acc = dont_filter_gc || !gc_is_reclaimed(txn, id);
+    TRACE("%s id %" PRIaTXN " len %zu", acc ? "acc" : "skip", id, len);
+    if (acc)
+      gc += len;
+    rc = outer_next(&cx.outer, &key, &data, MDBX_NEXT);
+  }
+  tASSERT(txn, rc == MDBX_NOTFOUND);
+
+  const size_t done_bitmap_size = (txn->n_dbi + CHAR_BIT - 1) / CHAR_BIT;
+  if (txn->parent) {
+    tASSERT(txn, txn->n_dbi == txn->parent->n_dbi && txn->n_dbi == txn->env->txn->n_dbi);
+#if MDBX_ENABLE_DBI_SPARSE
+    tASSERT(txn, txn->dbi_sparse == txn->parent->dbi_sparse && txn->dbi_sparse == txn->env->txn->dbi_sparse);
+#endif /* MDBX_ENABLE_DBI_SPARSE */
+  }
+
+  struct audit_ctx ctx = {0, alloca(done_bitmap_size)};
+  memset(ctx.done_bitmap, 0, done_bitmap_size);
+  ctx.used =
+      NUM_METAS + audit_db_used(dbi_dig(txn, FREE_DBI, nullptr)) + audit_db_used(dbi_dig(txn, MAIN_DBI, nullptr));
+
+  rc = mdbx_enumerate_tables(txn, audit_dbi, &ctx);
+  tASSERT(txn, rc == MDBX_SUCCESS);
+
+  for (size_t dbi = CORE_DBS; dbi < txn->n_dbi; ++dbi) {
+    if (ctx.done_bitmap[dbi / CHAR_BIT] & (1 << dbi % CHAR_BIT))
+      continue;
+    const tree_t *db = dbi_dig(txn, dbi, nullptr);
+    if (db)
+      ctx.used += audit_db_used(db);
+    else if (dbi_state(txn, dbi))
+      WARNING("audit %s@%" PRIaTXN ": unable account dbi %zd / \"%.*s\", state 0x%02x", txn->parent ? "nested-" : "",
+              txn->txnid, dbi, (int)env->kvs[dbi].name.iov_len, (const char *)env->kvs[dbi].name.iov_base,
+              dbi_state(txn, dbi));
+  }
+
+  if (pending + gc + ctx.used == txn->geo.first_unallocated)
+    return MDBX_SUCCESS;
+
+  if ((txn->flags & MDBX_TXN_RDONLY) == 0)
+    ERROR("audit @%" PRIaTXN ": %zu(pending) = %zu(loose) + "
+          "%zu(reclaimed) + %zu(retired-pending) - %zu(retired-stored)",
+          txn->txnid, pending, txn->wr.loose_count, MDBX_PNL_GETSIZE(txn->wr.repnl),
+          txn->wr.retired_pages ? MDBX_PNL_GETSIZE(txn->wr.retired_pages) : 0, retired_stored);
+  ERROR("audit @%" PRIaTXN ": %zu(pending) + %zu"
+        "(gc) + %zu(count) = %zu(total) <> %zu"
+        "(allocated)",
+        txn->txnid, pending, gc, ctx.used, pending + gc + ctx.used, (size_t)txn->geo.first_unallocated);
+  return MDBX_PROBLEM;
+}
+
+__cold int audit_ex(MDBX_txn *txn, size_t retired_stored, bool dont_filter_gc) {
+  MDBX_env *const env = txn->env;
+  int rc = osal_fastmutex_acquire(&env->dbi_lock);
+  if (likely(rc == MDBX_SUCCESS)) {
+    rc = audit_ex_locked(txn, retired_stored, dont_filter_gc);
+    ENSURE(txn->env, osal_fastmutex_release(&env->dbi_lock) == MDBX_SUCCESS);
+  }
+  return rc;
+}

src/bits.md

@@ -0,0 +1,34 @@
+N |   MASK  | ENV       | TXN          | DB       | PUT       | DBI        | NODE    | PAGE     | MRESIZE |
+--|---------|-----------|--------------|----------|-----------|------------|---------|----------|---------|
+0 |0000 0001|ALLOC_RSRV |TXN_FINISHED  |          |           |DBI_DIRTY   |N_BIG    |P_BRANCH  |         |
+1 |0000 0002|ALLOC_UNIMP|TXN_ERROR     |REVERSEKEY|N_TREE     |DBI_STALE   |N_TREE   |P_LEAF    |         |
+2 |0000 0004|ALLOC_COLSC|TXN_DIRTY     |DUPSORT   |           |DBI_FRESH   |N_DUP    |P_LARGE   |         |
+3 |0000 0008|ALLOC_SSCAN|TXN_SPILLS    |INTEGERKEY|           |DBI_CREAT   |         |P_META    |         |
+4 |0000 0010|ALLOC_FIFO |TXN_HAS_CHILD |DUPFIXED  |NOOVERWRITE|DBI_VALID   |         |P_BAD     |         |
+5 |0000 0020|           |TXN_PARKED    |INTEGERDUP|NODUPDATA  |            |         |P_DUPFIX  |         |
+6 |0000 0040|           |TXN_AUTOUNPARK|REVERSEDUP|CURRENT    |DBI_OLDEN   |         |P_SUBP    |         |
+7 |0000 0080|           |TXN_DRAINED_GC|DB_VALID  |ALLDUPS    |DBI_LINDO   |         |          |         |
+8 |0000 0100| _MAY_MOVE |TXN_CURSORS   |          |           |            |         |          | <=      |
+9 |0000 0200| _MAY_UNMAP|              |          |           |            |         |          | <=      |
+10|0000 0400|           |              |          |           |            |         |          |         |
+11|0000 0800|           |              |          |           |            |         |          |         |
+12|0000 1000|           |              |          |           |            |         |          |         |
+13|0000 2000|VALIDATION |              |          |           |            |         |P_SPILLED |         |
+14|0000 4000|NOSUBDIR   |              |          |           |            |         |P_LOOSE   |         |
+15|0000 8000|           |              |          |           |            |         |P_FROZEN  |         |
+16|0001 0000|SAFE_NOSYNC|TXN_NOSYNC    |          |RESERVE    |            |RESERVE  |          |         |
+17|0002 0000|RDONLY     |TXN_RDONLY    |          |APPEND     |            |APPEND   |          | <=      |
+18|0004 0000|NOMETASYNC |TXN_NOMETASYNC|CREATE    |APPENDDUP  |            |         |          |         |
+19|0008 0000|WRITEMAP   |<=            |          |MULTIPLE   |            |         |          | <=      |
+20|0010 0000|UTTERLY    |              |          |           |            |         |          | <=      |
+21|0020 0000|NOSTICKYTHR|<=            |          |           |            |         |          |         |
+22|0040 0000|EXCLUSIVE  |              |          |           |            |         |          |         |
+23|0080 0000|NORDAHEAD  |              |          |           |            |         |          |         |
+24|0100 0000|NOMEMINIT  |TXN_PREPARE   |          |           |            |         |          |         |
+25|0200 0000|COALESCE   |              |          |           |            |         |          |         |
+26|0400 0000|LIFORECLAIM|              |          |           |            |         |          |         |
+27|0800 0000|PAGEPERTURB|              |          |           |            |         |          |         |
+28|1000 0000|ENV_TXKEY  |TXN_TRY       |          |           |            |         |          |         |
+29|2000 0000|ENV_ACTIVE |              |          |           |            |         |          |         |
+30|4000 0000|ACCEDE     |SHRINK_ALLOWED|DB_ACCEDE |           |            |         |          |         |
+31|8000 0000|FATAL_ERROR|              |          |           |            |         |          |         |

src/cogs.c

@@ -0,0 +1,309 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+/*------------------------------------------------------------------------------
+ * Pack/Unpack 16-bit values for Grow step & Shrink threshold */
+
+MDBX_NOTHROW_CONST_FUNCTION static inline pgno_t me2v(size_t m, size_t e) {
+  assert(m < 2048 && e < 8);
+  return (pgno_t)(32768 + ((m + 1) << (e + 8)));
+}
+
+MDBX_NOTHROW_CONST_FUNCTION static inline uint16_t v2me(size_t v, size_t e) {
+  assert(v > (e ? me2v(2047, e - 1) : 32768));
+  assert(v <= me2v(2047, e));
+  size_t m = (v - 32768 + ((size_t)1 << (e + 8)) - 1) >> (e + 8);
+  m -= m > 0;
+  assert(m < 2048 && e < 8);
+  // f e d c b a 9 8 7 6 5 4 3 2 1 0
+  // 1 e e e m m m m m m m m m m m 1
+  const uint16_t pv = (uint16_t)(0x8001 + (e << 12) + (m << 1));
+  assert(pv != 65535);
+  return pv;
+}
+
+/* Convert 16-bit packed (exponential quantized) value to number of pages */
+pgno_t pv2pages(uint16_t pv) {
+  if ((pv & 0x8001) != 0x8001)
+    return pv;
+  if (pv == 65535)
+    return 65536;
+  // f e d c b a 9 8 7 6 5 4 3 2 1 0
+  // 1 e e e m m m m m m m m m m m 1
+  return me2v((pv >> 1) & 2047, (pv >> 12) & 7);
+}
+
+/* Convert number of pages to 16-bit packed (exponential quantized) value */
+uint16_t pages2pv(size_t pages) {
+  if (pages < 32769 || (pages < 65536 && (pages & 1) == 0))
+    return (uint16_t)pages;
+  if (pages <= me2v(2047, 0))
+    return v2me(pages, 0);
+  if (pages <= me2v(2047, 1))
+    return v2me(pages, 1);
+  if (pages <= me2v(2047, 2))
+    return v2me(pages, 2);
+  if (pages <= me2v(2047, 3))
+    return v2me(pages, 3);
+  if (pages <= me2v(2047, 4))
+    return v2me(pages, 4);
+  if (pages <= me2v(2047, 5))
+    return v2me(pages, 5);
+  if (pages <= me2v(2047, 6))
+    return v2me(pages, 6);
+  return (pages < me2v(2046, 7)) ? v2me(pages, 7) : 65533;
+}
+
+__cold bool pv2pages_verify(void) {
+  bool ok = true, dump_translation = false;
+  for (size_t i = 0; i < 65536; ++i) {
+    size_t pages = pv2pages(i);
+    size_t x = pages2pv(pages);
+    size_t xp = pv2pages(x);
+    if (pages != xp) {
+      ERROR("%zu => %zu => %zu => %zu\n", i, pages, x, xp);
+      ok = false;
+    } else if (dump_translation && !(x == i || (x % 2 == 0 && x < 65536))) {
+      DEBUG("%zu => %zu => %zu => %zu\n", i, pages, x, xp);
+    }
+  }
+  return ok;
+}
+
+/*----------------------------------------------------------------------------*/
+
+MDBX_NOTHROW_PURE_FUNCTION size_t bytes_align2os_bytes(const MDBX_env *env, size_t bytes) {
+  return ceil_powerof2(bytes, (env->ps > globals.sys_pagesize) ? env->ps : globals.sys_pagesize);
+}
+
+MDBX_NOTHROW_PURE_FUNCTION size_t pgno_align2os_bytes(const MDBX_env *env, size_t pgno) {
+  return ceil_powerof2(pgno2bytes(env, pgno), globals.sys_pagesize);
+}
+
+MDBX_NOTHROW_PURE_FUNCTION pgno_t pgno_align2os_pgno(const MDBX_env *env, size_t pgno) {
+  return bytes2pgno(env, pgno_align2os_bytes(env, pgno));
+}
+
+/*----------------------------------------------------------------------------*/
+
+MDBX_NOTHROW_PURE_FUNCTION static __always_inline int cmp_int_inline(const size_t expected_alignment, const MDBX_val *a,
+                                                                     const MDBX_val *b) {
+  if (likely(a->iov_len == b->iov_len)) {
+    if (sizeof(size_t) > 7 && likely(a->iov_len == 8))
+      return CMP2INT(unaligned_peek_u64(expected_alignment, a->iov_base),
+                     unaligned_peek_u64(expected_alignment, b->iov_base));
+    if (likely(a->iov_len == 4))
+      return CMP2INT(unaligned_peek_u32(expected_alignment, a->iov_base),
+                     unaligned_peek_u32(expected_alignment, b->iov_base));
+    if (sizeof(size_t) < 8 && likely(a->iov_len == 8))
+      return CMP2INT(unaligned_peek_u64(expected_alignment, a->iov_base),
+                     unaligned_peek_u64(expected_alignment, b->iov_base));
+  }
+  ERROR("mismatch and/or invalid size %p.%zu/%p.%zu for INTEGERKEY/INTEGERDUP", a->iov_base, a->iov_len, b->iov_base,
+        b->iov_len);
+  return 0;
+}
+
+MDBX_NOTHROW_PURE_FUNCTION __hot int cmp_int_unaligned(const MDBX_val *a, const MDBX_val *b) {
+  return cmp_int_inline(1, a, b);
+}
+
+#ifndef cmp_int_align2
+/* Compare two items pointing at 2-byte aligned unsigned int's. */
+MDBX_NOTHROW_PURE_FUNCTION __hot int cmp_int_align2(const MDBX_val *a, const MDBX_val *b) {
+  return cmp_int_inline(2, a, b);
+}
+#endif /* cmp_int_align2 */
+
+#ifndef cmp_int_align4
+/* Compare two items pointing at 4-byte aligned unsigned int's. */
+MDBX_NOTHROW_PURE_FUNCTION __hot int cmp_int_align4(const MDBX_val *a, const MDBX_val *b) {
+  return cmp_int_inline(4, a, b);
+}
+#endif /* cmp_int_align4 */
+
+/* Compare two items lexically */
+MDBX_NOTHROW_PURE_FUNCTION __hot int cmp_lexical(const MDBX_val *a, const MDBX_val *b) {
+  if (a->iov_len == b->iov_len)
+    return a->iov_len ? memcmp(a->iov_base, b->iov_base, a->iov_len) : 0;
+
+  const int diff_len = (a->iov_len < b->iov_len) ? -1 : 1;
+  const size_t shortest = (a->iov_len < b->iov_len) ? a->iov_len : b->iov_len;
+  int diff_data = shortest ? memcmp(a->iov_base, b->iov_base, shortest) : 0;
+  return likely(diff_data) ? diff_data : diff_len;
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static __always_inline unsigned tail3le(const uint8_t *p, size_t l) {
+  STATIC_ASSERT(sizeof(unsigned) > 2);
+  // 1: 0 0 0
+  // 2: 0 1 1
+  // 3: 0 1 2
+  return p[0] | p[l >> 1] << 8 | p[l - 1] << 16;
+}
+
+/* Compare two items in reverse byte order */
+MDBX_NOTHROW_PURE_FUNCTION __hot int cmp_reverse(const MDBX_val *a, const MDBX_val *b) {
+  size_t left = (a->iov_len < b->iov_len) ? a->iov_len : b->iov_len;
+  if (likely(left)) {
+    const uint8_t *pa = ptr_disp(a->iov_base, a->iov_len);
+    const uint8_t *pb = ptr_disp(b->iov_base, b->iov_len);
+    while (left >= sizeof(size_t)) {
+      pa -= sizeof(size_t);
+      pb -= sizeof(size_t);
+      left -= sizeof(size_t);
+      STATIC_ASSERT(sizeof(size_t) == 4 || sizeof(size_t) == 8);
+      if (sizeof(size_t) == 4) {
+        uint32_t xa = unaligned_peek_u32(1, pa);
+        uint32_t xb = unaligned_peek_u32(1, pb);
+#if __BYTE_ORDER__ != __ORDER_LITTLE_ENDIAN__
+        xa = osal_bswap32(xa);
+        xb = osal_bswap32(xb);
+#endif /* __BYTE_ORDER__ != __ORDER_BIG_ENDIAN__ */
+        if (xa != xb)
+          return (xa < xb) ? -1 : 1;
+      } else {
+        uint64_t xa = unaligned_peek_u64(1, pa);
+        uint64_t xb = unaligned_peek_u64(1, pb);
+#if __BYTE_ORDER__ != __ORDER_LITTLE_ENDIAN__
+        xa = osal_bswap64(xa);
+        xb = osal_bswap64(xb);
+#endif /* __BYTE_ORDER__ != __ORDER_BIG_ENDIAN__ */
+        if (xa != xb)
+          return (xa < xb) ? -1 : 1;
+      }
+    }
+    if (sizeof(size_t) == 8 && left >= 4) {
+      pa -= 4;
+      pb -= 4;
+      left -= 4;
+      uint32_t xa = unaligned_peek_u32(1, pa);
+      uint32_t xb = unaligned_peek_u32(1, pb);
+#if __BYTE_ORDER__ != __ORDER_LITTLE_ENDIAN__
+      xa = osal_bswap32(xa);
+      xb = osal_bswap32(xb);
+#endif /* __BYTE_ORDER__ != __ORDER_BIG_ENDIAN__ */
+      if (xa != xb)
+        return (xa < xb) ? -1 : 1;
+    }
+    if (left) {
+      unsigned xa = tail3le(pa - left, left);
+      unsigned xb = tail3le(pb - left, left);
+      if (xa != xb)
+        return (xa < xb) ? -1 : 1;
+    }
+  }
+  return CMP2INT(a->iov_len, b->iov_len);
+}
+
+/* Fast non-lexically comparator */
+MDBX_NOTHROW_PURE_FUNCTION __hot int cmp_lenfast(const MDBX_val *a, const MDBX_val *b) {
+  int diff = CMP2INT(a->iov_len, b->iov_len);
+  return (likely(diff) || a->iov_len == 0) ? diff : memcmp(a->iov_base, b->iov_base, a->iov_len);
+}
+
+MDBX_NOTHROW_PURE_FUNCTION __hot bool eq_fast_slowpath(const uint8_t *a, const uint8_t *b, size_t l) {
+  if (likely(l > 3)) {
+    if (MDBX_UNALIGNED_OK >= 4 && likely(l < 9))
+      return ((unaligned_peek_u32(1, a) - unaligned_peek_u32(1, b)) |
+              (unaligned_peek_u32(1, a + l - 4) - unaligned_peek_u32(1, b + l - 4))) == 0;
+    if (MDBX_UNALIGNED_OK >= 8 && sizeof(size_t) > 7 && likely(l < 17))
+      return ((unaligned_peek_u64(1, a) - unaligned_peek_u64(1, b)) |
+              (unaligned_peek_u64(1, a + l - 8) - unaligned_peek_u64(1, b + l - 8))) == 0;
+    return memcmp(a, b, l) == 0;
+  }
+  if (likely(l))
+    return tail3le(a, l) == tail3le(b, l);
+  return true;
+}
+
+int cmp_equal_or_greater(const MDBX_val *a, const MDBX_val *b) { return eq_fast(a, b) ? 0 : 1; }
+
+int cmp_equal_or_wrong(const MDBX_val *a, const MDBX_val *b) { return eq_fast(a, b) ? 0 : -1; }
+
+/*----------------------------------------------------------------------------*/
+
+__cold void update_mlcnt(const MDBX_env *env, const pgno_t new_aligned_mlocked_pgno, const bool lock_not_release) {
+  for (;;) {
+    const pgno_t mlock_pgno_before = atomic_load32(&env->mlocked_pgno, mo_AcquireRelease);
+    eASSERT(env, pgno_align2os_pgno(env, mlock_pgno_before) == mlock_pgno_before);
+    eASSERT(env, pgno_align2os_pgno(env, new_aligned_mlocked_pgno) == new_aligned_mlocked_pgno);
+    if (lock_not_release ? (mlock_pgno_before >= new_aligned_mlocked_pgno)
+                         : (mlock_pgno_before <= new_aligned_mlocked_pgno))
+      break;
+    if (likely(atomic_cas32(&((MDBX_env *)env)->mlocked_pgno, mlock_pgno_before, new_aligned_mlocked_pgno)))
+      for (;;) {
+        mdbx_atomic_uint32_t *const mlcnt = env->lck->mlcnt;
+        const int32_t snap_locked = atomic_load32(mlcnt + 0, mo_Relaxed);
+        const int32_t snap_unlocked = atomic_load32(mlcnt + 1, mo_Relaxed);
+        if (mlock_pgno_before == 0 && (snap_locked - snap_unlocked) < INT_MAX) {
+          eASSERT(env, lock_not_release);
+          if (unlikely(!atomic_cas32(mlcnt + 0, snap_locked, snap_locked + 1)))
+            continue;
+        }
+        if (new_aligned_mlocked_pgno == 0 && (snap_locked - snap_unlocked) > 0) {
+          eASSERT(env, !lock_not_release);
+          if (unlikely(!atomic_cas32(mlcnt + 1, snap_unlocked, snap_unlocked + 1)))
+            continue;
+        }
+        NOTICE("%s-pages %u..%u, mlocked-process(es) %u -> %u", lock_not_release ? "lock" : "unlock",
+               lock_not_release ? mlock_pgno_before : new_aligned_mlocked_pgno,
+               lock_not_release ? new_aligned_mlocked_pgno : mlock_pgno_before, snap_locked - snap_unlocked,
+               atomic_load32(mlcnt + 0, mo_Relaxed) - atomic_load32(mlcnt + 1, mo_Relaxed));
+        return;
+      }
+  }
+}
+
+__cold void munlock_after(const MDBX_env *env, const pgno_t aligned_pgno, const size_t end_bytes) {
+  if (atomic_load32(&env->mlocked_pgno, mo_AcquireRelease) > aligned_pgno) {
+    int err = MDBX_ENOSYS;
+    const size_t munlock_begin = pgno2bytes(env, aligned_pgno);
+    const size_t munlock_size = end_bytes - munlock_begin;
+    eASSERT(env, end_bytes % globals.sys_pagesize == 0 && munlock_begin % globals.sys_pagesize == 0 &&
+                     munlock_size % globals.sys_pagesize == 0);
+#if defined(_WIN32) || defined(_WIN64)
+    err = VirtualUnlock(ptr_disp(env->dxb_mmap.base, munlock_begin), munlock_size) ? MDBX_SUCCESS : (int)GetLastError();
+    if (err == ERROR_NOT_LOCKED)
+      err = MDBX_SUCCESS;
+#elif defined(_POSIX_MEMLOCK_RANGE)
+    err = munlock(ptr_disp(env->dxb_mmap.base, munlock_begin), munlock_size) ? errno : MDBX_SUCCESS;
+#endif
+    if (likely(err == MDBX_SUCCESS))
+      update_mlcnt(env, aligned_pgno, false);
+    else {
+#if defined(_WIN32) || defined(_WIN64)
+      WARNING("VirtualUnlock(%zu, %zu) error %d", munlock_begin, munlock_size, err);
+#else
+      WARNING("munlock(%zu, %zu) error %d", munlock_begin, munlock_size, err);
+#endif
+    }
+  }
+}
+
+__cold void munlock_all(const MDBX_env *env) {
+  munlock_after(env, 0, bytes_align2os_bytes(env, env->dxb_mmap.current));
+}
+
+/*----------------------------------------------------------------------------*/
+
+uint32_t combine_durability_flags(const uint32_t a, const uint32_t b) {
+  uint32_t r = a | b;
+
+  /* avoid false MDBX_UTTERLY_NOSYNC */
+  if (F_ISSET(r, MDBX_UTTERLY_NOSYNC) && !F_ISSET(a, MDBX_UTTERLY_NOSYNC) && !F_ISSET(b, MDBX_UTTERLY_NOSYNC))
+    r = (r - MDBX_UTTERLY_NOSYNC) | MDBX_SAFE_NOSYNC;
+
+  /* convert DEPRECATED_MAPASYNC to MDBX_SAFE_NOSYNC */
+  if ((r & (MDBX_WRITEMAP | DEPRECATED_MAPASYNC)) == (MDBX_WRITEMAP | DEPRECATED_MAPASYNC) &&
+      !F_ISSET(r, MDBX_UTTERLY_NOSYNC))
+    r = (r - DEPRECATED_MAPASYNC) | MDBX_SAFE_NOSYNC;
+
+  /* force MDBX_NOMETASYNC if NOSYNC enabled */
+  if (r & (MDBX_SAFE_NOSYNC | MDBX_UTTERLY_NOSYNC))
+    r |= MDBX_NOMETASYNC;
+
+  assert(!(F_ISSET(r, MDBX_UTTERLY_NOSYNC) && !F_ISSET(a, MDBX_UTTERLY_NOSYNC) && !F_ISSET(b, MDBX_UTTERLY_NOSYNC)));
+  return r;
+}

src/cogs.h

@@ -0,0 +1,507 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#pragma once
+
+#include "essentials.h"
+
+MDBX_NOTHROW_CONST_FUNCTION MDBX_INTERNAL pgno_t pv2pages(uint16_t pv);
+
+MDBX_NOTHROW_CONST_FUNCTION MDBX_INTERNAL uint16_t pages2pv(size_t pages);
+
+MDBX_MAYBE_UNUSED MDBX_INTERNAL bool pv2pages_verify(void);
+
+/*------------------------------------------------------------------------------
+ * Nodes, Keys & Values length limitation factors:
+ *
+ * BRANCH_NODE_MAX
+ *   Branch-page must contain at least two nodes, within each a key and a child
+ *   page number. But page can't be split if it contains less that 4 keys,
+ *   i.e. a page should not overflow before adding the fourth key. Therefore,
+ *   at least 3 branch-node should fit in the single branch-page. Further, the
+ *   first node of a branch-page doesn't contain a key, i.e. the first node
+ *   is always require space just for itself. Thus:
+ *       PAGESPACE = pagesize - page_hdr_len;
+ *       BRANCH_NODE_MAX = even_floor(
+ *         (PAGESPACE - sizeof(indx_t) - NODESIZE) / (3 - 1) - sizeof(indx_t));
+ *       KEYLEN_MAX = BRANCH_NODE_MAX - node_hdr_len;
+ *
+ * LEAF_NODE_MAX
+ *   Leaf-node must fit into single leaf-page, where a value could be placed on
+ *   a large/overflow page. However, may require to insert a nearly page-sized
+ *   node between two large nodes are already fill-up a page. In this case the
+ *   page must be split to two if some pair of nodes fits on one page, or
+ *   otherwise the page should be split to the THREE with a single node
+ *   per each of ones. Such 1-into-3 page splitting is costly and complex since
+ *   requires TWO insertion into the parent page, that could lead to split it
+ *   and so on up to the root. Therefore double-splitting is avoided here and
+ *   the maximum node size is half of a leaf page space:
+ *       LEAF_NODE_MAX = even_floor(PAGESPACE / 2 - sizeof(indx_t));
+ *       DATALEN_NO_OVERFLOW = LEAF_NODE_MAX - NODESIZE - KEYLEN_MAX;
+ *
+ *  - Table-node must fit into one leaf-page:
+ *       TABLE_NAME_MAX = LEAF_NODE_MAX - node_hdr_len - sizeof(tree_t);
+ *
+ *  - Dupsort values itself are a keys in a dupsort-table and couldn't be longer
+ *    than the KEYLEN_MAX. But dupsort node must not great than LEAF_NODE_MAX,
+ *    since dupsort value couldn't be placed on a large/overflow page:
+ *       DUPSORT_DATALEN_MAX = min(KEYLEN_MAX,
+ *                                 max(DATALEN_NO_OVERFLOW, sizeof(tree_t));
+ */
+
+#define PAGESPACE(pagesize) ((pagesize) - PAGEHDRSZ)
+
+#define BRANCH_NODE_MAX(pagesize)                                                                                      \
+  (EVEN_FLOOR((PAGESPACE(pagesize) - sizeof(indx_t) - NODESIZE) / (3 - 1) - sizeof(indx_t)))
+
+#define LEAF_NODE_MAX(pagesize) (EVEN_FLOOR(PAGESPACE(pagesize) / 2) - sizeof(indx_t))
+
+#define MAX_GC1OVPAGE(pagesize) (PAGESPACE(pagesize) / sizeof(pgno_t) - 1)
+
+MDBX_NOTHROW_CONST_FUNCTION static inline size_t keysize_max(size_t pagesize, MDBX_db_flags_t flags) {
+  assert(pagesize >= MDBX_MIN_PAGESIZE && pagesize <= MDBX_MAX_PAGESIZE && is_powerof2(pagesize));
+  STATIC_ASSERT(BRANCH_NODE_MAX(MDBX_MIN_PAGESIZE) - NODESIZE >= 8);
+  if (flags & MDBX_INTEGERKEY)
+    return 8 /* sizeof(uint64_t) */;
+
+  const intptr_t max_branch_key = BRANCH_NODE_MAX(pagesize) - NODESIZE;
+  STATIC_ASSERT(LEAF_NODE_MAX(MDBX_MIN_PAGESIZE) - NODESIZE -
+                    /* sizeof(uint64) as a key */ 8 >
+                sizeof(tree_t));
+  if (flags & (MDBX_DUPSORT | MDBX_DUPFIXED | MDBX_REVERSEDUP | MDBX_INTEGERDUP)) {
+    const intptr_t max_dupsort_leaf_key = LEAF_NODE_MAX(pagesize) - NODESIZE - sizeof(tree_t);
+    return (max_branch_key < max_dupsort_leaf_key) ? max_branch_key : max_dupsort_leaf_key;
+  }
+  return max_branch_key;
+}
+
+MDBX_NOTHROW_CONST_FUNCTION static inline size_t env_keysize_max(const MDBX_env *env, MDBX_db_flags_t flags) {
+  size_t size_max;
+  if (flags & MDBX_INTEGERKEY)
+    size_max = 8 /* sizeof(uint64_t) */;
+  else {
+    const intptr_t max_branch_key = env->branch_nodemax - NODESIZE;
+    STATIC_ASSERT(LEAF_NODE_MAX(MDBX_MIN_PAGESIZE) - NODESIZE -
+                      /* sizeof(uint64) as a key */ 8 >
+                  sizeof(tree_t));
+    if (flags & (MDBX_DUPSORT | MDBX_DUPFIXED | MDBX_REVERSEDUP | MDBX_INTEGERDUP)) {
+      const intptr_t max_dupsort_leaf_key = env->leaf_nodemax - NODESIZE - sizeof(tree_t);
+      size_max = (max_branch_key < max_dupsort_leaf_key) ? max_branch_key : max_dupsort_leaf_key;
+    } else
+      size_max = max_branch_key;
+  }
+  eASSERT(env, size_max == keysize_max(env->ps, flags));
+  return size_max;
+}
+
+MDBX_NOTHROW_CONST_FUNCTION static inline size_t keysize_min(MDBX_db_flags_t flags) {
+  return (flags & MDBX_INTEGERKEY) ? 4 /* sizeof(uint32_t) */ : 0;
+}
+
+MDBX_NOTHROW_CONST_FUNCTION static inline size_t valsize_min(MDBX_db_flags_t flags) {
+  if (flags & MDBX_INTEGERDUP)
+    return 4 /* sizeof(uint32_t) */;
+  else if (flags & MDBX_DUPFIXED)
+    return sizeof(indx_t);
+  else
+    return 0;
+}
+
+MDBX_NOTHROW_CONST_FUNCTION static inline size_t valsize_max(size_t pagesize, MDBX_db_flags_t flags) {
+  assert(pagesize >= MDBX_MIN_PAGESIZE && pagesize <= MDBX_MAX_PAGESIZE && is_powerof2(pagesize));
+
+  if (flags & MDBX_INTEGERDUP)
+    return 8 /* sizeof(uint64_t) */;
+
+  if (flags & (MDBX_DUPSORT | MDBX_DUPFIXED | MDBX_REVERSEDUP))
+    return keysize_max(pagesize, 0);
+
+  const unsigned page_ln2 = log2n_powerof2(pagesize);
+  const size_t hard = 0x7FF00000ul;
+  const size_t hard_pages = hard >> page_ln2;
+  STATIC_ASSERT(PAGELIST_LIMIT <= MAX_PAGENO);
+  const size_t pages_limit = PAGELIST_LIMIT / 4;
+  const size_t limit = (hard_pages < pages_limit) ? hard : (pages_limit << page_ln2);
+  return (limit < MAX_MAPSIZE / 2) ? limit : MAX_MAPSIZE / 2;
+}
+
+MDBX_NOTHROW_CONST_FUNCTION static inline size_t env_valsize_max(const MDBX_env *env, MDBX_db_flags_t flags) {
+  size_t size_max;
+  if (flags & MDBX_INTEGERDUP)
+    size_max = 8 /* sizeof(uint64_t) */;
+  else if (flags & (MDBX_DUPSORT | MDBX_DUPFIXED | MDBX_REVERSEDUP))
+    size_max = env_keysize_max(env, 0);
+  else {
+    const size_t hard = 0x7FF00000ul;
+    const size_t hard_pages = hard >> env->ps2ln;
+    STATIC_ASSERT(PAGELIST_LIMIT <= MAX_PAGENO);
+    const size_t pages_limit = PAGELIST_LIMIT / 4;
+    const size_t limit = (hard_pages < pages_limit) ? hard : (pages_limit << env->ps2ln);
+    size_max = (limit < MAX_MAPSIZE / 2) ? limit : MAX_MAPSIZE / 2;
+  }
+  eASSERT(env, size_max == valsize_max(env->ps, flags));
+  return size_max;
+}
+
+/*----------------------------------------------------------------------------*/
+
+MDBX_NOTHROW_PURE_FUNCTION static inline size_t leaf_size(const MDBX_env *env, const MDBX_val *key,
+                                                          const MDBX_val *data) {
+  size_t node_bytes = node_size(key, data);
+  if (node_bytes > env->leaf_nodemax)
+    /* put on large/overflow page */
+    node_bytes = node_size_len(key->iov_len, 0) + sizeof(pgno_t);
+
+  return node_bytes + sizeof(indx_t);
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline size_t branch_size(const MDBX_env *env, const MDBX_val *key) {
+  /* Size of a node in a branch page with a given key.
+   * This is just the node header plus the key, there is no data. */
+  size_t node_bytes = node_size(key, nullptr);
+  if (unlikely(node_bytes > env->branch_nodemax)) {
+    /* put on large/overflow page, not implemented */
+    mdbx_panic("node_size(key) %zu > %u branch_nodemax", node_bytes, env->branch_nodemax);
+    node_bytes = node_size(key, nullptr) + sizeof(pgno_t);
+  }
+
+  return node_bytes + sizeof(indx_t);
+}
+
+MDBX_NOTHROW_CONST_FUNCTION static inline uint16_t flags_db2sub(uint16_t db_flags) {
+  uint16_t sub_flags = db_flags & MDBX_DUPFIXED;
+
+  /* MDBX_INTEGERDUP => MDBX_INTEGERKEY */
+#define SHIFT_INTEGERDUP_TO_INTEGERKEY 2
+  STATIC_ASSERT((MDBX_INTEGERDUP >> SHIFT_INTEGERDUP_TO_INTEGERKEY) == MDBX_INTEGERKEY);
+  sub_flags |= (db_flags & MDBX_INTEGERDUP) >> SHIFT_INTEGERDUP_TO_INTEGERKEY;
+
+  /* MDBX_REVERSEDUP => MDBX_REVERSEKEY */
+#define SHIFT_REVERSEDUP_TO_REVERSEKEY 5
+  STATIC_ASSERT((MDBX_REVERSEDUP >> SHIFT_REVERSEDUP_TO_REVERSEKEY) == MDBX_REVERSEKEY);
+  sub_flags |= (db_flags & MDBX_REVERSEDUP) >> SHIFT_REVERSEDUP_TO_REVERSEKEY;
+
+  return sub_flags;
+}
+
+static inline bool check_table_flags(unsigned flags) {
+  switch (flags & ~(MDBX_REVERSEKEY | MDBX_INTEGERKEY)) {
+  default:
+    NOTICE("invalid db-flags 0x%x", flags);
+    return false;
+  case MDBX_DUPSORT:
+  case MDBX_DUPSORT | MDBX_REVERSEDUP:
+  case MDBX_DUPSORT | MDBX_DUPFIXED:
+  case MDBX_DUPSORT | MDBX_DUPFIXED | MDBX_REVERSEDUP:
+  case MDBX_DUPSORT | MDBX_DUPFIXED | MDBX_INTEGERDUP:
+  case MDBX_DUPSORT | MDBX_DUPFIXED | MDBX_INTEGERDUP | MDBX_REVERSEDUP:
+  case MDBX_DB_DEFAULTS:
+    return (flags & (MDBX_REVERSEKEY | MDBX_INTEGERKEY)) != (MDBX_REVERSEKEY | MDBX_INTEGERKEY);
+  }
+}
+
+static inline int tbl_setup_ifneed(const MDBX_env *env, volatile kvx_t *const kvx, const tree_t *const db) {
+  return likely(kvx->clc.v.lmax) ? MDBX_SUCCESS : tbl_setup(env, kvx, db);
+}
+
+/*----------------------------------------------------------------------------*/
+
+MDBX_NOTHROW_PURE_FUNCTION static inline size_t pgno2bytes(const MDBX_env *env, size_t pgno) {
+  eASSERT(env, (1u << env->ps2ln) == env->ps);
+  return ((size_t)pgno) << env->ps2ln;
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline page_t *pgno2page(const MDBX_env *env, size_t pgno) {
+  return ptr_disp(env->dxb_mmap.base, pgno2bytes(env, pgno));
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline pgno_t bytes2pgno(const MDBX_env *env, size_t bytes) {
+  eASSERT(env, (env->ps >> env->ps2ln) == 1);
+  return (pgno_t)(bytes >> env->ps2ln);
+}
+
+MDBX_NOTHROW_PURE_FUNCTION MDBX_INTERNAL size_t bytes_align2os_bytes(const MDBX_env *env, size_t bytes);
+
+MDBX_NOTHROW_PURE_FUNCTION MDBX_INTERNAL size_t pgno_align2os_bytes(const MDBX_env *env, size_t pgno);
+
+MDBX_NOTHROW_PURE_FUNCTION MDBX_INTERNAL pgno_t pgno_align2os_pgno(const MDBX_env *env, size_t pgno);
+
+MDBX_NOTHROW_PURE_FUNCTION static inline pgno_t largechunk_npages(const MDBX_env *env, size_t bytes) {
+  return bytes2pgno(env, PAGEHDRSZ - 1 + bytes) + 1;
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline MDBX_val get_key(const node_t *node) {
+  MDBX_val key;
+  key.iov_len = node_ks(node);
+  key.iov_base = node_key(node);
+  return key;
+}
+
+static inline void get_key_optional(const node_t *node, MDBX_val *keyptr /* __may_null */) {
+  if (keyptr)
+    *keyptr = get_key(node);
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline void *page_data(const page_t *mp) { return ptr_disp(mp, PAGEHDRSZ); }
+
+MDBX_NOTHROW_PURE_FUNCTION static inline const page_t *data_page(const void *data) {
+  return container_of(data, page_t, entries);
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline meta_t *page_meta(page_t *mp) { return (meta_t *)page_data(mp); }
+
+MDBX_NOTHROW_PURE_FUNCTION static inline size_t page_numkeys(const page_t *mp) {
+  assert(mp->lower <= mp->upper);
+  return mp->lower >> 1;
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline size_t page_room(const page_t *mp) {
+  assert(mp->lower <= mp->upper);
+  return mp->upper - mp->lower;
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline size_t page_space(const MDBX_env *env) {
+  STATIC_ASSERT(PAGEHDRSZ % 2 == 0);
+  return env->ps - PAGEHDRSZ;
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline size_t page_used(const MDBX_env *env, const page_t *mp) {
+  return page_space(env) - page_room(mp);
+}
+
+/* The percentage of space used in the page, in a percents. */
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline unsigned page_fill_percentum_x10(const MDBX_env *env,
+                                                                                            const page_t *mp) {
+  const size_t space = page_space(env);
+  return (unsigned)((page_used(env, mp) * 1000 + space / 2) / space);
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline node_t *page_node(const page_t *mp, size_t i) {
+  assert(page_type_compat(mp) == P_LEAF || page_type(mp) == P_BRANCH);
+  assert(page_numkeys(mp) > i);
+  assert(mp->entries[i] % 2 == 0);
+  return ptr_disp(mp, mp->entries[i] + PAGEHDRSZ);
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline void *page_dupfix_ptr(const page_t *mp, size_t i, size_t keysize) {
+  assert(page_type_compat(mp) == (P_LEAF | P_DUPFIX) && i == (indx_t)i && mp->dupfix_ksize == keysize);
+  (void)keysize;
+  return ptr_disp(mp, PAGEHDRSZ + mp->dupfix_ksize * (indx_t)i);
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline MDBX_val page_dupfix_key(const page_t *mp, size_t i, size_t keysize) {
+  MDBX_val r;
+  r.iov_base = page_dupfix_ptr(mp, i, keysize);
+  r.iov_len = mp->dupfix_ksize;
+  return r;
+}
+
+/*----------------------------------------------------------------------------*/
+
+MDBX_NOTHROW_PURE_FUNCTION MDBX_INTERNAL int cmp_int_unaligned(const MDBX_val *a, const MDBX_val *b);
+
+#if MDBX_UNALIGNED_OK < 2 || (MDBX_DEBUG || MDBX_FORCE_ASSERTIONS || !defined(NDEBUG))
+MDBX_NOTHROW_PURE_FUNCTION MDBX_INTERNAL int
+/* Compare two items pointing at 2-byte aligned unsigned int's. */
+cmp_int_align2(const MDBX_val *a, const MDBX_val *b);
+#else
+#define cmp_int_align2 cmp_int_unaligned
+#endif /* !MDBX_UNALIGNED_OK || debug */
+
+#if MDBX_UNALIGNED_OK < 4 || (MDBX_DEBUG || MDBX_FORCE_ASSERTIONS || !defined(NDEBUG))
+MDBX_NOTHROW_PURE_FUNCTION MDBX_INTERNAL int
+/* Compare two items pointing at 4-byte aligned unsigned int's. */
+cmp_int_align4(const MDBX_val *a, const MDBX_val *b);
+#else
+#define cmp_int_align4 cmp_int_unaligned
+#endif /* !MDBX_UNALIGNED_OK || debug */
+
+/* Compare two items lexically */
+MDBX_NOTHROW_PURE_FUNCTION MDBX_INTERNAL int cmp_lexical(const MDBX_val *a, const MDBX_val *b);
+
+/* Compare two items in reverse byte order */
+MDBX_NOTHROW_PURE_FUNCTION MDBX_INTERNAL int cmp_reverse(const MDBX_val *a, const MDBX_val *b);
+
+/* Fast non-lexically comparator */
+MDBX_NOTHROW_PURE_FUNCTION MDBX_INTERNAL int cmp_lenfast(const MDBX_val *a, const MDBX_val *b);
+
+MDBX_NOTHROW_PURE_FUNCTION MDBX_INTERNAL bool eq_fast_slowpath(const uint8_t *a, const uint8_t *b, size_t l);
+
+MDBX_NOTHROW_PURE_FUNCTION static inline bool eq_fast(const MDBX_val *a, const MDBX_val *b) {
+  return unlikely(a->iov_len == b->iov_len) && eq_fast_slowpath(a->iov_base, b->iov_base, a->iov_len);
+}
+
+MDBX_NOTHROW_PURE_FUNCTION MDBX_INTERNAL int cmp_equal_or_greater(const MDBX_val *a, const MDBX_val *b);
+
+MDBX_NOTHROW_PURE_FUNCTION MDBX_INTERNAL int cmp_equal_or_wrong(const MDBX_val *a, const MDBX_val *b);
+
+static inline MDBX_cmp_func *builtin_keycmp(MDBX_db_flags_t flags) {
+  return (flags & MDBX_REVERSEKEY) ? cmp_reverse : (flags & MDBX_INTEGERKEY) ? cmp_int_align2 : cmp_lexical;
+}
+
+static inline MDBX_cmp_func *builtin_datacmp(MDBX_db_flags_t flags) {
+  return !(flags & MDBX_DUPSORT)
+             ? cmp_lenfast
+             : ((flags & MDBX_INTEGERDUP) ? cmp_int_unaligned
+                                          : ((flags & MDBX_REVERSEDUP) ? cmp_reverse : cmp_lexical));
+}
+
+/*----------------------------------------------------------------------------*/
+
+MDBX_INTERNAL uint32_t combine_durability_flags(const uint32_t a, const uint32_t b);
+
+MDBX_CONST_FUNCTION static inline lck_t *lckless_stub(const MDBX_env *env) {
+  uintptr_t stub = (uintptr_t)&env->lckless_placeholder;
+  /* align to avoid false-positive alarm from UndefinedBehaviorSanitizer */
+  stub = (stub + MDBX_CACHELINE_SIZE - 1) & ~(MDBX_CACHELINE_SIZE - 1);
+  return (lck_t *)stub;
+}
+
+#if !(defined(_WIN32) || defined(_WIN64))
+MDBX_CONST_FUNCTION static inline int ignore_enosys(int err) {
+#ifdef ENOSYS
+  if (err == ENOSYS)
+    return MDBX_RESULT_TRUE;
+#endif /* ENOSYS */
+#ifdef ENOIMPL
+  if (err == ENOIMPL)
+    return MDBX_RESULT_TRUE;
+#endif /* ENOIMPL */
+#ifdef ENOTSUP
+  if (err == ENOTSUP)
+    return MDBX_RESULT_TRUE;
+#endif /* ENOTSUP */
+#ifdef ENOSUPP
+  if (err == ENOSUPP)
+    return MDBX_RESULT_TRUE;
+#endif /* ENOSUPP */
+#ifdef EOPNOTSUPP
+  if (err == EOPNOTSUPP)
+    return MDBX_RESULT_TRUE;
+#endif /* EOPNOTSUPP */
+  return err;
+}
+
+MDBX_MAYBE_UNUSED MDBX_CONST_FUNCTION static inline int ignore_enosys_and_eagain(int err) {
+  return (err == EAGAIN) ? MDBX_RESULT_TRUE : ignore_enosys(err);
+}
+
+MDBX_MAYBE_UNUSED MDBX_CONST_FUNCTION static inline int ignore_enosys_and_einval(int err) {
+  return (err == EINVAL) ? MDBX_RESULT_TRUE : ignore_enosys(err);
+}
+#endif /* defined(_WIN32) || defined(_WIN64) */
+
+static inline int check_env(const MDBX_env *env, const bool wanna_active) {
+  if (unlikely(!env))
+    return MDBX_EINVAL;
+
+  if (unlikely(env->signature.weak != env_signature))
+    return MDBX_EBADSIGN;
+
+  if (unlikely(env->flags & ENV_FATAL_ERROR))
+    return MDBX_PANIC;
+
+  if (wanna_active) {
+#if MDBX_ENV_CHECKPID
+    if (unlikely(env->pid != osal_getpid()) && env->pid) {
+      ((MDBX_env *)env)->flags |= ENV_FATAL_ERROR;
+      return MDBX_PANIC;
+    }
+#endif /* MDBX_ENV_CHECKPID */
+    if (unlikely((env->flags & ENV_ACTIVE) == 0))
+      return MDBX_EPERM;
+    eASSERT(env, env->dxb_mmap.base != nullptr);
+  }
+
+  return MDBX_SUCCESS;
+}
+
+static __always_inline int check_txn(const MDBX_txn *txn, int bad_bits) {
+  if (unlikely(!txn))
+    return MDBX_EINVAL;
+
+  if (unlikely(txn->signature != txn_signature))
+    return MDBX_EBADSIGN;
+
+  if (bad_bits) {
+    if (unlikely(!txn->env->dxb_mmap.base))
+      return MDBX_EPERM;
+
+    if (unlikely(txn->flags & bad_bits)) {
+      if ((bad_bits & MDBX_TXN_RDONLY) && unlikely(txn->flags & MDBX_TXN_RDONLY))
+        return MDBX_EACCESS;
+      if ((bad_bits & MDBX_TXN_PARKED) == 0)
+        return MDBX_BAD_TXN;
+      return txn_check_badbits_parked(txn, bad_bits);
+    }
+  }
+
+  tASSERT(txn, (txn->flags & MDBX_TXN_FINISHED) ||
+                   (txn->flags & MDBX_NOSTICKYTHREADS) == (txn->env->flags & MDBX_NOSTICKYTHREADS));
+#if MDBX_TXN_CHECKOWNER
+  if ((txn->flags & (MDBX_NOSTICKYTHREADS | MDBX_TXN_FINISHED)) != MDBX_NOSTICKYTHREADS &&
+      !(bad_bits /* abort/reset/txn-break */ == 0 &&
+        ((txn->flags & (MDBX_TXN_RDONLY | MDBX_TXN_FINISHED)) == (MDBX_TXN_RDONLY | MDBX_TXN_FINISHED))) &&
+      unlikely(txn->owner != osal_thread_self()))
+    return txn->owner ? MDBX_THREAD_MISMATCH : MDBX_BAD_TXN;
+#endif /* MDBX_TXN_CHECKOWNER */
+
+  return MDBX_SUCCESS;
+}
+
+static inline int check_txn_rw(const MDBX_txn *txn, int bad_bits) {
+  return check_txn(txn, (bad_bits | MDBX_TXN_RDONLY) & ~MDBX_TXN_PARKED);
+}
+
+/*----------------------------------------------------------------------------*/
+
+MDBX_INTERNAL void mincore_clean_cache(const MDBX_env *const env);
+
+MDBX_INTERNAL void update_mlcnt(const MDBX_env *env, const pgno_t new_aligned_mlocked_pgno,
+                                const bool lock_not_release);
+
+MDBX_INTERNAL void munlock_after(const MDBX_env *env, const pgno_t aligned_pgno, const size_t end_bytes);
+
+MDBX_INTERNAL void munlock_all(const MDBX_env *env);
+
+/*----------------------------------------------------------------------------*/
+/* Cache coherence and mmap invalidation */
+#ifndef MDBX_CPU_WRITEBACK_INCOHERENT
+#error "The MDBX_CPU_WRITEBACK_INCOHERENT must be defined before"
+#elif MDBX_CPU_WRITEBACK_INCOHERENT
+#define osal_flush_incoherent_cpu_writeback() osal_memory_barrier()
+#else
+#define osal_flush_incoherent_cpu_writeback() osal_compiler_barrier()
+#endif /* MDBX_CPU_WRITEBACK_INCOHERENT */
+
+MDBX_MAYBE_UNUSED static inline void osal_flush_incoherent_mmap(const void *addr, size_t nbytes,
+                                                                const intptr_t pagesize) {
+#ifndef MDBX_MMAP_INCOHERENT_FILE_WRITE
+#error "The MDBX_MMAP_INCOHERENT_FILE_WRITE must be defined before"
+#elif MDBX_MMAP_INCOHERENT_FILE_WRITE
+  char *const begin = (char *)(-pagesize & (intptr_t)addr);
+  char *const end = (char *)(-pagesize & (intptr_t)((char *)addr + nbytes + pagesize - 1));
+  int err = msync(begin, end - begin, MS_SYNC | MS_INVALIDATE) ? errno : 0;
+  eASSERT(nullptr, err == 0);
+  (void)err;
+#else
+  (void)pagesize;
+#endif /* MDBX_MMAP_INCOHERENT_FILE_WRITE */
+
+#ifndef MDBX_MMAP_INCOHERENT_CPU_CACHE
+#error "The MDBX_MMAP_INCOHERENT_CPU_CACHE must be defined before"
+#elif MDBX_MMAP_INCOHERENT_CPU_CACHE
+#ifdef DCACHE
+  /* MIPS has cache coherency issues.
+   * Note: for any nbytes >= on-chip cache size, entire is flushed. */
+  cacheflush((void *)addr, nbytes, DCACHE);
+#else
+#error "Oops, cacheflush() not available"
+#endif /* DCACHE */
+#endif /* MDBX_MMAP_INCOHERENT_CPU_CACHE */
+
+#if !MDBX_MMAP_INCOHERENT_FILE_WRITE && !MDBX_MMAP_INCOHERENT_CPU_CACHE
+  (void)addr;
+  (void)nbytes;
+#endif
+}

src/coherency.c

@@ -0,0 +1,170 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+/* check against https://libmdbx.dqdkfa.ru/dead-github/issues/269 */
+static bool coherency_check(const MDBX_env *env, const txnid_t txnid, const volatile tree_t *trees,
+                            const volatile meta_t *meta, bool report) {
+  const txnid_t freedb_mod_txnid = trees[FREE_DBI].mod_txnid;
+  const txnid_t maindb_mod_txnid = trees[MAIN_DBI].mod_txnid;
+  const pgno_t last_pgno = meta->geometry.now;
+
+  const pgno_t freedb_root_pgno = trees[FREE_DBI].root;
+  const page_t *freedb_root =
+      (env->dxb_mmap.base && freedb_root_pgno < last_pgno) ? pgno2page(env, freedb_root_pgno) : nullptr;
+
+  const pgno_t maindb_root_pgno = trees[MAIN_DBI].root;
+  const page_t *maindb_root =
+      (env->dxb_mmap.base && maindb_root_pgno < last_pgno) ? pgno2page(env, maindb_root_pgno) : nullptr;
+  const uint64_t magic_and_version = unaligned_peek_u64_volatile(4, &meta->magic_and_version);
+
+  bool ok = true;
+  if (freedb_root_pgno != P_INVALID && unlikely(freedb_root_pgno >= last_pgno)) {
+    if (report)
+      WARNING("catch invalid %s-db root %" PRIaPGNO " for meta_txnid %" PRIaTXN " %s", "free", freedb_root_pgno, txnid,
+              (env->stuck_meta < 0) ? "(workaround for incoherent flaw of unified page/buffer cache)"
+                                    : "(wagering meta)");
+    ok = false;
+  }
+  if (maindb_root_pgno != P_INVALID && unlikely(maindb_root_pgno >= last_pgno)) {
+    if (report)
+      WARNING("catch invalid %s-db root %" PRIaPGNO " for meta_txnid %" PRIaTXN " %s", "main", maindb_root_pgno, txnid,
+              (env->stuck_meta < 0) ? "(workaround for incoherent flaw of unified page/buffer cache)"
+                                    : "(wagering meta)");
+    ok = false;
+  }
+  if (unlikely(txnid < freedb_mod_txnid ||
+               (!freedb_mod_txnid && freedb_root && likely(magic_and_version == MDBX_DATA_MAGIC)))) {
+    if (report)
+      WARNING(
+          "catch invalid %s-db.mod_txnid %" PRIaTXN " for meta_txnid %" PRIaTXN " %s", "free", freedb_mod_txnid, txnid,
+          (env->stuck_meta < 0) ? "(workaround for incoherent flaw of unified page/buffer cache)" : "(wagering meta)");
+    ok = false;
+  }
+  if (unlikely(txnid < maindb_mod_txnid ||
+               (!maindb_mod_txnid && maindb_root && likely(magic_and_version == MDBX_DATA_MAGIC)))) {
+    if (report)
+      WARNING(
+          "catch invalid %s-db.mod_txnid %" PRIaTXN " for meta_txnid %" PRIaTXN " %s", "main", maindb_mod_txnid, txnid,
+          (env->stuck_meta < 0) ? "(workaround for incoherent flaw of unified page/buffer cache)" : "(wagering meta)");
+    ok = false;
+  }
+
+  /* Проверяем отметки внутри корневых страниц только если сами страницы
+   * в пределах текущего отображения. Иначе возможны SIGSEGV до переноса
+   * вызова coherency_check_head() после dxb_resize() внутри txn_renew(). */
+  if (likely(freedb_root && freedb_mod_txnid &&
+             (size_t)ptr_dist(env->dxb_mmap.base, freedb_root) < env->dxb_mmap.limit)) {
+    VALGRIND_MAKE_MEM_DEFINED(freedb_root, sizeof(freedb_root->txnid));
+    MDBX_ASAN_UNPOISON_MEMORY_REGION(freedb_root, sizeof(freedb_root->txnid));
+    const txnid_t root_txnid = freedb_root->txnid;
+    if (unlikely(root_txnid != freedb_mod_txnid)) {
+      if (report)
+        WARNING("catch invalid root_page %" PRIaPGNO " mod_txnid %" PRIaTXN " for %s-db.mod_txnid %" PRIaTXN " %s",
+                freedb_root_pgno, root_txnid, "free", freedb_mod_txnid,
+                (env->stuck_meta < 0) ? "(workaround for incoherent flaw of "
+                                        "unified page/buffer cache)"
+                                      : "(wagering meta)");
+      ok = false;
+    }
+  }
+  if (likely(maindb_root && maindb_mod_txnid &&
+             (size_t)ptr_dist(env->dxb_mmap.base, maindb_root) < env->dxb_mmap.limit)) {
+    VALGRIND_MAKE_MEM_DEFINED(maindb_root, sizeof(maindb_root->txnid));
+    MDBX_ASAN_UNPOISON_MEMORY_REGION(maindb_root, sizeof(maindb_root->txnid));
+    const txnid_t root_txnid = maindb_root->txnid;
+    if (unlikely(root_txnid != maindb_mod_txnid)) {
+      if (report)
+        WARNING("catch invalid root_page %" PRIaPGNO " mod_txnid %" PRIaTXN " for %s-db.mod_txnid %" PRIaTXN " %s",
+                maindb_root_pgno, root_txnid, "main", maindb_mod_txnid,
+                (env->stuck_meta < 0) ? "(workaround for incoherent flaw of "
+                                        "unified page/buffer cache)"
+                                      : "(wagering meta)");
+      ok = false;
+    }
+  }
+  if (unlikely(!ok) && report)
+    env->lck->pgops.incoherence.weak =
+        (env->lck->pgops.incoherence.weak >= INT32_MAX) ? INT32_MAX : env->lck->pgops.incoherence.weak + 1;
+  return ok;
+}
+
+__cold int coherency_timeout(uint64_t *timestamp, intptr_t pgno, const MDBX_env *env) {
+  if (likely(timestamp && *timestamp == 0))
+    *timestamp = osal_monotime();
+  else if (unlikely(!timestamp || osal_monotime() - *timestamp > osal_16dot16_to_monotime(65536 / 10))) {
+    if (pgno >= 0 && pgno != env->stuck_meta)
+      ERROR("bailout waiting for %" PRIuSIZE " page arrival %s", pgno,
+            "(workaround for incoherent flaw of unified page/buffer cache)");
+    else if (env->stuck_meta < 0)
+      ERROR("bailout waiting for valid snapshot (%s)", "workaround for incoherent flaw of unified page/buffer cache");
+    return MDBX_PROBLEM;
+  }
+
+  osal_memory_fence(mo_AcquireRelease, true);
+#if defined(_WIN32) || defined(_WIN64)
+  SwitchToThread();
+#elif defined(__linux__) || defined(__gnu_linux__) || defined(_UNIX03_SOURCE)
+  sched_yield();
+#elif (defined(_GNU_SOURCE) && __GLIBC_PREREQ(2, 1)) || defined(_OPEN_THREADS)
+  pthread_yield();
+#else
+  usleep(42);
+#endif
+  return MDBX_RESULT_TRUE;
+}
+
+/* check with timeout as the workaround
+ * for https://libmdbx.dqdkfa.ru/dead-github/issues/269 */
+__hot int coherency_fetch_head(MDBX_txn *txn, const meta_ptr_t head, uint64_t *timestamp) {
+  /* Copy the DB info and flags */
+  txn->txnid = head.txnid;
+  txn->geo = head.ptr_c->geometry;
+  memcpy(txn->dbs, &head.ptr_c->trees, sizeof(head.ptr_c->trees));
+  STATIC_ASSERT(sizeof(head.ptr_c->trees) == CORE_DBS * sizeof(tree_t));
+  VALGRIND_MAKE_MEM_UNDEFINED(txn->dbs + CORE_DBS, txn->env->max_dbi - CORE_DBS);
+  txn->canary = head.ptr_c->canary;
+
+  if (unlikely(!coherency_check(txn->env, head.txnid, txn->dbs, head.ptr_v, *timestamp == 0) ||
+               txn->txnid != meta_txnid(head.ptr_v)))
+    return coherency_timeout(timestamp, -1, txn->env);
+
+  if (unlikely(txn->dbs[FREE_DBI].flags != MDBX_INTEGERKEY)) {
+    if ((txn->dbs[FREE_DBI].flags & DB_PERSISTENT_FLAGS) != MDBX_INTEGERKEY ||
+        unaligned_peek_u64(4, &head.ptr_c->magic_and_version) == MDBX_DATA_MAGIC) {
+      ERROR("unexpected/invalid db-flags 0x%x for %s", txn->dbs[FREE_DBI].flags, "GC/FreeDB");
+      return MDBX_INCOMPATIBLE;
+    }
+    txn->dbs[FREE_DBI].flags &= DB_PERSISTENT_FLAGS;
+  }
+  tASSERT(txn, txn->dbs[FREE_DBI].flags == MDBX_INTEGERKEY);
+  tASSERT(txn, check_table_flags(txn->dbs[MAIN_DBI].flags));
+  return MDBX_SUCCESS;
+}
+
+int coherency_check_written(const MDBX_env *env, const txnid_t txnid, const volatile meta_t *meta, const intptr_t pgno,
+                            uint64_t *timestamp) {
+  const bool report = !(timestamp && *timestamp);
+  const txnid_t head_txnid = meta_txnid(meta);
+  if (likely(head_txnid >= MIN_TXNID && head_txnid >= txnid)) {
+    if (likely(coherency_check(env, head_txnid, &meta->trees.gc, meta, report))) {
+      eASSERT(env, meta->trees.gc.flags == MDBX_INTEGERKEY);
+      eASSERT(env, check_table_flags(meta->trees.main.flags));
+      return MDBX_SUCCESS;
+    }
+  } else if (report) {
+    env->lck->pgops.incoherence.weak =
+        (env->lck->pgops.incoherence.weak >= INT32_MAX) ? INT32_MAX : env->lck->pgops.incoherence.weak + 1;
+    WARNING("catch %s txnid %" PRIaTXN " for meta_%" PRIaPGNO " %s",
+            (head_txnid < MIN_TXNID) ? "invalid" : "unexpected", head_txnid,
+            bytes2pgno(env, ptr_dist(meta, env->dxb_mmap.base)),
+            "(workaround for incoherent flaw of unified page/buffer cache)");
+  }
+  return coherency_timeout(timestamp, pgno, env);
+}
+
+bool coherency_check_meta(const MDBX_env *env, const volatile meta_t *meta, bool report) {
+  uint64_t timestamp = 0;
+  return coherency_check_written(env, 0, meta, -1, report ? &timestamp : nullptr) == MDBX_SUCCESS;
+}

src/config.h.in

@@ -0,0 +1,88 @@
+/* This is CMake-template for libmdbx's config.h
+ ******************************************************************************/
+
+/* *INDENT-OFF* */
+/* clang-format off */
+
+#cmakedefine LTO_ENABLED
+#cmakedefine ENABLE_MEMCHECK
+#cmakedefine ENABLE_GPROF
+#cmakedefine ENABLE_GCOV
+#cmakedefine ENABLE_ASAN
+#cmakedefine ENABLE_UBSAN
+#cmakedefine01 MDBX_FORCE_ASSERTIONS
+#if !defined(MDBX_BUILD_TEST) && !defined(MDBX_BUILD_CXX)
+#cmakedefine01 MDBX_BUILD_CXX
+#endif
+
+/* Common */
+#cmakedefine01 MDBX_TXN_CHECKOWNER
+#cmakedefine MDBX_ENV_CHECKPID_AUTO
+#ifndef MDBX_ENV_CHECKPID_AUTO
+#cmakedefine01 MDBX_ENV_CHECKPID
+#endif
+#cmakedefine MDBX_LOCKING_AUTO
+#ifndef MDBX_LOCKING_AUTO
+#cmakedefine MDBX_LOCKING @MDBX_LOCKING@
+#endif
+#cmakedefine MDBX_TRUST_RTC_AUTO
+#ifndef MDBX_TRUST_RTC_AUTO
+#cmakedefine01 MDBX_TRUST_RTC
+#endif
+#cmakedefine01 MDBX_DISABLE_VALIDATION
+#cmakedefine01 MDBX_AVOID_MSYNC
+#cmakedefine01 MDBX_ENABLE_REFUND
+#cmakedefine01 MDBX_ENABLE_BIGFOOT
+#cmakedefine01 MDBX_ENABLE_PGOP_STAT
+#cmakedefine01 MDBX_ENABLE_PROFGC
+#cmakedefine01 MDBX_ENABLE_DBI_SPARSE
+#cmakedefine01 MDBX_ENABLE_DBI_LOCKFREE
+
+/* Windows */
+#if defined(MDBX_BUILD_TEST) || !defined(MDBX_BUILD_CXX) || MDBX_BUILD_CXX
+#define MDBX_WITHOUT_MSVC_CRT 0
+#else
+#cmakedefine01 MDBX_WITHOUT_MSVC_CRT
+#endif /* MDBX_WITHOUT_MSVC_CRT */
+
+/* MacOS & iOS */
+#cmakedefine01 MDBX_APPLE_SPEED_INSTEADOF_DURABILITY
+
+/* POSIX */
+#cmakedefine01 MDBX_DISABLE_GNU_SOURCE
+
+#cmakedefine MDBX_USE_OFDLOCKS_AUTO
+#ifndef MDBX_USE_OFDLOCKS_AUTO
+#cmakedefine01 MDBX_USE_OFDLOCKS
+#endif /* MDBX_USE_OFDLOCKS */
+
+#cmakedefine MDBX_MMAP_NEEDS_JOLT_AUTO
+#ifndef MDBX_MMAP_NEEDS_JOLT_AUTO
+#cmakedefine01 MDBX_MMAP_NEEDS_JOLT
+#endif /* MDBX_MMAP_NEEDS_JOLT */
+
+#cmakedefine01 MDBX_USE_MINCORE
+
+/* Build Info */
+#ifndef MDBX_BUILD_TIMESTAMP
+#cmakedefine MDBX_BUILD_TIMESTAMP "@MDBX_BUILD_TIMESTAMP@"
+#endif
+#ifndef MDBX_BUILD_TARGET
+#cmakedefine MDBX_BUILD_TARGET "@MDBX_BUILD_TARGET@"
+#endif
+#ifndef MDBX_BUILD_TYPE
+#cmakedefine MDBX_BUILD_TYPE "@MDBX_BUILD_TYPE@"
+#endif
+#ifndef MDBX_BUILD_COMPILER
+#cmakedefine MDBX_BUILD_COMPILER "@MDBX_BUILD_COMPILER@"
+#endif
+#ifndef MDBX_BUILD_FLAGS
+#cmakedefine MDBX_BUILD_FLAGS "@MDBX_BUILD_FLAGS@"
+#endif
+#ifndef MDBX_BUILD_METADATA
+#cmakedefine MDBX_BUILD_METADATA "@MDBX_BUILD_METADATA@"
+#endif
+#cmakedefine MDBX_BUILD_SOURCERY @MDBX_BUILD_SOURCERY@
+
+/* *INDENT-ON* */
+/* clang-format on */

src/cursor.h

@@ -0,0 +1,391 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#pragma once
+
+#include "essentials.h"
+
+/* Состояние курсора.
+ *
+ * плохой/poor:
+ *  - неустановленный курсор с незаполненым стеком;
+ *  - следует пропускать во всех циклах отслеживания/корректировки
+ *    позиций курсоров;
+ *  - допускаются только операции предполагающие установку абсолютной позиции;
+ *  - в остальных случаях возвращается ENODATA.
+ *
+ *    У таких курсоров top = -1 и flags < 0, что позволяет дешево проверять и
+ *    пропускать такие курсоры в циклах отслеживания/корректировки по условию
+ *    probe_cursor->top < this_cursor->top.
+ *
+ * пустой/hollow:
+ *  - частично инициализированный курсор, но без доступной пользователю позиции,
+ *    поэтому нельзя выполнить какую-либо операцию без абсолютного (не
+ *    относительного) позиционирования;
+ *  - ki[top] может быть некорректным, в том числе >= page_numkeys(pg[top]).
+ *
+ *    У таких курсоров top >= 0, но flags < 0 (есть флажок z_hollow).
+ *
+ * установленный/pointed:
+ *  - полностью инициализированный курсор с конкретной позицией с данными;
+ *  - можно прочитать текущую строку, удалить её, либо выполнить
+ *    относительное перемещение;
+ *  - может иметь флажки z_after_delete, z_eof_hard и z_eof_soft;
+ *  - наличие z_eof_soft означает что курсор перемещен за пределы данных,
+ *    поэтому нелья прочитать текущие данные, либо удалить их.
+ *
+ *    У таких курсоров top >= 0 и flags >= 0 (нет флажка z_hollow).
+ *
+ * наполненный данными/filled:
+ *  - это установленный/pointed курсор без флагов z_eof_soft;
+ *  - за курсором есть даные, возможны CRUD операции в текущей позиции.
+ *
+ *    У таких курсоров top >= 0 и (unsigned)flags < z_eof_soft.
+ *
+ * Изменения состояния.
+ *
+ *  - Сбрасывается состояние курсора посредством top_and_flags |= z_poor_mark,
+ *    что равносильно top = -1 вместе с flags |= z_poor_mark;
+ *  - При позиционировании курсора сначала устанавливается top, а flags
+ *    только в самом конце при отсутстви ошибок.
+ *  - Повторное позиционирование first/last может начинаться
+ *    с установки/обнуления только top без сброса flags, что позволяет работать
+ *    быстрому пути внутри tree_search_finalize().
+ *
+ *  - Заморочки с концом данных:
+ *     - mdbx_cursor_get(NEXT) выполняет две операции (перемещение и чтение),
+ *       поэтому перемещение на последнюю строку строку всегда успешно,
+ *       а ошибка возвращается только при последующем next().
+ *       Однако, из-за этой двойственности семантика ситуации возврата ошибки
+ *       из mdbx_cursor_get(NEXT) допускает разночтение/неопределенность, ибо
+ *       не понятно к чему относится ошибка:
+ *        - Если к чтению данных, то курсор перемещен и стоит после последней
+ *          строки. Соответственно, чтение в текущей позиции запрещено,
+ *          а при выполнении prev() курсор вернется на последнюю строку;
+ *        - Если же ошибка относится к перемещению, то курсор не перемещен и
+ *          остается на последней строке. Соответственно, чтение в текущей
+ *          позиции допустимо, а при выполнении prev() курсор встанет
+ *          на пред-последнюю строку.
+ *        - Пикантность в том, что пользователи (так или иначе) полагаются
+ *          на оба варианта поведения, при этом конечно ожидают что после
+ *          ошибки MDBX_NEXT функция mdbx_cursor_eof() будет возвращать true.
+ *     - далее добавляется схожая ситуация с MDBX_GET_RANGE, MDBX_LOWERBOUND,
+ *       MDBX_GET_BOTH_RANGE и MDBX_UPPERBOUND. Тут при неуспехе поиска курсор
+ *       может/должен стоять после последней строки.
+ *     - далее добавляется MDBX_LAST. Тут курсор должен стоять на последней
+ *       строке и допускать чтение в текузщей позиции,
+ *       но mdbx_cursor_eof() должен возвращать true.
+ *
+ *    Решение = делаем два флажка z_eof_soft и z_eof_hard:
+ *     - Когда установлен только z_eof_soft,
+ *       функция mdbx_cursor_eof() возвращает true, но допускается
+ *       чтение данных в текущей позиции, а prev() передвигает курсор
+ *       на пред-последнюю строку.
+ *     - Когда установлен z_eof_hard, чтение данных в текущей позиции
+ *       не допускается, и mdbx_cursor_eof() также возвращает true,
+ *       а prev() устанавливает курсора на последюю строку. */
+enum cursor_state {
+  /* Это вложенный курсор для вложенного дерева/страницы и является
+     inner-элементом struct cursor_couple. */
+  z_inner = 0x01,
+
+  /* Происходит подготовка к обновлению GC,
+     поэтому можно брать страницы из GC даже для FREE_DBI. */
+  z_gcu_preparation = 0x02,
+
+  /* Курсор только-что создан, поэтому допускается авто-установка
+     в начало/конец, вместо возврата ошибки. */
+  z_fresh = 0x04,
+
+  /* Предыдущей операцией было удаление, поэтому курсор уже физически указывает
+     на следующий элемент и соответствующая операция перемещения должна
+     игнорироваться. */
+  z_after_delete = 0x08,
+
+  /* */
+  z_disable_tree_search_fastpath = 0x10,
+
+  /* Курсор логически в конце данных, но физически на последней строке,
+   * ki[top] == page_numkeys(pg[top]) - 1 и читать данные в текущей позиции. */
+  z_eof_soft = 0x20,
+
+  /* Курсор логически за концом данных, поэтому следующий переход "назад"
+     должен игнорироваться и/или приводить к установке на последнюю строку.
+     В текущем же состоянии нельзя делать CRUD операции. */
+  z_eof_hard = 0x40,
+
+  /* За курсором нет данных, логически его позиция не определена,
+     нельзя делать CRUD операции в текущей позиции.
+     Относительное перемещение запрещено. */
+  z_hollow = -128 /* 0x80 */,
+
+  /* Маски для сброса/установки состояния. */
+  z_clear_mask = z_inner | z_gcu_preparation,
+  z_poor_mark = z_eof_hard | z_hollow | z_disable_tree_search_fastpath,
+  z_fresh_mark = z_poor_mark | z_fresh
+};
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool is_inner(const MDBX_cursor *mc) {
+  return (mc->flags & z_inner) != 0;
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool is_poor(const MDBX_cursor *mc) {
+  const bool r = mc->top < 0;
+  cASSERT(mc, r == (mc->top_and_flags < 0));
+  if (r && mc->subcur)
+    cASSERT(mc, mc->subcur->cursor.flags < 0 && mc->subcur->cursor.top < 0);
+  return r;
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool is_pointed(const MDBX_cursor *mc) {
+  const bool r = mc->top >= 0;
+  cASSERT(mc, r == (mc->top_and_flags >= 0));
+  if (!r && mc->subcur)
+    cASSERT(mc, is_poor(&mc->subcur->cursor));
+  return r;
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool is_hollow(const MDBX_cursor *mc) {
+  const bool r = mc->flags < 0;
+  if (!r) {
+    cASSERT(mc, mc->top >= 0);
+    cASSERT(mc, (mc->flags & z_eof_hard) || mc->ki[mc->top] < page_numkeys(mc->pg[mc->top]));
+  } else if (mc->subcur)
+    cASSERT(mc, is_poor(&mc->subcur->cursor) || (is_pointed(mc) && mc->subcur->cursor.flags < 0));
+  return r;
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool is_eof(const MDBX_cursor *mc) {
+  const bool r = z_eof_soft <= (uint8_t)mc->flags;
+  return r;
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool is_filled(const MDBX_cursor *mc) {
+  const bool r = z_eof_hard > (uint8_t)mc->flags;
+  return r;
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool inner_filled(const MDBX_cursor *mc) {
+  return mc->subcur && is_filled(&mc->subcur->cursor);
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool inner_pointed(const MDBX_cursor *mc) {
+  return mc->subcur && is_pointed(&mc->subcur->cursor);
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool inner_hollow(const MDBX_cursor *mc) {
+  const bool r = !mc->subcur || is_hollow(&mc->subcur->cursor);
+#if MDBX_DEBUG || MDBX_FORCE_ASSERTIONS
+  if (!r) {
+    cASSERT(mc, is_filled(mc));
+    const page_t *mp = mc->pg[mc->top];
+    const node_t *node = page_node(mp, mc->ki[mc->top]);
+    cASSERT(mc, node_flags(node) & N_DUP);
+  }
+#endif /* MDBX_DEBUG || MDBX_FORCE_ASSERTIONS */
+  return r;
+}
+
+MDBX_MAYBE_UNUSED static inline void inner_gone(MDBX_cursor *mc) {
+  if (mc->subcur) {
+    TRACE("reset inner cursor %p", __Wpedantic_format_voidptr(&mc->subcur->cursor));
+    mc->subcur->nested_tree.root = 0;
+    mc->subcur->cursor.top_and_flags = z_inner | z_poor_mark;
+  }
+}
+
+MDBX_MAYBE_UNUSED static inline void be_poor(MDBX_cursor *mc) {
+  const bool inner = is_inner(mc);
+  if (inner) {
+    mc->tree->root = 0;
+    mc->top_and_flags = z_inner | z_poor_mark;
+  } else {
+    mc->top_and_flags |= z_poor_mark;
+    inner_gone(mc);
+  }
+  cASSERT(mc, is_poor(mc) && !is_pointed(mc) && !is_filled(mc));
+  cASSERT(mc, inner == is_inner(mc));
+}
+
+MDBX_MAYBE_UNUSED static inline void be_filled(MDBX_cursor *mc) {
+  cASSERT(mc, mc->top >= 0);
+  cASSERT(mc, mc->ki[mc->top] < page_numkeys(mc->pg[mc->top]));
+  const bool inner = is_inner(mc);
+  mc->flags &= z_clear_mask;
+  cASSERT(mc, is_filled(mc));
+  cASSERT(mc, inner == is_inner(mc));
+}
+
+MDBX_MAYBE_UNUSED static inline bool is_related(const MDBX_cursor *base, const MDBX_cursor *scan) {
+  cASSERT(base, base->top >= 0);
+  return base->top <= scan->top && base != scan;
+}
+
+/* Флаги контроля/проверки курсора. */
+enum cursor_checking {
+  z_branch = 0x01 /* same as P_BRANCH for check_leaf_type() */,
+  z_leaf = 0x02 /* same as P_LEAF for check_leaf_type() */,
+  z_largepage = 0x04 /* same as P_LARGE for check_leaf_type() */,
+  z_updating = 0x08 /* update/rebalance pending */,
+  z_ignord = 0x10 /* don't check keys ordering */,
+  z_dupfix = 0x20 /* same as P_DUPFIX for check_leaf_type() */,
+  z_retiring = 0x40 /* refs to child pages may be invalid */,
+  z_pagecheck = 0x80 /* perform page checking, see MDBX_VALIDATION */
+};
+
+MDBX_INTERNAL int __must_check_result cursor_validate(const MDBX_cursor *mc);
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline size_t cursor_dbi(const MDBX_cursor *mc) {
+  cASSERT(mc, mc->txn && mc->txn->signature == txn_signature);
+  size_t dbi = mc->dbi_state - mc->txn->dbi_state;
+  cASSERT(mc, dbi < mc->txn->env->n_dbi);
+  return dbi;
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool cursor_dbi_changed(const MDBX_cursor *mc) {
+  return dbi_changed(mc->txn, cursor_dbi(mc));
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline uint8_t *cursor_dbi_state(const MDBX_cursor *mc) {
+  return mc->dbi_state;
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool cursor_is_gc(const MDBX_cursor *mc) {
+  return mc->dbi_state == mc->txn->dbi_state + FREE_DBI;
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool cursor_is_main(const MDBX_cursor *mc) {
+  return mc->dbi_state == mc->txn->dbi_state + MAIN_DBI;
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool cursor_is_core(const MDBX_cursor *mc) {
+  return mc->dbi_state < mc->txn->dbi_state + CORE_DBS;
+}
+
+MDBX_MAYBE_UNUSED static inline int cursor_dbi_dbg(const MDBX_cursor *mc) {
+  /* Debugging output value of a cursor's DBI: Negative for a sub-cursor. */
+  const int dbi = cursor_dbi(mc);
+  return (mc->flags & z_inner) ? -dbi : dbi;
+}
+
+MDBX_MAYBE_UNUSED static inline int __must_check_result cursor_push(MDBX_cursor *mc, page_t *mp, indx_t ki) {
+  TRACE("pushing page %" PRIaPGNO " on db %d cursor %p", mp->pgno, cursor_dbi_dbg(mc), __Wpedantic_format_voidptr(mc));
+  if (unlikely(mc->top >= CURSOR_STACK_SIZE - 1)) {
+    be_poor(mc);
+    mc->txn->flags |= MDBX_TXN_ERROR;
+    return MDBX_CURSOR_FULL;
+  }
+  mc->top += 1;
+  mc->pg[mc->top] = mp;
+  mc->ki[mc->top] = ki;
+  return MDBX_SUCCESS;
+}
+
+MDBX_MAYBE_UNUSED static inline void cursor_pop(MDBX_cursor *mc) {
+  TRACE("popped page %" PRIaPGNO " off db %d cursor %p", mc->pg[mc->top]->pgno, cursor_dbi_dbg(mc),
+        __Wpedantic_format_voidptr(mc));
+  cASSERT(mc, mc->top >= 0);
+  mc->top -= 1;
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline bool check_leaf_type(const MDBX_cursor *mc, const page_t *mp) {
+  return (((page_type(mp) ^ mc->checking) & (z_branch | z_leaf | z_largepage | z_dupfix)) == 0);
+}
+
+MDBX_INTERNAL int cursor_check(const MDBX_cursor *mc, int txn_bad_bits);
+
+/* без необходимости доступа к данным, без активации припаркованных транзакций. */
+static inline int cursor_check_pure(const MDBX_cursor *mc) {
+  return cursor_check(mc, MDBX_TXN_BLOCKED - MDBX_TXN_PARKED);
+}
+
+/* для чтения данных, с активацией припаркованных транзакций. */
+static inline int cursor_check_ro(const MDBX_cursor *mc) { return cursor_check(mc, MDBX_TXN_BLOCKED); }
+
+/* для записи данных. */
+static inline int cursor_check_rw(const MDBX_cursor *mc) {
+  return cursor_check(mc, (MDBX_TXN_BLOCKED - MDBX_TXN_PARKED) | MDBX_TXN_RDONLY);
+}
+
+MDBX_INTERNAL MDBX_cursor *cursor_eot(MDBX_cursor *cursor, MDBX_txn *txn);
+MDBX_INTERNAL int cursor_shadow(MDBX_cursor *cursor, MDBX_txn *nested, const size_t dbi);
+
+MDBX_INTERNAL MDBX_cursor *cursor_cpstk(const MDBX_cursor *csrc, MDBX_cursor *cdst);
+
+MDBX_INTERNAL int __must_check_result cursor_ops(MDBX_cursor *mc, MDBX_val *key, MDBX_val *data,
+                                                 const MDBX_cursor_op op);
+
+MDBX_INTERNAL int __must_check_result cursor_check_multiple(MDBX_cursor *mc, const MDBX_val *key, MDBX_val *data,
+                                                            unsigned flags);
+
+MDBX_INTERNAL int __must_check_result cursor_put_checklen(MDBX_cursor *mc, const MDBX_val *key, MDBX_val *data,
+                                                          unsigned flags);
+
+MDBX_INTERNAL int __must_check_result cursor_put(MDBX_cursor *mc, const MDBX_val *key, MDBX_val *data, unsigned flags);
+
+MDBX_INTERNAL int __must_check_result cursor_validate_updating(MDBX_cursor *mc);
+
+MDBX_INTERNAL int __must_check_result cursor_del(MDBX_cursor *mc, unsigned flags);
+
+MDBX_INTERNAL int __must_check_result cursor_sibling_left(MDBX_cursor *mc);
+MDBX_INTERNAL int __must_check_result cursor_sibling_right(MDBX_cursor *mc);
+
+typedef struct cursor_set_result {
+  int err;
+  bool exact;
+} csr_t;
+
+MDBX_INTERNAL csr_t cursor_seek(MDBX_cursor *mc, MDBX_val *key, MDBX_val *data, MDBX_cursor_op op);
+
+MDBX_INTERNAL int __must_check_result inner_first(MDBX_cursor *__restrict mc, MDBX_val *__restrict data);
+MDBX_INTERNAL int __must_check_result inner_last(MDBX_cursor *__restrict mc, MDBX_val *__restrict data);
+MDBX_INTERNAL int __must_check_result outer_first(MDBX_cursor *__restrict mc, MDBX_val *__restrict key,
+                                                  MDBX_val *__restrict data);
+MDBX_INTERNAL int __must_check_result outer_last(MDBX_cursor *__restrict mc, MDBX_val *__restrict key,
+                                                 MDBX_val *__restrict data);
+
+MDBX_INTERNAL int __must_check_result inner_next(MDBX_cursor *__restrict mc, MDBX_val *__restrict data);
+MDBX_INTERNAL int __must_check_result inner_prev(MDBX_cursor *__restrict mc, MDBX_val *__restrict data);
+MDBX_INTERNAL int __must_check_result outer_next(MDBX_cursor *__restrict mc, MDBX_val *__restrict key,
+                                                 MDBX_val *__restrict data, MDBX_cursor_op op);
+MDBX_INTERNAL int __must_check_result outer_prev(MDBX_cursor *__restrict mc, MDBX_val *__restrict key,
+                                                 MDBX_val *__restrict data, MDBX_cursor_op op);
+
+MDBX_INTERNAL int cursor_init4walk(cursor_couple_t *couple, const MDBX_txn *const txn, tree_t *const tree,
+                                   kvx_t *const kvx);
+
+MDBX_INTERNAL int __must_check_result cursor_init(MDBX_cursor *mc, const MDBX_txn *txn, size_t dbi);
+
+MDBX_INTERNAL int __must_check_result cursor_dupsort_setup(MDBX_cursor *mc, const node_t *node, const page_t *mp);
+
+MDBX_INTERNAL int __must_check_result cursor_touch(MDBX_cursor *const mc, const MDBX_val *key, const MDBX_val *data);
+
+/*----------------------------------------------------------------------------*/
+
+/* Update sub-page pointer, if any, in mc->subcur.
+ * Needed when the node which contains the sub-page may have moved.
+ * Called with mp = mc->pg[mc->top], ki = mc->ki[mc->top]. */
+MDBX_MAYBE_UNUSED static inline void cursor_inner_refresh(const MDBX_cursor *mc, const page_t *mp, unsigned ki) {
+  cASSERT(mc, is_leaf(mp));
+  const node_t *node = page_node(mp, ki);
+  if ((node_flags(node) & (N_DUP | N_TREE)) == N_DUP)
+    mc->subcur->cursor.pg[0] = node_data(node);
+}
+
+MDBX_MAYBE_UNUSED MDBX_INTERNAL bool cursor_is_tracked(const MDBX_cursor *mc);
+
+static inline void cursor_reset(cursor_couple_t *couple) {
+  couple->outer.top_and_flags = z_fresh_mark;
+  couple->inner.cursor.top_and_flags = z_fresh_mark | z_inner;
+}
+
+static inline void cursor_drown(cursor_couple_t *couple) {
+  couple->outer.top_and_flags = z_poor_mark;
+  couple->inner.cursor.top_and_flags = z_poor_mark | z_inner;
+  couple->outer.txn = nullptr;
+  couple->inner.cursor.txn = nullptr;
+  couple->outer.tree = nullptr;
+  /* сохраняем clc-указатель, так он используется для вычисления dbi в mdbx_cursor_renew(). */
+  couple->outer.dbi_state = nullptr;
+  couple->inner.cursor.dbi_state = nullptr;
+}

src/dbi.c

@@ -0,0 +1,692 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+#if MDBX_ENABLE_DBI_SPARSE
+size_t dbi_bitmap_ctz_fallback(const MDBX_txn *txn, intptr_t bmi) {
+  tASSERT(txn, bmi > 0);
+  bmi &= -bmi;
+  if (sizeof(txn->dbi_sparse[0]) > 4) {
+    static const uint8_t debruijn_ctz64[64] = {0,  1,  2,  53, 3,  7,  54, 27, 4,  38, 41, 8,  34, 55, 48, 28,
+                                               62, 5,  39, 46, 44, 42, 22, 9,  24, 35, 59, 56, 49, 18, 29, 11,
+                                               63, 52, 6,  26, 37, 40, 33, 47, 61, 45, 43, 21, 23, 58, 17, 10,
+                                               51, 25, 36, 32, 60, 20, 57, 16, 50, 31, 19, 15, 30, 14, 13, 12};
+    return debruijn_ctz64[(UINT64_C(0x022FDD63CC95386D) * (uint64_t)bmi) >> 58];
+  } else {
+    static const uint8_t debruijn_ctz32[32] = {0,  1,  28, 2,  29, 14, 24, 3, 30, 22, 20, 15, 25, 17, 4,  8,
+                                               31, 27, 13, 23, 21, 19, 16, 7, 26, 12, 18, 6,  11, 5,  10, 9};
+    return debruijn_ctz32[(UINT32_C(0x077CB531) * (uint32_t)bmi) >> 27];
+  }
+}
+#endif /* MDBX_ENABLE_DBI_SPARSE */
+
+struct dbi_snap_result dbi_snap(const MDBX_env *env, const size_t dbi) {
+  eASSERT(env, dbi < env->n_dbi);
+  struct dbi_snap_result r;
+  uint32_t snap = atomic_load32(&env->dbi_seqs[dbi], mo_AcquireRelease);
+  do {
+    r.sequence = snap;
+    r.flags = env->dbs_flags[dbi];
+    snap = atomic_load32(&env->dbi_seqs[dbi], mo_AcquireRelease);
+  } while (unlikely(snap != r.sequence));
+  return r;
+}
+
+__noinline int dbi_import(MDBX_txn *txn, const size_t dbi) {
+  const MDBX_env *const env = txn->env;
+  if (dbi >= env->n_dbi || !env->dbs_flags[dbi])
+    return MDBX_BAD_DBI;
+
+#if MDBX_ENABLE_DBI_SPARSE
+  const size_t bitmap_chunk = CHAR_BIT * sizeof(txn->dbi_sparse[0]);
+  const size_t bitmap_indx = dbi / bitmap_chunk;
+  const size_t bitmap_mask = (size_t)1 << dbi % bitmap_chunk;
+  if (dbi >= txn->n_dbi) {
+    for (size_t i = (txn->n_dbi + bitmap_chunk - 1) / bitmap_chunk; bitmap_indx >= i; ++i)
+      txn->dbi_sparse[i] = 0;
+    eASSERT(env, (txn->dbi_sparse[bitmap_indx] & bitmap_mask) == 0);
+    MDBX_txn *scan = txn;
+    do {
+      eASSERT(env, scan->dbi_sparse == txn->dbi_sparse);
+      eASSERT(env, scan->n_dbi < dbi + 1);
+      scan->n_dbi = (unsigned)dbi + 1;
+      scan->dbi_state[dbi] = 0;
+      scan = scan->parent;
+    } while (scan /* && scan->dbi_sparse == txn->dbi_sparse */);
+    txn->dbi_sparse[bitmap_indx] |= bitmap_mask;
+    goto lindo;
+  }
+  if ((txn->dbi_sparse[bitmap_indx] & bitmap_mask) == 0) {
+    MDBX_txn *scan = txn;
+    do {
+      eASSERT(env, scan->dbi_sparse == txn->dbi_sparse);
+      eASSERT(env, scan->n_dbi == txn->n_dbi);
+      scan->dbi_state[dbi] = 0;
+      scan = scan->parent;
+    } while (scan /* && scan->dbi_sparse == txn->dbi_sparse */);
+    txn->dbi_sparse[bitmap_indx] |= bitmap_mask;
+    goto lindo;
+  }
+#else
+  if (dbi >= txn->n_dbi) {
+    size_t i = txn->n_dbi;
+    do
+      txn->dbi_state[i] = 0;
+    while (dbi >= ++i);
+    txn->n_dbi = i;
+    goto lindo;
+  }
+#endif /* MDBX_ENABLE_DBI_SPARSE */
+
+  if (!txn->dbi_state[dbi]) {
+  lindo:
+    /* dbi-слот еще не инициализирован в транзакции, а хендл не использовался */
+    txn->cursors[dbi] = nullptr;
+    MDBX_txn *const parent = txn->parent;
+    if (parent) {
+      /* вложенная пишущая транзакция */
+      int rc = dbi_check(parent, dbi);
+      /* копируем состояние dbi-хендла очищая new-флаги. */
+      eASSERT(env, txn->dbi_seqs == parent->dbi_seqs);
+      txn->dbi_state[dbi] = parent->dbi_state[dbi] & ~(DBI_FRESH | DBI_CREAT | DBI_DIRTY);
+      if (likely(rc == MDBX_SUCCESS)) {
+        txn->dbs[dbi] = parent->dbs[dbi];
+        rc = txn_shadow_cursors(parent, dbi);
+      }
+      return rc;
+    }
+    txn->dbi_seqs[dbi] = 0;
+    txn->dbi_state[dbi] = DBI_LINDO;
+  } else {
+    eASSERT(env, txn->dbi_seqs[dbi] != env->dbi_seqs[dbi].weak);
+    if (unlikely((txn->dbi_state[dbi] & (DBI_VALID | DBI_OLDEN)) || txn->cursors[dbi])) {
+      /* хендл уже использовался в транзакции, но был закрыт или переоткрыт,
+       * либо при явном пере-открытии хендла есть висячие курсоры */
+      eASSERT(env, (txn->dbi_state[dbi] & DBI_STALE) == 0);
+      txn->dbi_seqs[dbi] = env->dbi_seqs[dbi].weak;
+      txn->dbi_state[dbi] = DBI_OLDEN | DBI_LINDO;
+      return txn->cursors[dbi] ? MDBX_DANGLING_DBI : MDBX_BAD_DBI;
+    }
+  }
+
+  /* хендл не использовался в транзакции, либо явно пере-отрывается при
+   * отсутствии висячих курсоров */
+  eASSERT(env, (txn->dbi_state[dbi] & DBI_LINDO) && !txn->cursors[dbi]);
+
+  /* читаем актуальные флаги и sequence */
+  struct dbi_snap_result snap = dbi_snap(env, dbi);
+  txn->dbi_seqs[dbi] = snap.sequence;
+  if (snap.flags & DB_VALID) {
+    txn->dbs[dbi].flags = snap.flags & DB_PERSISTENT_FLAGS;
+    txn->dbi_state[dbi] = DBI_LINDO | DBI_VALID | DBI_STALE;
+    return MDBX_SUCCESS;
+  }
+  return MDBX_BAD_DBI;
+}
+
+int dbi_defer_release(MDBX_env *const env, defer_free_item_t *const chain) {
+  size_t length = 0;
+  defer_free_item_t *obsolete_chain = nullptr;
+#if MDBX_ENABLE_DBI_LOCKFREE
+  const uint64_t now = osal_monotime();
+  defer_free_item_t **scan = &env->defer_free;
+  if (env->defer_free) {
+    const uint64_t threshold_1second = osal_16dot16_to_monotime(1 * 65536);
+    do {
+      defer_free_item_t *item = *scan;
+      if (now - item->timestamp < threshold_1second) {
+        scan = &item->next;
+        length += 1;
+      } else {
+        *scan = item->next;
+        item->next = obsolete_chain;
+        obsolete_chain = item;
+      }
+    } while (*scan);
+  }
+
+  eASSERT(env, *scan == nullptr);
+  if (chain) {
+    defer_free_item_t *item = chain;
+    do {
+      item->timestamp = now;
+      item = item->next;
+    } while (item);
+    *scan = chain;
+  }
+#else  /* MDBX_ENABLE_DBI_LOCKFREE */
+  obsolete_chain = chain;
+#endif /* MDBX_ENABLE_DBI_LOCKFREE */
+
+  ENSURE(env, osal_fastmutex_release(&env->dbi_lock) == MDBX_SUCCESS);
+  if (length > 42) {
+#if defined(_WIN32) || defined(_WIN64)
+    SwitchToThread();
+#else
+    sched_yield();
+#endif /* Windows */
+  }
+  while (obsolete_chain) {
+    defer_free_item_t *item = obsolete_chain;
+    obsolete_chain = obsolete_chain->next;
+    osal_free(item);
+  }
+  return chain ? MDBX_SUCCESS : MDBX_BAD_DBI;
+}
+
+/* Export or close DBI handles opened in this txn. */
+int dbi_update(MDBX_txn *txn, bool keep) {
+  MDBX_env *const env = txn->env;
+  tASSERT(txn, !txn->parent && txn == env->basal_txn);
+  bool locked = false;
+  defer_free_item_t *defer_chain = nullptr;
+  TXN_FOREACH_DBI_USER(txn, dbi) {
+    if (likely((txn->dbi_state[dbi] & DBI_CREAT) == 0))
+      continue;
+    if (!locked) {
+      int err = osal_fastmutex_acquire(&env->dbi_lock);
+      if (unlikely(err != MDBX_SUCCESS))
+        return err;
+      locked = true;
+      if (dbi >= env->n_dbi)
+        /* хендл был закрыт из другого потока пока захватывали блокировку */
+        continue;
+    }
+    tASSERT(txn, dbi < env->n_dbi);
+    if (keep) {
+      env->dbs_flags[dbi] = txn->dbs[dbi].flags | DB_VALID;
+    } else {
+      uint32_t seq = dbi_seq_next(env, dbi);
+      defer_free_item_t *item = env->kvs[dbi].name.iov_base;
+      if (item) {
+        env->dbs_flags[dbi] = 0;
+        env->kvs[dbi].name.iov_len = 0;
+        env->kvs[dbi].name.iov_base = nullptr;
+        atomic_store32(&env->dbi_seqs[dbi], seq, mo_AcquireRelease);
+        osal_flush_incoherent_cpu_writeback();
+        item->next = defer_chain;
+        defer_chain = item;
+      } else {
+        eASSERT(env, env->kvs[dbi].name.iov_len == 0);
+        eASSERT(env, env->dbs_flags[dbi] == 0);
+      }
+    }
+  }
+
+  if (locked) {
+    size_t i = env->n_dbi;
+    eASSERT(env, env->n_dbi >= CORE_DBS);
+    while ((env->dbs_flags[i - 1] & DB_VALID) == 0) {
+      --i;
+      eASSERT(env, i >= CORE_DBS);
+      eASSERT(env, !env->dbs_flags[i] && !env->kvs[i].name.iov_len && !env->kvs[i].name.iov_base);
+    }
+    env->n_dbi = (unsigned)i;
+    dbi_defer_release(env, defer_chain);
+  }
+  return MDBX_SUCCESS;
+}
+
+int dbi_bind(MDBX_txn *txn, const size_t dbi, unsigned user_flags, MDBX_cmp_func *keycmp, MDBX_cmp_func *datacmp) {
+  const MDBX_env *const env = txn->env;
+  eASSERT(env, dbi < txn->n_dbi && dbi < env->n_dbi);
+  eASSERT(env, dbi_state(txn, dbi) & DBI_LINDO);
+  eASSERT(env, env->dbs_flags[dbi] != DB_POISON);
+  if ((env->dbs_flags[dbi] & DB_VALID) == 0) {
+    eASSERT(env, !env->kvs[dbi].clc.k.cmp && !env->kvs[dbi].clc.v.cmp && !env->kvs[dbi].name.iov_len &&
+                     !env->kvs[dbi].name.iov_base && !env->kvs[dbi].clc.k.lmax && !env->kvs[dbi].clc.k.lmin &&
+                     !env->kvs[dbi].clc.v.lmax && !env->kvs[dbi].clc.v.lmin);
+  } else {
+    eASSERT(env, !(txn->dbi_state[dbi] & DBI_VALID) || (txn->dbs[dbi].flags | DB_VALID) == env->dbs_flags[dbi]);
+    eASSERT(env, env->kvs[dbi].name.iov_base || dbi < CORE_DBS);
+  }
+
+  /* Если dbi уже использовался, то корректными считаем четыре варианта:
+   * 1) user_flags равны MDBX_DB_ACCEDE
+   *   = предполагаем что пользователь открывает существующую table,
+   *     при этом код проверки не позволит установить другие компараторы.
+   * 2) user_flags нулевые, а оба компаратора пустые/нулевые или равны текущим
+   *   = предполагаем что пользователь открывает существующую table
+   *     старым способом с нулевыми с флагами по-умолчанию.
+   * 3) user_flags совпадают, а компараторы не заданы или те же
+   *    = предполагаем что пользователь открывает table указывая все параметры;
+   * 4) user_flags отличаются, но table пустая и задан флаг MDBX_CREATE
+   *    = предполагаем что пользователь пересоздает table;
+   */
+  if ((user_flags & ~MDBX_CREATE) != (unsigned)(env->dbs_flags[dbi] & DB_PERSISTENT_FLAGS)) {
+    /* flags are differs, check other conditions */
+    if ((!user_flags && (!keycmp || keycmp == env->kvs[dbi].clc.k.cmp) &&
+         (!datacmp || datacmp == env->kvs[dbi].clc.v.cmp)) ||
+        user_flags == MDBX_DB_ACCEDE) {
+      user_flags = env->dbs_flags[dbi] & DB_PERSISTENT_FLAGS;
+    } else if ((user_flags & MDBX_CREATE) == 0)
+      return /* FIXME: return extended info */ MDBX_INCOMPATIBLE;
+    else {
+      if (txn->dbi_state[dbi] & DBI_STALE) {
+        eASSERT(env, env->dbs_flags[dbi] & DB_VALID);
+        int err = tbl_fetch(txn, dbi);
+        if (unlikely(err == MDBX_SUCCESS))
+          return err;
+      }
+      eASSERT(env, ((env->dbs_flags[dbi] ^ txn->dbs[dbi].flags) & DB_PERSISTENT_FLAGS) == 0);
+      eASSERT(env, (txn->dbi_state[dbi] & (DBI_LINDO | DBI_VALID | DBI_STALE)) == (DBI_LINDO | DBI_VALID));
+      if (unlikely(txn->dbs[dbi].leaf_pages))
+        return /* FIXME: return extended info */ MDBX_INCOMPATIBLE;
+
+      /* Пересоздаём table если там пусто */
+      if (unlikely(txn->cursors[dbi]))
+        return MDBX_DANGLING_DBI;
+      env->dbs_flags[dbi] = DB_POISON;
+      atomic_store32(&env->dbi_seqs[dbi], dbi_seq_next(env, dbi), mo_AcquireRelease);
+
+      const uint32_t seq = dbi_seq_next(env, dbi);
+      const uint16_t db_flags = user_flags & DB_PERSISTENT_FLAGS;
+      eASSERT(env, txn->dbs[dbi].height == 0 && txn->dbs[dbi].items == 0 && txn->dbs[dbi].root == P_INVALID);
+      env->kvs[dbi].clc.k.cmp = keycmp ? keycmp : builtin_keycmp(user_flags);
+      env->kvs[dbi].clc.v.cmp = datacmp ? datacmp : builtin_datacmp(user_flags);
+      txn->dbs[dbi].flags = db_flags;
+      txn->dbs[dbi].dupfix_size = 0;
+      if (unlikely(tbl_setup(env, &env->kvs[dbi], &txn->dbs[dbi]))) {
+        txn->dbi_state[dbi] = DBI_LINDO;
+        txn->flags |= MDBX_TXN_ERROR;
+        return MDBX_PROBLEM;
+      }
+
+      env->dbs_flags[dbi] = db_flags | DB_VALID;
+      atomic_store32(&env->dbi_seqs[dbi], seq, mo_AcquireRelease);
+      txn->dbi_seqs[dbi] = seq;
+      txn->dbi_state[dbi] = DBI_LINDO | DBI_VALID | DBI_CREAT | DBI_DIRTY;
+      txn->flags |= MDBX_TXN_DIRTY;
+    }
+  }
+
+  if (!keycmp)
+    keycmp = (env->dbs_flags[dbi] & DB_VALID) ? env->kvs[dbi].clc.k.cmp : builtin_keycmp(user_flags);
+  if (env->kvs[dbi].clc.k.cmp != keycmp) {
+    if (env->dbs_flags[dbi] & DB_VALID)
+      return MDBX_EINVAL;
+    env->kvs[dbi].clc.k.cmp = keycmp;
+  }
+
+  if (!datacmp)
+    datacmp = (env->dbs_flags[dbi] & DB_VALID) ? env->kvs[dbi].clc.v.cmp : builtin_datacmp(user_flags);
+  if (env->kvs[dbi].clc.v.cmp != datacmp) {
+    if (env->dbs_flags[dbi] & DB_VALID)
+      return MDBX_EINVAL;
+    env->kvs[dbi].clc.v.cmp = datacmp;
+  }
+
+  return MDBX_SUCCESS;
+}
+
+static inline size_t dbi_namelen(const MDBX_val name) {
+  return (name.iov_len > sizeof(defer_free_item_t)) ? name.iov_len : sizeof(defer_free_item_t);
+}
+
+static int dbi_open_locked(MDBX_txn *txn, unsigned user_flags, MDBX_dbi *dbi, MDBX_cmp_func *keycmp,
+                           MDBX_cmp_func *datacmp, MDBX_val name) {
+  MDBX_env *const env = txn->env;
+
+  /* Cannot mix named table(s) with DUPSORT flags */
+  tASSERT(txn, (txn->dbi_state[MAIN_DBI] & (DBI_LINDO | DBI_VALID | DBI_STALE)) == (DBI_LINDO | DBI_VALID));
+  if (unlikely(txn->dbs[MAIN_DBI].flags & MDBX_DUPSORT)) {
+    if (unlikely((user_flags & MDBX_CREATE) == 0))
+      return MDBX_NOTFOUND;
+    if (unlikely(txn->dbs[MAIN_DBI].leaf_pages))
+      /* В MainDB есть записи, либо она уже использовалась. */
+      return MDBX_INCOMPATIBLE;
+
+    /* Пересоздаём MainDB когда там пусто. */
+    tASSERT(txn,
+            txn->dbs[MAIN_DBI].height == 0 && txn->dbs[MAIN_DBI].items == 0 && txn->dbs[MAIN_DBI].root == P_INVALID);
+    if (unlikely(txn->cursors[MAIN_DBI]))
+      return MDBX_DANGLING_DBI;
+    env->dbs_flags[MAIN_DBI] = DB_POISON;
+    atomic_store32(&env->dbi_seqs[MAIN_DBI], dbi_seq_next(env, MAIN_DBI), mo_AcquireRelease);
+
+    const uint32_t seq = dbi_seq_next(env, MAIN_DBI);
+    const uint16_t main_flags = txn->dbs[MAIN_DBI].flags & (MDBX_REVERSEKEY | MDBX_INTEGERKEY);
+    env->kvs[MAIN_DBI].clc.k.cmp = builtin_keycmp(main_flags);
+    env->kvs[MAIN_DBI].clc.v.cmp = builtin_datacmp(main_flags);
+    txn->dbs[MAIN_DBI].flags = main_flags;
+    txn->dbs[MAIN_DBI].dupfix_size = 0;
+    int err = tbl_setup(env, &env->kvs[MAIN_DBI], &txn->dbs[MAIN_DBI]);
+    if (unlikely(err != MDBX_SUCCESS)) {
+      txn->dbi_state[MAIN_DBI] = DBI_LINDO;
+      txn->flags |= MDBX_TXN_ERROR;
+      env->flags |= ENV_FATAL_ERROR;
+      return err;
+    }
+    env->dbs_flags[MAIN_DBI] = main_flags | DB_VALID;
+    txn->dbi_seqs[MAIN_DBI] = atomic_store32(&env->dbi_seqs[MAIN_DBI], seq, mo_AcquireRelease);
+    txn->dbi_state[MAIN_DBI] |= DBI_DIRTY;
+    txn->flags |= MDBX_TXN_DIRTY;
+  }
+
+  tASSERT(txn, env->kvs[MAIN_DBI].clc.k.cmp);
+
+  /* Is the DB already open? */
+  size_t slot = env->n_dbi;
+  for (size_t scan = CORE_DBS; scan < env->n_dbi; ++scan) {
+    if ((env->dbs_flags[scan] & DB_VALID) == 0) {
+      /* Remember this free slot */
+      slot = (slot < scan) ? slot : scan;
+      continue;
+    }
+    if (!env->kvs[MAIN_DBI].clc.k.cmp(&name, &env->kvs[scan].name)) {
+      slot = scan;
+      int err = dbi_check(txn, slot);
+      if (err == MDBX_BAD_DBI && txn->dbi_state[slot] == (DBI_OLDEN | DBI_LINDO)) {
+        /* хендл использовался, стал невалидным,
+         * но теперь явно пере-открывается в этой транзакци */
+        eASSERT(env, !txn->cursors[slot]);
+        txn->dbi_state[slot] = DBI_LINDO;
+        err = dbi_check(txn, slot);
+      }
+      if (err == MDBX_SUCCESS) {
+        err = dbi_bind(txn, slot, user_flags, keycmp, datacmp);
+        if (likely(err == MDBX_SUCCESS)) {
+          goto done;
+        }
+      }
+      return err;
+    }
+  }
+
+  /* Fail, if no free slot and max hit */
+  if (unlikely(slot >= env->max_dbi))
+    return MDBX_DBS_FULL;
+
+  if (env->n_dbi == slot)
+    eASSERT(env, !env->dbs_flags[slot] && !env->kvs[slot].name.iov_len && !env->kvs[slot].name.iov_base);
+
+  env->dbs_flags[slot] = DB_POISON;
+  atomic_store32(&env->dbi_seqs[slot], dbi_seq_next(env, slot), mo_AcquireRelease);
+  memset(&env->kvs[slot], 0, sizeof(env->kvs[slot]));
+  if (env->n_dbi == slot)
+    env->n_dbi = (unsigned)slot + 1;
+  eASSERT(env, slot < env->n_dbi);
+
+  int err = dbi_check(txn, slot);
+  eASSERT(env, err == MDBX_BAD_DBI);
+  if (err != MDBX_BAD_DBI)
+    return MDBX_PROBLEM;
+
+  /* Find the DB info */
+  MDBX_val body;
+  cursor_couple_t cx;
+  int rc = cursor_init(&cx.outer, txn, MAIN_DBI);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return rc;
+  rc = cursor_seek(&cx.outer, &name, &body, MDBX_SET).err;
+  if (unlikely(rc != MDBX_SUCCESS)) {
+    if (rc != MDBX_NOTFOUND || !(user_flags & MDBX_CREATE))
+      return rc;
+  } else {
+    /* make sure this is actually a table */
+    node_t *node = page_node(cx.outer.pg[cx.outer.top], cx.outer.ki[cx.outer.top]);
+    if (unlikely((node_flags(node) & (N_DUP | N_TREE)) != N_TREE))
+      return MDBX_INCOMPATIBLE;
+    if (!MDBX_DISABLE_VALIDATION && unlikely(body.iov_len != sizeof(tree_t))) {
+      ERROR("%s/%d: %s %zu", "MDBX_CORRUPTED", MDBX_CORRUPTED, "invalid table node size", body.iov_len);
+      return MDBX_CORRUPTED;
+    }
+    memcpy(&txn->dbs[slot], body.iov_base, sizeof(tree_t));
+  }
+
+  /* Done here so we cannot fail after creating a new DB */
+  defer_free_item_t *const clone = osal_malloc(dbi_namelen(name));
+  if (unlikely(!clone))
+    return MDBX_ENOMEM;
+  memcpy(clone, name.iov_base, name.iov_len);
+  name.iov_base = clone;
+
+  uint8_t dbi_state = DBI_LINDO | DBI_VALID | DBI_FRESH;
+  if (unlikely(rc)) {
+    /* MDBX_NOTFOUND and MDBX_CREATE: Create new DB */
+    tASSERT(txn, rc == MDBX_NOTFOUND);
+    body.iov_base = memset(&txn->dbs[slot], 0, body.iov_len = sizeof(tree_t));
+    txn->dbs[slot].root = P_INVALID;
+    txn->dbs[slot].mod_txnid = txn->txnid;
+    txn->dbs[slot].flags = user_flags & DB_PERSISTENT_FLAGS;
+    cx.outer.next = txn->cursors[MAIN_DBI];
+    txn->cursors[MAIN_DBI] = &cx.outer;
+    rc = cursor_put_checklen(&cx.outer, &name, &body, N_TREE | MDBX_NOOVERWRITE);
+    txn->cursors[MAIN_DBI] = cx.outer.next;
+    if (unlikely(rc != MDBX_SUCCESS))
+      goto bailout;
+
+    dbi_state |= DBI_DIRTY | DBI_CREAT;
+    txn->flags |= MDBX_TXN_DIRTY;
+    tASSERT(txn, (txn->dbi_state[MAIN_DBI] & DBI_DIRTY) != 0);
+  }
+
+  /* Got info, register DBI in this txn */
+  const uint32_t seq = dbi_seq_next(env, slot);
+  eASSERT(env, env->dbs_flags[slot] == DB_POISON && !txn->cursors[slot] &&
+                   (txn->dbi_state[slot] & (DBI_LINDO | DBI_VALID)) == DBI_LINDO);
+  txn->dbi_state[slot] = dbi_state;
+  memcpy(&txn->dbs[slot], body.iov_base, sizeof(txn->dbs[slot]));
+  env->dbs_flags[slot] = txn->dbs[slot].flags;
+  rc = dbi_bind(txn, slot, user_flags, keycmp, datacmp);
+  if (unlikely(rc != MDBX_SUCCESS))
+    goto bailout;
+
+  env->kvs[slot].name = name;
+  env->dbs_flags[slot] = txn->dbs[slot].flags | DB_VALID;
+  txn->dbi_seqs[slot] = atomic_store32(&env->dbi_seqs[slot], seq, mo_AcquireRelease);
+
+done:
+  *dbi = (MDBX_dbi)slot;
+  tASSERT(txn, slot < txn->n_dbi && (env->dbs_flags[slot] & DB_VALID) != 0);
+  eASSERT(env, dbi_check(txn, slot) == MDBX_SUCCESS);
+  return MDBX_SUCCESS;
+
+bailout:
+  eASSERT(env, !txn->cursors[slot] && !env->kvs[slot].name.iov_len && !env->kvs[slot].name.iov_base);
+  txn->dbi_state[slot] &= DBI_LINDO | DBI_OLDEN;
+  env->dbs_flags[slot] = 0;
+  osal_free(clone);
+  if (slot + 1 == env->n_dbi)
+    txn->n_dbi = env->n_dbi = (unsigned)slot;
+  return rc;
+}
+
+int dbi_open(MDBX_txn *txn, const MDBX_val *const name, unsigned user_flags, MDBX_dbi *dbi, MDBX_cmp_func *keycmp,
+             MDBX_cmp_func *datacmp) {
+  if (unlikely(!dbi))
+    return MDBX_EINVAL;
+  *dbi = 0;
+
+  if (user_flags != MDBX_ACCEDE && unlikely(!check_table_flags(user_flags & ~MDBX_CREATE)))
+    return MDBX_EINVAL;
+
+  int rc = check_txn(txn, MDBX_TXN_BLOCKED);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return rc;
+
+  if ((user_flags & MDBX_CREATE) && unlikely(txn->flags & MDBX_TXN_RDONLY))
+    return MDBX_EACCESS;
+
+  /* main table? */
+  if (unlikely(name == MDBX_CHK_MAIN || name->iov_base == MDBX_CHK_MAIN)) {
+    rc = dbi_bind(txn, MAIN_DBI, user_flags, keycmp, datacmp);
+    if (likely(rc == MDBX_SUCCESS))
+      *dbi = MAIN_DBI;
+    return rc;
+  }
+  if (unlikely(name == MDBX_CHK_GC || name->iov_base == MDBX_CHK_GC)) {
+    rc = dbi_bind(txn, FREE_DBI, user_flags, keycmp, datacmp);
+    if (likely(rc == MDBX_SUCCESS))
+      *dbi = FREE_DBI;
+    return rc;
+  }
+  if (unlikely(name == MDBX_CHK_META || name->iov_base == MDBX_CHK_META))
+    return MDBX_EINVAL;
+  if (unlikely(name->iov_len > txn->env->leaf_nodemax - NODESIZE - sizeof(tree_t)))
+    return MDBX_EINVAL;
+
+#if MDBX_ENABLE_DBI_LOCKFREE
+  /* Is the DB already open? */
+  const MDBX_env *const env = txn->env;
+  size_t free_slot = env->n_dbi;
+  for (size_t i = CORE_DBS; i < env->n_dbi; ++i) {
+  retry:
+    if ((env->dbs_flags[i] & DB_VALID) == 0) {
+      free_slot = i;
+      continue;
+    }
+
+    const uint32_t snap_seq = atomic_load32(&env->dbi_seqs[i], mo_AcquireRelease);
+    const uint16_t snap_flags = env->dbs_flags[i];
+    const MDBX_val snap_name = env->kvs[i].name;
+    if (user_flags != MDBX_ACCEDE &&
+        (((user_flags ^ snap_flags) & DB_PERSISTENT_FLAGS) || (keycmp && keycmp != env->kvs[i].clc.k.cmp) ||
+         (datacmp && datacmp != env->kvs[i].clc.v.cmp)))
+      continue;
+    const uint32_t main_seq = atomic_load32(&env->dbi_seqs[MAIN_DBI], mo_AcquireRelease);
+    MDBX_cmp_func *const snap_cmp = env->kvs[MAIN_DBI].clc.k.cmp;
+    if (unlikely(!(snap_flags & DB_VALID) || !snap_name.iov_base || !snap_name.iov_len || !snap_cmp))
+      continue;
+
+    const bool name_match = snap_cmp(&snap_name, name) == 0;
+    osal_flush_incoherent_cpu_writeback();
+    if (unlikely(snap_seq != atomic_load32(&env->dbi_seqs[i], mo_AcquireRelease) ||
+                 main_seq != atomic_load32(&env->dbi_seqs[MAIN_DBI], mo_AcquireRelease) ||
+                 snap_flags != env->dbs_flags[i] || snap_name.iov_base != env->kvs[i].name.iov_base ||
+                 snap_name.iov_len != env->kvs[i].name.iov_len))
+      goto retry;
+    if (name_match) {
+      rc = dbi_check(txn, i);
+      if (rc == MDBX_BAD_DBI && txn->dbi_state[i] == (DBI_OLDEN | DBI_LINDO)) {
+        /* хендл использовался, стал невалидным,
+         * но теперь явно пере-открывается в этой транзакци */
+        eASSERT(env, !txn->cursors[i]);
+        txn->dbi_state[i] = DBI_LINDO;
+        rc = dbi_check(txn, i);
+      }
+      if (likely(rc == MDBX_SUCCESS)) {
+        rc = dbi_bind(txn, i, user_flags, keycmp, datacmp);
+        if (likely(rc == MDBX_SUCCESS))
+          *dbi = (MDBX_dbi)i;
+      }
+      return rc;
+    }
+  }
+
+  /* Fail, if no free slot and max hit */
+  if (unlikely(free_slot >= env->max_dbi))
+    return MDBX_DBS_FULL;
+#endif /* MDBX_ENABLE_DBI_LOCKFREE */
+
+  rc = osal_fastmutex_acquire(&txn->env->dbi_lock);
+  if (likely(rc == MDBX_SUCCESS)) {
+    rc = dbi_open_locked(txn, user_flags, dbi, keycmp, datacmp, *name);
+    ENSURE(txn->env, osal_fastmutex_release(&txn->env->dbi_lock) == MDBX_SUCCESS);
+  }
+  return rc;
+}
+
+__cold struct dbi_rename_result dbi_rename_locked(MDBX_txn *txn, MDBX_dbi dbi, MDBX_val new_name) {
+  struct dbi_rename_result pair;
+  pair.defer = nullptr;
+  pair.err = dbi_check(txn, dbi);
+  if (unlikely(pair.err != MDBX_SUCCESS))
+    return pair;
+
+  MDBX_env *const env = txn->env;
+  MDBX_val old_name = env->kvs[dbi].name;
+  if (env->kvs[MAIN_DBI].clc.k.cmp(&new_name, &old_name) == 0 && MDBX_DEBUG == 0)
+    return pair;
+
+  cursor_couple_t cx;
+  pair.err = cursor_init(&cx.outer, txn, MAIN_DBI);
+  if (unlikely(pair.err != MDBX_SUCCESS))
+    return pair;
+  pair.err = cursor_seek(&cx.outer, &new_name, nullptr, MDBX_SET).err;
+  if (unlikely(pair.err != MDBX_NOTFOUND)) {
+    pair.err = (pair.err == MDBX_SUCCESS) ? MDBX_KEYEXIST : pair.err;
+    return pair;
+  }
+
+  pair.defer = osal_malloc(dbi_namelen(new_name));
+  if (unlikely(!pair.defer)) {
+    pair.err = MDBX_ENOMEM;
+    return pair;
+  }
+  new_name.iov_base = memcpy(pair.defer, new_name.iov_base, new_name.iov_len);
+
+  cx.outer.next = txn->cursors[MAIN_DBI];
+  txn->cursors[MAIN_DBI] = &cx.outer;
+
+  MDBX_val data = {&txn->dbs[dbi], sizeof(tree_t)};
+  pair.err = cursor_put_checklen(&cx.outer, &new_name, &data, N_TREE | MDBX_NOOVERWRITE);
+  if (likely(pair.err == MDBX_SUCCESS)) {
+    pair.err = cursor_seek(&cx.outer, &old_name, nullptr, MDBX_SET).err;
+    if (likely(pair.err == MDBX_SUCCESS))
+      pair.err = cursor_del(&cx.outer, N_TREE);
+    if (likely(pair.err == MDBX_SUCCESS)) {
+      pair.defer = env->kvs[dbi].name.iov_base;
+      env->kvs[dbi].name = new_name;
+    } else
+      txn->flags |= MDBX_TXN_ERROR;
+  }
+
+  txn->cursors[MAIN_DBI] = cx.outer.next;
+  return pair;
+}
+
+static defer_free_item_t *dbi_close_locked(MDBX_env *env, MDBX_dbi dbi) {
+  eASSERT(env, dbi >= CORE_DBS);
+  if (unlikely(dbi >= env->n_dbi))
+    return nullptr;
+
+  const uint32_t seq = dbi_seq_next(env, dbi);
+  defer_free_item_t *defer_item = env->kvs[dbi].name.iov_base;
+  if (likely(defer_item)) {
+    env->dbs_flags[dbi] = 0;
+    env->kvs[dbi].name.iov_len = 0;
+    env->kvs[dbi].name.iov_base = nullptr;
+    atomic_store32(&env->dbi_seqs[dbi], seq, mo_AcquireRelease);
+    osal_flush_incoherent_cpu_writeback();
+    defer_item->next = nullptr;
+
+    if (env->n_dbi == dbi + 1) {
+      size_t i = env->n_dbi;
+      do {
+        --i;
+        eASSERT(env, i >= CORE_DBS);
+        eASSERT(env, !env->dbs_flags[i] && !env->kvs[i].name.iov_len && !env->kvs[i].name.iov_base);
+      } while (i > CORE_DBS && !env->kvs[i - 1].name.iov_base);
+      env->n_dbi = (unsigned)i;
+    }
+  }
+
+  return defer_item;
+}
+
+__cold const tree_t *dbi_dig(const MDBX_txn *txn, const size_t dbi, tree_t *fallback) {
+  const MDBX_txn *dig = txn;
+  do {
+    tASSERT(txn, txn->n_dbi == dig->n_dbi);
+    const uint8_t state = dbi_state(dig, dbi);
+    if (state & DBI_LINDO)
+      switch (state & (DBI_VALID | DBI_STALE | DBI_OLDEN)) {
+      case DBI_VALID:
+      case DBI_OLDEN:
+        return dig->dbs + dbi;
+      case 0:
+        return fallback;
+      case DBI_VALID | DBI_STALE:
+      case DBI_OLDEN | DBI_STALE:
+        break;
+      default:
+        tASSERT(txn, !!"unexpected dig->dbi_state[dbi]");
+      }
+    dig = dig->parent;
+  } while (dig);
+  return fallback;
+}
+
+int dbi_close_release(MDBX_env *env, MDBX_dbi dbi) { return dbi_defer_release(env, dbi_close_locked(env, dbi)); }

src/dbi.h

@@ -0,0 +1,147 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#pragma once
+
+#include "essentials.h"
+
+#if MDBX_ENABLE_DBI_SPARSE
+
+MDBX_NOTHROW_CONST_FUNCTION MDBX_MAYBE_UNUSED MDBX_INTERNAL size_t dbi_bitmap_ctz_fallback(const MDBX_txn *txn,
+                                                                                           intptr_t bmi);
+
+static inline size_t dbi_bitmap_ctz(const MDBX_txn *txn, intptr_t bmi) {
+  tASSERT(txn, bmi > 0);
+  STATIC_ASSERT(sizeof(bmi) >= sizeof(txn->dbi_sparse[0]));
+#if __GNUC_PREREQ(4, 1) || __has_builtin(__builtin_ctzl)
+  if (sizeof(txn->dbi_sparse[0]) <= sizeof(int))
+    return __builtin_ctz((int)bmi);
+  if (sizeof(txn->dbi_sparse[0]) == sizeof(long))
+    return __builtin_ctzl((long)bmi);
+#if (defined(__SIZEOF_LONG_LONG__) && __SIZEOF_LONG_LONG__ == 8) || __has_builtin(__builtin_ctzll)
+  return __builtin_ctzll(bmi);
+#endif /* have(long long) && long long == uint64_t */
+#endif /* GNU C */
+
+#if defined(_MSC_VER)
+  unsigned long index;
+  if (sizeof(txn->dbi_sparse[0]) > 4) {
+#if defined(_M_AMD64) || defined(_M_ARM64) || defined(_M_X64)
+    _BitScanForward64(&index, bmi);
+    return index;
+#else
+    if (bmi > UINT32_MAX) {
+      _BitScanForward(&index, (uint32_t)((uint64_t)bmi >> 32));
+      return index;
+    }
+#endif
+  }
+  _BitScanForward(&index, (uint32_t)bmi);
+  return index;
+#endif /* MSVC */
+
+  return dbi_bitmap_ctz_fallback(txn, bmi);
+}
+
+static inline bool dbi_foreach_step(const MDBX_txn *const txn, size_t *bitmap_item, size_t *dbi) {
+  const size_t bitmap_chunk = CHAR_BIT * sizeof(txn->dbi_sparse[0]);
+  if (*bitmap_item & 1) {
+    *bitmap_item >>= 1;
+    return txn->dbi_state[*dbi] != 0;
+  }
+  if (*bitmap_item) {
+    size_t bitmap_skip = dbi_bitmap_ctz(txn, *bitmap_item);
+    *bitmap_item >>= bitmap_skip;
+    *dbi += bitmap_skip - 1;
+  } else {
+    *dbi = (*dbi - 1) | (bitmap_chunk - 1);
+    *bitmap_item = txn->dbi_sparse[(1 + *dbi) / bitmap_chunk];
+    if (*bitmap_item == 0)
+      *dbi += bitmap_chunk;
+  }
+  return false;
+}
+
+/* LY: Макрос целенаправленно сделан с одним циклом, чтобы сохранить возможность
+ * использования оператора break */
+#define TXN_FOREACH_DBI_FROM(TXN, I, FROM)                                                                             \
+  for (size_t bitmap_item = TXN->dbi_sparse[0] >> FROM, I = FROM; I < TXN->n_dbi; ++I)                                 \
+    if (dbi_foreach_step(TXN, &bitmap_item, &I))
+
+#else
+
+#define TXN_FOREACH_DBI_FROM(TXN, I, FROM)                                                                             \
+  for (size_t I = FROM; I < TXN->n_dbi; ++I)                                                                           \
+    if (TXN->dbi_state[I])
+
+#endif /* MDBX_ENABLE_DBI_SPARSE */
+
+#define TXN_FOREACH_DBI_ALL(TXN, I) TXN_FOREACH_DBI_FROM(TXN, I, 0)
+#define TXN_FOREACH_DBI_USER(TXN, I) TXN_FOREACH_DBI_FROM(TXN, I, CORE_DBS)
+
+MDBX_INTERNAL int dbi_import(MDBX_txn *txn, const size_t dbi);
+
+struct dbi_snap_result {
+  uint32_t sequence;
+  unsigned flags;
+};
+MDBX_INTERNAL struct dbi_snap_result dbi_snap(const MDBX_env *env, const size_t dbi);
+
+MDBX_INTERNAL int dbi_update(MDBX_txn *txn, bool keep);
+
+static inline uint8_t dbi_state(const MDBX_txn *txn, const size_t dbi) {
+  STATIC_ASSERT((int)DBI_DIRTY == MDBX_DBI_DIRTY && (int)DBI_STALE == MDBX_DBI_STALE &&
+                (int)DBI_FRESH == MDBX_DBI_FRESH && (int)DBI_CREAT == MDBX_DBI_CREAT);
+
+#if MDBX_ENABLE_DBI_SPARSE
+  const size_t bitmap_chunk = CHAR_BIT * sizeof(txn->dbi_sparse[0]);
+  const size_t bitmap_indx = dbi / bitmap_chunk;
+  const size_t bitmap_mask = (size_t)1 << dbi % bitmap_chunk;
+  return likely(dbi < txn->n_dbi && (txn->dbi_sparse[bitmap_indx] & bitmap_mask) != 0) ? txn->dbi_state[dbi] : 0;
+#else
+  return likely(dbi < txn->n_dbi) ? txn->dbi_state[dbi] : 0;
+#endif /* MDBX_ENABLE_DBI_SPARSE */
+}
+
+static inline bool dbi_changed(const MDBX_txn *txn, const size_t dbi) {
+  const MDBX_env *const env = txn->env;
+  eASSERT(env, dbi_state(txn, dbi) & DBI_LINDO);
+  const uint32_t snap_seq = atomic_load32(&env->dbi_seqs[dbi], mo_AcquireRelease);
+  return unlikely(snap_seq != txn->dbi_seqs[dbi]);
+}
+
+static inline int dbi_check(const MDBX_txn *txn, const size_t dbi) {
+  const uint8_t state = dbi_state(txn, dbi);
+  if (likely((state & DBI_LINDO) != 0 && !dbi_changed(txn, dbi)))
+    return (state & DBI_VALID) ? MDBX_SUCCESS : MDBX_BAD_DBI;
+
+  /* Медленный путь: ленивая до-инициализацяи и импорт */
+  return dbi_import((MDBX_txn *)txn, dbi);
+}
+
+static inline uint32_t dbi_seq_next(const MDBX_env *const env, size_t dbi) {
+  uint32_t v = atomic_load32(&env->dbi_seqs[dbi], mo_AcquireRelease) + 1;
+  return v ? v : 1;
+}
+
+MDBX_INTERNAL int dbi_open(MDBX_txn *txn, const MDBX_val *const name, unsigned user_flags, MDBX_dbi *dbi,
+                           MDBX_cmp_func *keycmp, MDBX_cmp_func *datacmp);
+
+MDBX_INTERNAL int dbi_bind(MDBX_txn *txn, const size_t dbi, unsigned user_flags, MDBX_cmp_func *keycmp,
+                           MDBX_cmp_func *datacmp);
+
+typedef struct defer_free_item {
+  struct defer_free_item *next;
+  uint64_t timestamp;
+} defer_free_item_t;
+
+MDBX_INTERNAL int dbi_defer_release(MDBX_env *const env, defer_free_item_t *const chain);
+MDBX_INTERNAL int dbi_close_release(MDBX_env *env, MDBX_dbi dbi);
+MDBX_INTERNAL const tree_t *dbi_dig(const MDBX_txn *txn, const size_t dbi, tree_t *fallback);
+
+struct dbi_rename_result {
+  defer_free_item_t *defer;
+  int err;
+};
+
+MDBX_INTERNAL struct dbi_rename_result dbi_rename_locked(MDBX_txn *txn, MDBX_dbi dbi, MDBX_val new_name);

src/debug_begin.h

@@ -0,0 +1,36 @@
+#if defined(__GNUC__) && !defined(__LCC__)
+
+#pragma push_macro("TRACE")
+#pragma push_macro("DEBUG")
+#pragma push_macro("VERBOSE")
+#pragma push_macro("NOTICE")
+#pragma push_macro("WARNING")
+#pragma push_macro("ERROR")
+#pragma push_macro("eASSERT")
+
+#undef TRACE
+#define TRACE(fmt, ...) debug_log(MDBX_LOG_TRACE, __func__, __LINE__, fmt "\n", __VA_ARGS__)
+
+#undef DEBUG
+#define DEBUG(fmt, ...) debug_log(MDBX_LOG_DEBUG, __func__, __LINE__, fmt "\n", __VA_ARGS__)
+
+#undef VERBOSE
+#define VERBOSE(fmt, ...) debug_log(MDBX_LOG_VERBOSE, __func__, __LINE__, fmt "\n", __VA_ARGS__)
+
+#undef NOTICE
+#define NOTICE(fmt, ...) debug_log(MDBX_LOG_NOTICE, __func__, __LINE__, fmt "\n", __VA_ARGS__)
+
+#undef WARNING
+#define WARNING(fmt, ...) debug_log(MDBX_LOG_WARN, __func__, __LINE__, fmt "\n", __VA_ARGS__)
+
+#undef ERROR
+#define ERROR(fmt, ...) debug_log(MDBX_LOG_ERROR, __func__, __LINE__, fmt "\n", __VA_ARGS__)
+
+#undef eASSERT
+#define eASSERT(env, expr) ENSURE(env, expr)
+
+#if !defined(__clang__)
+#pragma GCC optimize("-Og")
+#endif
+
+#endif /* GCC only */

src/debug_end.h

@@ -0,0 +1,15 @@
+#if defined(__GNUC__) && !defined(__LCC__)
+
+#pragma pop_macro("TRACE")
+#pragma pop_macro("DEBUG")
+#pragma pop_macro("VERBOSE")
+#pragma pop_macro("NOTICE")
+#pragma pop_macro("WARNING")
+#pragma pop_macro("ERROR")
+#pragma pop_macro("eASSERT")
+
+#if !defined(__clang__)
+#pragma GCC reset_options
+#endif
+
+#endif /* GCC only */

src/defs.h

@@ -1,428 +0,0 @@
-/*
- * Copyright 2015-2018 Leonid Yuriev <leo@yuriev.ru>
- * and other libmdbx authors: please see AUTHORS file.
- * All rights reserved.
- *
- * Redistribution and use in source and binary forms, with or without
- * modification, are permitted only as authorized by the OpenLDAP
- * Public License.
- *
- * A copy of this license is available in the file LICENSE in the
- * top-level directory of the distribution or, alternatively, at
- * <http://www.OpenLDAP.org/license.html>.
- */
-
-#pragma once
-/* *INDENT-OFF* */
-/* clang-format off */
-
-#ifndef __GNUC_PREREQ
-#   if defined(__GNUC__) && defined(__GNUC_MINOR__)
-#       define __GNUC_PREREQ(maj, min) \
-          ((__GNUC__ << 16) + __GNUC_MINOR__ >= ((maj) << 16) + (min))
-#   else
-#       define __GNUC_PREREQ(maj, min) (0)
-#   endif
-#endif /* __GNUC_PREREQ */
-
-#ifndef __CLANG_PREREQ
-#   ifdef __clang__
-#       define __CLANG_PREREQ(maj,min) \
-          ((__clang_major__ << 16) + __clang_minor__ >= ((maj) << 16) + (min))
-#   else
-#       define __CLANG_PREREQ(maj,min) (0)
-#   endif
-#endif /* __CLANG_PREREQ */
-
-#ifndef __GLIBC_PREREQ
-#   if defined(__GLIBC__) && defined(__GLIBC_MINOR__)
-#       define __GLIBC_PREREQ(maj, min) \
-          ((__GLIBC__ << 16) + __GLIBC_MINOR__ >= ((maj) << 16) + (min))
-#   else
-#       define __GLIBC_PREREQ(maj, min) (0)
-#   endif
-#endif /* __GLIBC_PREREQ */
-
-#ifndef __has_attribute
-#   define __has_attribute(x) (0)
-#endif
-
-#ifndef __has_feature
-#   define __has_feature(x) (0)
-#endif
-
-#ifndef __has_extension
-#   define __has_extension(x) (0)
-#endif
-
-#ifndef __has_builtin
-#   define __has_builtin(x) (0)
-#endif
-
-#ifndef __has_warning
-#   define __has_warning(x) (0)
-#endif
-
-#ifndef __has_include
-#   define __has_include(x) (0)
-#endif
-
-#if __has_feature(thread_sanitizer)
-#   define __SANITIZE_THREAD__ 1
-#endif
-
-#if __has_feature(address_sanitizer)
-#   define __SANITIZE_ADDRESS__ 1
-#endif
-
-/*----------------------------------------------------------------------------*/
-
-#ifndef __extern_C
-#   ifdef __cplusplus
-#       define __extern_C extern "C"
-#   else
-#       define __extern_C
-#   endif
-#endif /* __extern_C */
-
-#ifndef __cplusplus
-#   ifndef bool
-#       define bool _Bool
-#   endif
-#   ifndef true
-#       define true (1)
-#   endif
-#   ifndef false
-#       define false (0)
-#   endif
-#endif
-
-#if !defined(nullptr) && !defined(__cplusplus) || (__cplusplus < 201103L && !defined(_MSC_VER))
-#   define nullptr NULL
-#endif
-
-/*----------------------------------------------------------------------------*/
-
-#ifndef __alwaysinline
-#   if defined(__GNUC__) || __has_attribute(always_inline)
-#       define __alwaysinline __inline __attribute__((always_inline))
-#   elif defined(_MSC_VER)
-#       define __alwaysinline __forceinline
-#   else
-#       define __alwaysinline
-#   endif
-#endif /* __alwaysinline */
-
-#ifndef __noinline
-#   if defined(__GNUC__) || __has_attribute(noinline)
-#       define __noinline __attribute__((noinline))
-#   elif defined(_MSC_VER)
-#       define __noinline __declspec(noinline)
-#   elif defined(__SUNPRO_C) || defined(__sun) || defined(sun)
-#       define __noinline inline
-#   elif !defined(__INTEL_COMPILER)
-#       define __noinline /* FIXME ? */
-#   endif
-#endif /* __noinline */
-
-#ifndef __must_check_result
-#   if defined(__GNUC__) || __has_attribute(warn_unused_result)
-#       define __must_check_result __attribute__((warn_unused_result))
-#   else
-#       define __must_check_result
-#   endif
-#endif /* __must_check_result */
-
-#ifndef __deprecated
-#   if defined(__GNUC__) || __has_attribute(deprecated)
-#       define __deprecated __attribute__((deprecated))
-#   elif defined(_MSC_VER)
-#       define __deprecated __declspec(deprecated)
-#   else
-#       define __deprecated
-#   endif
-#endif /* __deprecated */
-
-#if !defined(__noop) && !defined(_MSC_VER)
-#	ifdef __cplusplus
-		static inline void __noop_consume_args() {}
-		template <typename First, typename... Rest>
-		static inline void
-		__noop_consume_args(const First &first, const Rest &... rest) {
-			(void) first; __noop_consume_args(rest...);
-		}
-#		define __noop(...) __noop_consume_args(__VA_ARGS__)
-#	elif defined(__GNUC__) && (!defined(__STRICT_ANSI__) || !__STRICT_ANSI__)
-		static __inline void __noop_consume_args(void* anchor, ...) {
-			(void) anchor;
-		}
-#		define __noop(...) __noop_consume_args(0, ##__VA_ARGS__)
-#	else
-#		define __noop(...) do {} while(0)
-#	endif
-#endif /* __noop */
-
-#ifndef __fallthrough
-#	if __GNUC_PREREQ(7, 0) || __has_attribute(fallthrough)
-#		define __fallthrough __attribute__((fallthrough))
-#	else
-#		define __fallthrough __noop()
-#	endif
-#endif /* __fallthrough */
-
-#ifndef __unreachable
-#	if __GNUC_PREREQ(4,5)
-#		define __unreachable() __builtin_unreachable()
-#	elif defined(_MSC_VER)
-#		define __unreachable() __assume(0)
-#	else
-#		define __unreachable() __noop()
-#	endif
-#endif /* __unreachable */
-
-#ifndef __prefetch
-#	if defined(__GNUC__) || defined(__clang__)
-#		define __prefetch(ptr) __builtin_prefetch(ptr)
-#	else
-#		define __prefetch(ptr) __noop(ptr)
-#	endif
-#endif /* __prefetch */
-
-#ifndef __noreturn
-#   if defined(__GNUC__) || __has_attribute(noreturn)
-#       define __noreturn __attribute__((noreturn))
-#   elif defined(_MSC_VER)
-#       define __noreturn __declspec(noreturn)
-#   else
-#       define __noreturn
-#   endif
-#endif /* __noreturn */
-
-#ifndef __nothrow
-#   if defined(__GNUC__) || __has_attribute(nothrow)
-#       define __nothrow __attribute__((nothrow))
-#   elif defined(_MSC_VER) && defined(__cplusplus)
-#       define __nothrow __declspec(nothrow)
-#   else
-#       define __nothrow
-#   endif
-#endif /* __nothrow */
-
-#ifndef __pure_function
-    /* Many functions have no effects except the return value and their
-     * return value depends only on the parameters and/or global variables.
-     * Such a function can be subject to common subexpression elimination
-     * and loop optimization just as an arithmetic operator would be.
-     * These functions should be declared with the attribute pure. */
-#   if defined(__GNUC__) || __has_attribute(pure)
-#       define __pure_function __attribute__((pure))
-#   else
-#       define __pure_function
-#   endif
-#endif /* __pure_function */
-
-#ifndef __const_function
-    /* Many functions do not examine any values except their arguments,
-     * and have no effects except the return value. Basically this is just
-     * slightly more strict class than the PURE attribute, since function
-     * is not allowed to read global memory.
-     *
-     * Note that a function that has pointer arguments and examines the
-     * data pointed to must not be declared const. Likewise, a function
-     * that calls a non-const function usually must not be const.
-     * It does not make sense for a const function to return void. */
-#   if defined(__GNUC__) || __has_attribute(const)
-#       define __const_function __attribute__((const))
-#   else
-#       define __const_function
-#   endif
-#endif /* __const_function */
-
-#ifndef __dll_hidden
-#   if defined(__GNUC__) || __has_attribute(visibility)
-#       define __hidden __attribute__((visibility("hidden")))
-#   else
-#       define __hidden
-#   endif
-#endif /* __dll_hidden */
-
-#ifndef __optimize
-#   if defined(__OPTIMIZE__)
-#     if defined(__clang__) && !__has_attribute(optimize)
-#           define __optimize(ops)
-#       elif defined(__GNUC__) || __has_attribute(optimize)
-#           define __optimize(ops) __attribute__((optimize(ops)))
-#       else
-#           define __optimize(ops)
-#       endif
-#   else
-#           define __optimize(ops)
-#   endif
-#endif /* __optimize */
-
-#ifndef __hot
-#   if defined(__OPTIMIZE__)
-#       if defined(__e2k__)
-#           define __hot __attribute__((hot)) __optimize(3)
-#       elif defined(__clang__) && !__has_attribute(hot)
-            /* just put frequently used functions in separate section */
-#           define __hot __attribute__((section("text.hot"))) __optimize("O3")
-#       elif defined(__GNUC__) || __has_attribute(hot)
-#           define __hot __attribute__((hot)) __optimize("O3")
-#       else
-#           define __hot  __optimize("O3")
-#       endif
-#   else
-#       define __hot
-#   endif
-#endif /* __hot */
-
-#ifndef __cold
-#   if defined(__OPTIMIZE__)
-#       if defined(__e2k__)
-#           define __cold __attribute__((cold)) __optimize(1)
-#       elif defined(__clang__) && !__has_attribute(cold)
-            /* just put infrequently used functions in separate section */
-#           define __cold __attribute__((section("text.unlikely"))) __optimize("Os")
-#       elif defined(__GNUC__) || __has_attribute(cold)
-#           define __cold __attribute__((cold)) __optimize("Os")
-#       else
-#           define __cold __optimize("Os")
-#       endif
-#   else
-#       define __cold
-#   endif
-#endif /* __cold */
-
-#ifndef __flatten
-#   if defined(__OPTIMIZE__) && (defined(__GNUC__) || __has_attribute(flatten))
-#       define __flatten __attribute__((flatten))
-#   else
-#       define __flatten
-#   endif
-#endif /* __flatten */
-
-#ifndef likely
-#   if defined(__GNUC__) || defined(__clang__)
-#       define likely(cond) __builtin_expect(!!(cond), 1)
-#   else
-#       define likely(x) (x)
-#   endif
-#endif /* likely */
-
-#ifndef unlikely
-#   if defined(__GNUC__) || defined(__clang__)
-#       define unlikely(cond) __builtin_expect(!!(cond), 0)
-#   else
-#       define unlikely(x) (x)
-#   endif
-#endif /* unlikely */
-
-/* Wrapper around __func__, which is a C99 feature */
-#if defined(__STDC_VERSION__) && __STDC_VERSION__ >= 199901L
-#   define mdbx_func_ __func__
-#elif (defined(__GNUC__) && __GNUC__ >= 2) || defined(__clang__) || defined(_MSC_VER)
-#   define mdbx_func_ __FUNCTION__
-#else
-#   define mdbx_func_ "<mdbx_unknown>"
-#endif
-
-#if defined(__GNUC__) || __has_attribute(format)
-#define __printf_args(format_index, first_arg)                                 \
-  __attribute__((format(printf, format_index, first_arg)))
-#else
-#define __printf_args(format_index, first_arg)
-#endif
-
-/*----------------------------------------------------------------------------*/
-
-#if defined(USE_VALGRIND)
-#   include <valgrind/memcheck.h>
-#   ifndef VALGRIND_DISABLE_ADDR_ERROR_REPORTING_IN_RANGE
-        /* LY: available since Valgrind 3.10 */
-#       define VALGRIND_DISABLE_ADDR_ERROR_REPORTING_IN_RANGE(a,s)
-#       define VALGRIND_ENABLE_ADDR_ERROR_REPORTING_IN_RANGE(a,s)
-#   endif
-#elif !defined(RUNNING_ON_VALGRIND)
-#   define VALGRIND_CREATE_MEMPOOL(h,r,z)
-#   define VALGRIND_DESTROY_MEMPOOL(h)
-#   define VALGRIND_MEMPOOL_TRIM(h,a,s)
-#   define VALGRIND_MEMPOOL_ALLOC(h,a,s)
-#   define VALGRIND_MEMPOOL_FREE(h,a)
-#   define VALGRIND_MEMPOOL_CHANGE(h,a,b,s)
-#   define VALGRIND_MAKE_MEM_NOACCESS(a,s)
-#   define VALGRIND_MAKE_MEM_DEFINED(a,s)
-#   define VALGRIND_MAKE_MEM_UNDEFINED(a,s)
-#   define VALGRIND_DISABLE_ADDR_ERROR_REPORTING_IN_RANGE(a,s)
-#   define VALGRIND_ENABLE_ADDR_ERROR_REPORTING_IN_RANGE(a,s)
-#   define VALGRIND_CHECK_MEM_IS_ADDRESSABLE(a,s) (0)
-#   define VALGRIND_CHECK_MEM_IS_DEFINED(a,s) (0)
-#   define RUNNING_ON_VALGRIND (0)
-#endif /* USE_VALGRIND */
-
-#ifdef __SANITIZE_ADDRESS__
-#   include <sanitizer/asan_interface.h>
-#elif !defined(ASAN_POISON_MEMORY_REGION)
-#   define ASAN_POISON_MEMORY_REGION(addr, size) \
-        ((void)(addr), (void)(size))
-#   define ASAN_UNPOISON_MEMORY_REGION(addr, size) \
-        ((void)(addr), (void)(size))
-#endif /* __SANITIZE_ADDRESS__ */
-
-/*----------------------------------------------------------------------------*/
-
-#ifndef ARRAY_LENGTH
-#   ifdef __cplusplus
-        template <typename T, size_t N>
-        char (&__ArraySizeHelper(T (&array)[N]))[N];
-#       define ARRAY_LENGTH(array) (sizeof(::__ArraySizeHelper(array)))
-#   else
-#       define ARRAY_LENGTH(array) (sizeof(array) / sizeof(array[0]))
-#   endif
-#endif /* ARRAY_LENGTH */
-
-#ifndef ARRAY_END
-#   define ARRAY_END(array) (&array[ARRAY_LENGTH(array)])
-#endif /* ARRAY_END */
-
-#ifndef STRINGIFY
-#   define STRINGIFY_HELPER(x) #x
-#   define STRINGIFY(x) STRINGIFY_HELPER(x)
-#endif /* STRINGIFY */
-
-#ifndef offsetof
-#   define offsetof(type, member)  __builtin_offsetof(type, member)
-#endif /* offsetof */
-
-#ifndef container_of
-#   define container_of(ptr, type, member) \
-        ((type *)((char *)(ptr) - offsetof(type, member)))
-#endif /* container_of */
-
-#define MDBX_TETRAD(a, b, c, d)                                                \
-  ((uint32_t)(a) << 24 | (uint32_t)(b) << 16 | (uint32_t)(c) << 8 | (d))
-
-#define MDBX_STRING_TETRAD(str) MDBX_TETRAD(str[0], str[1], str[2], str[3])
-
-#define FIXME "FIXME: " __FILE__ ", " STRINGIFY(__LINE__)
-
-#ifndef STATIC_ASSERT_MSG
-#   if (defined(__STDC_VERSION__) && __STDC_VERSION__ >= 201112L) \
-          || __has_feature(c_static_assert)
-#       define STATIC_ASSERT_MSG(expr, msg) _Static_assert(expr, msg)
-#   elif defined(static_assert)
-#       define STATIC_ASSERT_MSG(expr, msg) static_assert(expr, msg)
-#   elif defined(_MSC_VER)
-#       include <crtdbg.h>
-#       define STATIC_ASSERT_MSG(expr, msg) _STATIC_ASSERT(expr)
-#   else
-#       define STATIC_ASSERT_MSG(expr, msg) switch (0) {case 0:case (expr):;}
-#   endif
-#endif /* STATIC_ASSERT */
-
-#ifndef STATIC_ASSERT
-#   define STATIC_ASSERT(expr) STATIC_ASSERT_MSG(expr, #expr)
-#endif
-
-/* *INDENT-ON* */
-/* clang-format on */

src/dpl.c

@@ -0,0 +1,488 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+static inline size_t dpl_size2bytes(ptrdiff_t size) {
+  assert(size > CURSOR_STACK_SIZE && (size_t)size <= PAGELIST_LIMIT);
+#if MDBX_DPL_PREALLOC_FOR_RADIXSORT
+  size += size;
+#endif /* MDBX_DPL_PREALLOC_FOR_RADIXSORT */
+  STATIC_ASSERT(MDBX_ASSUME_MALLOC_OVERHEAD + sizeof(dpl_t) +
+                    (PAGELIST_LIMIT * (MDBX_DPL_PREALLOC_FOR_RADIXSORT + 1)) * sizeof(dp_t) +
+                    MDBX_PNL_GRANULATE * sizeof(void *) * 2 <
+                SIZE_MAX / 4 * 3);
+  size_t bytes = ceil_powerof2(MDBX_ASSUME_MALLOC_OVERHEAD + sizeof(dpl_t) + size * sizeof(dp_t),
+                               MDBX_PNL_GRANULATE * sizeof(void *) * 2) -
+                 MDBX_ASSUME_MALLOC_OVERHEAD;
+  return bytes;
+}
+
+static inline size_t dpl_bytes2size(const ptrdiff_t bytes) {
+  size_t size = (bytes - sizeof(dpl_t)) / sizeof(dp_t);
+#if MDBX_DPL_PREALLOC_FOR_RADIXSORT
+  size >>= 1;
+#endif /* MDBX_DPL_PREALLOC_FOR_RADIXSORT */
+  assert(size > CURSOR_STACK_SIZE && size <= PAGELIST_LIMIT + MDBX_PNL_GRANULATE);
+  return size;
+}
+
+void dpl_free(MDBX_txn *txn) {
+  if (likely(txn->wr.dirtylist)) {
+    osal_free(txn->wr.dirtylist);
+    txn->wr.dirtylist = nullptr;
+  }
+}
+
+dpl_t *dpl_reserve(MDBX_txn *txn, size_t size) {
+  tASSERT(txn, (txn->flags & MDBX_TXN_RDONLY) == 0);
+  tASSERT(txn, (txn->flags & MDBX_WRITEMAP) == 0 || MDBX_AVOID_MSYNC);
+
+  size_t bytes = dpl_size2bytes((size < PAGELIST_LIMIT) ? size : PAGELIST_LIMIT);
+  dpl_t *const dl = osal_realloc(txn->wr.dirtylist, bytes);
+  if (likely(dl)) {
+#ifdef osal_malloc_usable_size
+    bytes = osal_malloc_usable_size(dl);
+#endif /* osal_malloc_usable_size */
+    dl->detent = dpl_bytes2size(bytes);
+    tASSERT(txn, txn->wr.dirtylist == nullptr || dl->length <= dl->detent);
+    txn->wr.dirtylist = dl;
+  }
+  return dl;
+}
+
+int dpl_alloc(MDBX_txn *txn) {
+  tASSERT(txn, (txn->flags & MDBX_TXN_RDONLY) == 0);
+  tASSERT(txn, (txn->flags & MDBX_WRITEMAP) == 0 || MDBX_AVOID_MSYNC);
+
+  const size_t wanna = (txn->env->options.dp_initial < txn->geo.upper) ? txn->env->options.dp_initial : txn->geo.upper;
+#if MDBX_FORCE_ASSERTIONS || MDBX_DEBUG
+  if (txn->wr.dirtylist)
+    /* обнуляем чтобы не сработал ассерт внутри dpl_reserve() */
+    txn->wr.dirtylist->sorted = txn->wr.dirtylist->length = 0;
+#endif /* asertions enabled */
+  if (unlikely(!txn->wr.dirtylist || txn->wr.dirtylist->detent < wanna || txn->wr.dirtylist->detent > wanna + wanna) &&
+      unlikely(!dpl_reserve(txn, wanna)))
+    return MDBX_ENOMEM;
+
+  /* LY: wr.dirtylist не может быть nullptr, так как либо уже выделен, либо будет выделен в dpl_reserve(). */
+  /* coverity[var_deref_model] */
+  dpl_clear(txn->wr.dirtylist);
+  return MDBX_SUCCESS;
+}
+
+#define MDBX_DPL_EXTRACT_KEY(ptr) ((ptr)->pgno)
+RADIXSORT_IMPL(dp, dp_t, MDBX_DPL_EXTRACT_KEY, MDBX_DPL_PREALLOC_FOR_RADIXSORT, 1)
+
+#define DP_SORT_CMP(first, last) ((first).pgno < (last).pgno)
+SORT_IMPL(dp_sort, false, dp_t, DP_SORT_CMP)
+
+__hot __noinline dpl_t *dpl_sort_slowpath(const MDBX_txn *txn) {
+  tASSERT(txn, (txn->flags & MDBX_TXN_RDONLY) == 0);
+  tASSERT(txn, (txn->flags & MDBX_WRITEMAP) == 0 || MDBX_AVOID_MSYNC);
+
+  dpl_t *dl = txn->wr.dirtylist;
+  assert(dl->items[0].pgno == 0 && dl->items[dl->length + 1].pgno == P_INVALID);
+  const size_t unsorted = dl->length - dl->sorted;
+  if (likely(unsorted < MDBX_RADIXSORT_THRESHOLD) || unlikely(!dp_radixsort(dl->items + 1, dl->length))) {
+    if (dl->sorted > unsorted / 4 + 4 &&
+        (MDBX_DPL_PREALLOC_FOR_RADIXSORT || dl->length + unsorted < dl->detent + dpl_gap_mergesort)) {
+      dp_t *const sorted_begin = dl->items + 1;
+      dp_t *const sorted_end = sorted_begin + dl->sorted;
+      dp_t *const end =
+          dl->items + (MDBX_DPL_PREALLOC_FOR_RADIXSORT ? dl->length + dl->length + 1 : dl->detent + dpl_reserve_gap);
+      dp_t *const tmp = end - unsorted;
+      assert(dl->items + dl->length + 1 < tmp);
+      /* copy unsorted to the end of allocated space and sort it */
+      memcpy(tmp, sorted_end, unsorted * sizeof(dp_t));
+      dp_sort(tmp, tmp + unsorted);
+      /* merge two parts from end to begin */
+      dp_t *__restrict w = dl->items + dl->length;
+      dp_t *__restrict l = dl->items + dl->sorted;
+      dp_t *__restrict r = end - 1;
+      do {
+        const bool cmp = expect_with_probability(l->pgno > r->pgno, 0, .5);
+#if defined(__LCC__) || __CLANG_PREREQ(13, 0) || !MDBX_HAVE_CMOV
+        *w = cmp ? *l-- : *r--;
+#else
+        *w = cmp ? *l : *r;
+        l -= cmp;
+        r += (ptrdiff_t)cmp - 1;
+#endif
+      } while (likely(--w > l));
+      assert(r == tmp - 1);
+      assert(dl->items[0].pgno == 0 && dl->items[dl->length + 1].pgno == P_INVALID);
+      if (ASSERT_ENABLED())
+        for (size_t i = 0; i <= dl->length; ++i)
+          assert(dl->items[i].pgno < dl->items[i + 1].pgno);
+    } else {
+      dp_sort(dl->items + 1, dl->items + dl->length + 1);
+      assert(dl->items[0].pgno == 0 && dl->items[dl->length + 1].pgno == P_INVALID);
+    }
+  } else {
+    assert(dl->items[0].pgno == 0 && dl->items[dl->length + 1].pgno == P_INVALID);
+  }
+  dl->sorted = dl->length;
+  return dl;
+}
+
+/* Returns the index of the first dirty-page whose pgno
+ * member is greater than or equal to id. */
+#define DP_SEARCH_CMP(dp, id) ((dp).pgno < (id))
+SEARCH_IMPL(dp_bsearch, dp_t, pgno_t, DP_SEARCH_CMP)
+
+__hot __noinline size_t dpl_search(const MDBX_txn *txn, pgno_t pgno) {
+  tASSERT(txn, (txn->flags & MDBX_TXN_RDONLY) == 0);
+  tASSERT(txn, (txn->flags & MDBX_WRITEMAP) == 0 || MDBX_AVOID_MSYNC);
+
+  dpl_t *dl = txn->wr.dirtylist;
+  assert(dl->items[0].pgno == 0 && dl->items[dl->length + 1].pgno == P_INVALID);
+  if (AUDIT_ENABLED()) {
+    for (const dp_t *ptr = dl->items + dl->sorted; --ptr > dl->items;) {
+      assert(ptr[0].pgno < ptr[1].pgno);
+      assert(ptr[0].pgno >= NUM_METAS);
+    }
+  }
+
+  switch (dl->length - dl->sorted) {
+  default:
+    /* sort a whole */
+    dpl_sort_slowpath(txn);
+    break;
+  case 0:
+    /* whole sorted cases */
+    break;
+
+#define LINEAR_SEARCH_CASE(N)                                                                                          \
+  case N:                                                                                                              \
+    if (dl->items[dl->length - N + 1].pgno == pgno)                                                                    \
+      return dl->length - N + 1;                                                                                       \
+    __fallthrough
+
+    /* use linear scan until the threshold */
+    LINEAR_SEARCH_CASE(7); /* fall through */
+    LINEAR_SEARCH_CASE(6); /* fall through */
+    LINEAR_SEARCH_CASE(5); /* fall through */
+    LINEAR_SEARCH_CASE(4); /* fall through */
+    LINEAR_SEARCH_CASE(3); /* fall through */
+    LINEAR_SEARCH_CASE(2); /* fall through */
+  case 1:
+    if (dl->items[dl->length].pgno == pgno)
+      return dl->length;
+    /* continue bsearch on the sorted part */
+    break;
+  }
+  return dp_bsearch(dl->items + 1, dl->sorted, pgno) - dl->items;
+}
+
+const page_t *debug_dpl_find(const MDBX_txn *txn, const pgno_t pgno) {
+  tASSERT(txn, (txn->flags & MDBX_TXN_RDONLY) == 0);
+  const dpl_t *dl = txn->wr.dirtylist;
+  if (dl) {
+    tASSERT(txn, (txn->flags & MDBX_WRITEMAP) == 0 || MDBX_AVOID_MSYNC);
+    assert(dl->items[0].pgno == 0 && dl->items[dl->length + 1].pgno == P_INVALID);
+    for (size_t i = dl->length; i > dl->sorted; --i)
+      if (dl->items[i].pgno == pgno)
+        return dl->items[i].ptr;
+
+    if (dl->sorted) {
+      const size_t i = dp_bsearch(dl->items + 1, dl->sorted, pgno) - dl->items;
+      if (dl->items[i].pgno == pgno)
+        return dl->items[i].ptr;
+    }
+  } else {
+    tASSERT(txn, (txn->flags & MDBX_WRITEMAP) != 0 && !MDBX_AVOID_MSYNC);
+  }
+  return nullptr;
+}
+
+void dpl_remove_ex(const MDBX_txn *txn, size_t i, size_t npages) {
+  tASSERT(txn, (txn->flags & MDBX_TXN_RDONLY) == 0);
+  tASSERT(txn, (txn->flags & MDBX_WRITEMAP) == 0 || MDBX_AVOID_MSYNC);
+
+  dpl_t *dl = txn->wr.dirtylist;
+  assert((intptr_t)i > 0 && i <= dl->length);
+  assert(dl->items[0].pgno == 0 && dl->items[dl->length + 1].pgno == P_INVALID);
+  dl->pages_including_loose -= npages;
+  dl->sorted -= dl->sorted >= i;
+  dl->length -= 1;
+  memmove(dl->items + i, dl->items + i + 1, (dl->length - i + 2) * sizeof(dl->items[0]));
+  assert(dl->items[0].pgno == 0 && dl->items[dl->length + 1].pgno == P_INVALID);
+}
+
+int __must_check_result dpl_append(MDBX_txn *txn, pgno_t pgno, page_t *page, size_t npages) {
+  tASSERT(txn, (txn->flags & MDBX_TXN_RDONLY) == 0);
+  tASSERT(txn, (txn->flags & MDBX_WRITEMAP) == 0 || MDBX_AVOID_MSYNC);
+  const dp_t dp = {page, pgno, (pgno_t)npages};
+  if ((txn->flags & MDBX_WRITEMAP) == 0) {
+    size_t *const ptr = ptr_disp(page, -(ptrdiff_t)sizeof(size_t));
+    *ptr = txn->wr.dirtylru;
+  }
+
+  dpl_t *dl = txn->wr.dirtylist;
+  tASSERT(txn, dl->length <= PAGELIST_LIMIT + MDBX_PNL_GRANULATE);
+  tASSERT(txn, dl->items[0].pgno == 0 && dl->items[dl->length + 1].pgno == P_INVALID);
+  if (AUDIT_ENABLED()) {
+    for (size_t i = dl->length; i > 0; --i) {
+      assert(dl->items[i].pgno != dp.pgno);
+      if (unlikely(dl->items[i].pgno == dp.pgno)) {
+        ERROR("Page %u already exist in the DPL at %zu", dp.pgno, i);
+        return MDBX_PROBLEM;
+      }
+    }
+  }
+
+  if (unlikely(dl->length == dl->detent)) {
+    if (unlikely(dl->detent >= PAGELIST_LIMIT)) {
+      ERROR("DPL is full (PAGELIST_LIMIT %zu)", PAGELIST_LIMIT);
+      return MDBX_TXN_FULL;
+    }
+    const size_t size = (dl->detent < MDBX_PNL_INITIAL * 42) ? dl->detent + dl->detent : dl->detent + dl->detent / 2;
+    dl = dpl_reserve(txn, size);
+    if (unlikely(!dl))
+      return MDBX_ENOMEM;
+    tASSERT(txn, dl->length < dl->detent);
+  }
+
+  /* Сортировка нужна для быстрого поиска, используем несколько тактик:
+   *  1) Сохраняем упорядоченность при естественной вставке в нужном порядке.
+   *  2) Добавляем в не-сортированный хвост, который сортируем и сливаем
+   *     с отсортированной головой по необходимости, а пока хвост короткий
+   *     ищем в нём сканированием, избегая большой пересортировки.
+   *  3) Если не-сортированный хвост короткий, а добавляемый элемент близок
+   *     к концу отсортированной головы, то выгоднее сразу вставить элемент
+   *     в нужное место.
+   *
+   * Алгоритмически:
+   *  - добавлять в не-сортированный хвост следует только если вставка сильно
+   *    дорогая, т.е. если целевая позиция элемента сильно далека от конца;
+   *  - для быстрой проверки достаточно сравнить добавляемый элемент с отстоящим
+   *    от конца на максимально-приемлемое расстояние;
+   *  - если список короче, либо элемент в этой позиции меньше вставляемого,
+   *    то следует перемещать элементы и вставлять в отсортированную голову;
+   *  - если не-сортированный хвост длиннее, либо элемент в этой позиции больше,
+   *    то следует добавлять в не-сортированный хвост. */
+
+  dl->pages_including_loose += npages;
+  dp_t *i = dl->items + dl->length;
+
+  const ptrdiff_t pivot = (ptrdiff_t)dl->length - dpl_insertion_threshold;
+#if MDBX_HAVE_CMOV
+  const pgno_t pivot_pgno =
+      dl->items[(dl->length < dpl_insertion_threshold) ? 0 : dl->length - dpl_insertion_threshold].pgno;
+#endif /* MDBX_HAVE_CMOV */
+
+  /* copy the stub beyond the end */
+  i[2] = i[1];
+  dl->length += 1;
+
+  if (likely(pivot <= (ptrdiff_t)dl->sorted) &&
+#if MDBX_HAVE_CMOV
+      pivot_pgno < dp.pgno) {
+#else
+      (pivot <= 0 || dl->items[pivot].pgno < dp.pgno)) {
+#endif /* MDBX_HAVE_CMOV */
+    dl->sorted += 1;
+
+    /* сдвигаем несортированный хвост */
+    while (i >= dl->items + dl->sorted) {
+#if !defined(__GNUC__) /* пытаемся избежать вызова memmove() */
+      i[1] = *i;
+#elif MDBX_WORDBITS == 64 && (defined(__SIZEOF_INT128__) || (defined(_INTEGRAL_MAX_BITS) && _INTEGRAL_MAX_BITS >= 128))
+      STATIC_ASSERT(sizeof(dp) == sizeof(__uint128_t));
+      ((__uint128_t *)i)[1] = *(volatile __uint128_t *)i;
+#else
+    i[1].ptr = i->ptr;
+    i[1].pgno = i->pgno;
+    i[1].npages = i->npages;
+#endif
+      --i;
+    }
+    /* ищем нужную позицию сдвигая отсортированные элементы */
+    while (i->pgno > pgno) {
+      tASSERT(txn, i > dl->items);
+      i[1] = *i;
+      --i;
+    }
+    tASSERT(txn, i->pgno < dp.pgno);
+  }
+
+  i[1] = dp;
+  assert(dl->items[0].pgno == 0 && dl->items[dl->length + 1].pgno == P_INVALID);
+  assert(dl->sorted <= dl->length);
+  return MDBX_SUCCESS;
+}
+
+__cold bool dpl_check(MDBX_txn *txn) {
+  tASSERT(txn, (txn->flags & MDBX_TXN_RDONLY) == 0);
+  const dpl_t *const dl = txn->wr.dirtylist;
+  if (!dl) {
+    tASSERT(txn, (txn->flags & MDBX_WRITEMAP) != 0 && !MDBX_AVOID_MSYNC);
+    return true;
+  }
+  tASSERT(txn, (txn->flags & MDBX_WRITEMAP) == 0 || MDBX_AVOID_MSYNC);
+
+  assert(dl->items[0].pgno == 0 && dl->items[dl->length + 1].pgno == P_INVALID);
+  tASSERT(txn,
+          txn->wr.dirtyroom + dl->length == (txn->parent ? txn->parent->wr.dirtyroom : txn->env->options.dp_limit));
+
+  if (!AUDIT_ENABLED())
+    return true;
+
+  size_t loose = 0, pages = 0;
+  for (size_t i = dl->length; i > 0; --i) {
+    const page_t *const dp = dl->items[i].ptr;
+    if (!dp)
+      continue;
+
+    tASSERT(txn, dp->pgno == dl->items[i].pgno);
+    if (unlikely(dp->pgno != dl->items[i].pgno))
+      return false;
+
+    if ((txn->flags & MDBX_WRITEMAP) == 0) {
+      const uint32_t age = dpl_age(txn, i);
+      tASSERT(txn, age < UINT32_MAX / 3);
+      if (unlikely(age > UINT32_MAX / 3))
+        return false;
+    }
+
+    tASSERT(txn, dp->flags == P_LOOSE || is_modifable(txn, dp));
+    if (dp->flags == P_LOOSE) {
+      loose += 1;
+    } else if (unlikely(!is_modifable(txn, dp)))
+      return false;
+
+    const unsigned num = dpl_npages(dl, i);
+    pages += num;
+    tASSERT(txn, txn->geo.first_unallocated >= dp->pgno + num);
+    if (unlikely(txn->geo.first_unallocated < dp->pgno + num))
+      return false;
+
+    if (i < dl->sorted) {
+      tASSERT(txn, dl->items[i + 1].pgno >= dp->pgno + num);
+      if (unlikely(dl->items[i + 1].pgno < dp->pgno + num))
+        return false;
+    }
+
+    const size_t rpa = pnl_search(txn->wr.repnl, dp->pgno, txn->geo.first_unallocated);
+    tASSERT(txn, rpa > MDBX_PNL_GETSIZE(txn->wr.repnl) || txn->wr.repnl[rpa] != dp->pgno);
+    if (rpa <= MDBX_PNL_GETSIZE(txn->wr.repnl) && unlikely(txn->wr.repnl[rpa] == dp->pgno))
+      return false;
+    if (num > 1) {
+      const size_t rpb = pnl_search(txn->wr.repnl, dp->pgno + num - 1, txn->geo.first_unallocated);
+      tASSERT(txn, rpa == rpb);
+      if (unlikely(rpa != rpb))
+        return false;
+    }
+  }
+
+  tASSERT(txn, loose == txn->wr.loose_count);
+  if (unlikely(loose != txn->wr.loose_count))
+    return false;
+
+  tASSERT(txn, pages == dl->pages_including_loose);
+  if (unlikely(pages != dl->pages_including_loose))
+    return false;
+
+  for (size_t i = 1; i <= MDBX_PNL_GETSIZE(txn->wr.retired_pages); ++i) {
+    const page_t *const dp = debug_dpl_find(txn, txn->wr.retired_pages[i]);
+    tASSERT(txn, !dp);
+    if (unlikely(dp))
+      return false;
+  }
+
+  return true;
+}
+
+/*----------------------------------------------------------------------------*/
+
+__noinline void dpl_lru_reduce(MDBX_txn *txn) {
+  VERBOSE("lru-reduce %u -> %u", txn->wr.dirtylru, txn->wr.dirtylru >> 1);
+  tASSERT(txn, (txn->flags & (MDBX_TXN_RDONLY | MDBX_WRITEMAP)) == 0);
+  do {
+    txn->wr.dirtylru >>= 1;
+    dpl_t *dl = txn->wr.dirtylist;
+    for (size_t i = 1; i <= dl->length; ++i) {
+      size_t *const ptr = ptr_disp(dl->items[i].ptr, -(ptrdiff_t)sizeof(size_t));
+      *ptr >>= 1;
+    }
+    txn = txn->parent;
+  } while (txn);
+}
+
+void dpl_sift(MDBX_txn *const txn, pnl_t pl, const bool spilled) {
+  tASSERT(txn, (txn->flags & MDBX_TXN_RDONLY) == 0);
+  tASSERT(txn, (txn->flags & MDBX_WRITEMAP) == 0 || MDBX_AVOID_MSYNC);
+  if (MDBX_PNL_GETSIZE(pl) && txn->wr.dirtylist->length) {
+    tASSERT(txn, pnl_check_allocated(pl, (size_t)txn->geo.first_unallocated << spilled));
+    dpl_t *dl = dpl_sort(txn);
+
+    /* Scanning in ascend order */
+    const intptr_t step = MDBX_PNL_ASCENDING ? 1 : -1;
+    const intptr_t begin = MDBX_PNL_ASCENDING ? 1 : MDBX_PNL_GETSIZE(pl);
+    const intptr_t end = MDBX_PNL_ASCENDING ? MDBX_PNL_GETSIZE(pl) + 1 : 0;
+    tASSERT(txn, pl[begin] <= pl[end - step]);
+
+    size_t w, r = dpl_search(txn, pl[begin] >> spilled);
+    tASSERT(txn, dl->sorted == dl->length);
+    for (intptr_t i = begin; r <= dl->length;) { /* scan loop */
+      assert(i != end);
+      tASSERT(txn, !spilled || (pl[i] & 1) == 0);
+      pgno_t pl_pgno = pl[i] >> spilled;
+      pgno_t dp_pgno = dl->items[r].pgno;
+      if (likely(dp_pgno != pl_pgno)) {
+        const bool cmp = dp_pgno < pl_pgno;
+        r += cmp;
+        i += cmp ? 0 : step;
+        if (likely(i != end))
+          continue;
+        return;
+      }
+
+      /* update loop */
+      unsigned npages;
+      w = r;
+    remove_dl:
+      npages = dpl_npages(dl, r);
+      dl->pages_including_loose -= npages;
+      if (!MDBX_AVOID_MSYNC || !(txn->flags & MDBX_WRITEMAP))
+        page_shadow_release(txn->env, dl->items[r].ptr, npages);
+      ++r;
+    next_i:
+      i += step;
+      if (unlikely(i == end)) {
+        while (r <= dl->length)
+          dl->items[w++] = dl->items[r++];
+      } else {
+        while (r <= dl->length) {
+          assert(i != end);
+          tASSERT(txn, !spilled || (pl[i] & 1) == 0);
+          pl_pgno = pl[i] >> spilled;
+          dp_pgno = dl->items[r].pgno;
+          if (dp_pgno < pl_pgno)
+            dl->items[w++] = dl->items[r++];
+          else if (dp_pgno > pl_pgno)
+            goto next_i;
+          else
+            goto remove_dl;
+        }
+      }
+      dl->sorted = dpl_setlen(dl, w - 1);
+      txn->wr.dirtyroom += r - w;
+      tASSERT(txn, txn->wr.dirtyroom + txn->wr.dirtylist->length ==
+                       (txn->parent ? txn->parent->wr.dirtyroom : txn->env->options.dp_limit));
+      return;
+    }
+  }
+}
+
+void dpl_release_shadows(MDBX_txn *txn) {
+  tASSERT(txn, (txn->flags & (MDBX_TXN_RDONLY | MDBX_WRITEMAP)) == 0);
+  MDBX_env *env = txn->env;
+  dpl_t *const dl = txn->wr.dirtylist;
+
+  for (size_t i = 1; i <= dl->length; i++)
+    page_shadow_release(env, dl->items[i].ptr, dpl_npages(dl, i));
+
+  dpl_clear(dl);
+}

src/dpl.h

@@ -0,0 +1,134 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#pragma once
+
+#include "essentials.h"
+
+static inline size_t dpl_setlen(dpl_t *dl, size_t len) {
+  static const page_t dpl_stub_pageE = {INVALID_TXNID,
+                                        0,
+                                        P_BAD,
+                                        {0},
+                                        /* pgno */ ~(pgno_t)0};
+  assert(dpl_stub_pageE.flags == P_BAD && dpl_stub_pageE.pgno == P_INVALID);
+  dl->length = len;
+  dl->items[len + 1].ptr = (page_t *)&dpl_stub_pageE;
+  dl->items[len + 1].pgno = P_INVALID;
+  dl->items[len + 1].npages = 1;
+  return len;
+}
+
+static inline void dpl_clear(dpl_t *dl) {
+  static const page_t dpl_stub_pageB = {INVALID_TXNID,
+                                        0,
+                                        P_BAD,
+                                        {0},
+                                        /* pgno */ 0};
+  assert(dpl_stub_pageB.flags == P_BAD && dpl_stub_pageB.pgno == 0);
+  dl->sorted = dpl_setlen(dl, 0);
+  dl->pages_including_loose = 0;
+  dl->items[0].ptr = (page_t *)&dpl_stub_pageB;
+  dl->items[0].pgno = 0;
+  dl->items[0].npages = 1;
+  assert(dl->items[0].pgno == 0 && dl->items[dl->length + 1].pgno == P_INVALID);
+}
+
+MDBX_INTERNAL int __must_check_result dpl_alloc(MDBX_txn *txn);
+
+MDBX_INTERNAL void dpl_free(MDBX_txn *txn);
+
+MDBX_INTERNAL dpl_t *dpl_reserve(MDBX_txn *txn, size_t size);
+
+MDBX_INTERNAL __noinline dpl_t *dpl_sort_slowpath(const MDBX_txn *txn);
+
+static inline dpl_t *dpl_sort(const MDBX_txn *txn) {
+  tASSERT(txn, (txn->flags & MDBX_TXN_RDONLY) == 0);
+  tASSERT(txn, (txn->flags & MDBX_WRITEMAP) == 0 || MDBX_AVOID_MSYNC);
+
+  dpl_t *dl = txn->wr.dirtylist;
+  tASSERT(txn, dl->length <= PAGELIST_LIMIT);
+  tASSERT(txn, dl->sorted <= dl->length);
+  tASSERT(txn, dl->items[0].pgno == 0 && dl->items[dl->length + 1].pgno == P_INVALID);
+  return likely(dl->sorted == dl->length) ? dl : dpl_sort_slowpath(txn);
+}
+
+MDBX_NOTHROW_PURE_FUNCTION MDBX_INTERNAL __noinline size_t dpl_search(const MDBX_txn *txn, pgno_t pgno);
+
+MDBX_MAYBE_UNUSED MDBX_INTERNAL const page_t *debug_dpl_find(const MDBX_txn *txn, const pgno_t pgno);
+
+MDBX_NOTHROW_PURE_FUNCTION static inline unsigned dpl_npages(const dpl_t *dl, size_t i) {
+  assert(0 <= (intptr_t)i && i <= dl->length);
+  unsigned n = dl->items[i].npages;
+  assert(n == (is_largepage(dl->items[i].ptr) ? dl->items[i].ptr->pages : 1));
+  return n;
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline pgno_t dpl_endpgno(const dpl_t *dl, size_t i) {
+  return dpl_npages(dl, i) + dl->items[i].pgno;
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline bool dpl_intersect(const MDBX_txn *txn, pgno_t pgno, size_t npages) {
+  tASSERT(txn, (txn->flags & MDBX_TXN_RDONLY) == 0);
+  tASSERT(txn, (txn->flags & MDBX_WRITEMAP) == 0 || MDBX_AVOID_MSYNC);
+
+  dpl_t *dl = txn->wr.dirtylist;
+  tASSERT(txn, dl->sorted == dl->length);
+  tASSERT(txn, dl->items[0].pgno == 0 && dl->items[dl->length + 1].pgno == P_INVALID);
+  size_t const n = dpl_search(txn, pgno);
+  tASSERT(txn, n >= 1 && n <= dl->length + 1);
+  tASSERT(txn, pgno <= dl->items[n].pgno);
+  tASSERT(txn, pgno > dl->items[n - 1].pgno);
+  const bool rc =
+      /* intersection with founded */ pgno + npages > dl->items[n].pgno ||
+      /* intersection with prev */ dpl_endpgno(dl, n - 1) > pgno;
+  if (ASSERT_ENABLED()) {
+    bool check = false;
+    for (size_t i = 1; i <= dl->length; ++i) {
+      const page_t *const dp = dl->items[i].ptr;
+      if (!(dp->pgno /* begin */ >= /* end */ pgno + npages || dpl_endpgno(dl, i) /* end */ <= /* begin */ pgno))
+        check |= true;
+    }
+    tASSERT(txn, check == rc);
+  }
+  return rc;
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline size_t dpl_exist(const MDBX_txn *txn, pgno_t pgno) {
+  tASSERT(txn, (txn->flags & MDBX_WRITEMAP) == 0 || MDBX_AVOID_MSYNC);
+  dpl_t *dl = txn->wr.dirtylist;
+  size_t i = dpl_search(txn, pgno);
+  tASSERT(txn, (int)i > 0);
+  return (dl->items[i].pgno == pgno) ? i : 0;
+}
+
+MDBX_INTERNAL void dpl_remove_ex(const MDBX_txn *txn, size_t i, size_t npages);
+
+static inline void dpl_remove(const MDBX_txn *txn, size_t i) {
+  dpl_remove_ex(txn, i, dpl_npages(txn->wr.dirtylist, i));
+}
+
+MDBX_INTERNAL int __must_check_result dpl_append(MDBX_txn *txn, pgno_t pgno, page_t *page, size_t npages);
+
+MDBX_MAYBE_UNUSED MDBX_INTERNAL bool dpl_check(MDBX_txn *txn);
+
+MDBX_NOTHROW_PURE_FUNCTION static inline uint32_t dpl_age(const MDBX_txn *txn, size_t i) {
+  tASSERT(txn, (txn->flags & (MDBX_TXN_RDONLY | MDBX_WRITEMAP)) == 0);
+  const dpl_t *dl = txn->wr.dirtylist;
+  assert((intptr_t)i > 0 && i <= dl->length);
+  size_t *const ptr = ptr_disp(dl->items[i].ptr, -(ptrdiff_t)sizeof(size_t));
+  return txn->wr.dirtylru - (uint32_t)*ptr;
+}
+
+MDBX_INTERNAL void dpl_lru_reduce(MDBX_txn *txn);
+
+static inline uint32_t dpl_lru_turn(MDBX_txn *txn) {
+  txn->wr.dirtylru += 1;
+  if (unlikely(txn->wr.dirtylru > UINT32_MAX / 3) && (txn->flags & MDBX_WRITEMAP) == 0)
+    dpl_lru_reduce(txn);
+  return txn->wr.dirtylru;
+}
+
+MDBX_INTERNAL void dpl_sift(MDBX_txn *const txn, pnl_t pl, const bool spilled);
+
+MDBX_INTERNAL void dpl_release_shadows(MDBX_txn *txn);

src/env.c

@@ -0,0 +1,602 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+MDBX_txn *env_owned_wrtxn(const MDBX_env *env) {
+  if (likely(env->basal_txn)) {
+    const bool is_owned = (env->flags & MDBX_NOSTICKYTHREADS) ? (env->basal_txn->owner != 0)
+                                                              : (env->basal_txn->owner == osal_thread_self());
+    if (is_owned)
+      return env->txn ? env->txn : env->basal_txn;
+  }
+  return nullptr;
+}
+
+int env_page_auxbuffer(MDBX_env *env) {
+  const int err = env->page_auxbuf
+                      ? MDBX_SUCCESS
+                      : osal_memalign_alloc(globals.sys_pagesize, env->ps * (size_t)NUM_METAS, &env->page_auxbuf);
+  if (likely(err == MDBX_SUCCESS)) {
+    memset(env->page_auxbuf, -1, env->ps * (size_t)2);
+    memset(ptr_disp(env->page_auxbuf, env->ps * (size_t)2), 0, env->ps);
+  }
+  return err;
+}
+
+__cold unsigned env_setup_pagesize(MDBX_env *env, const size_t pagesize) {
+  STATIC_ASSERT(PTRDIFF_MAX > MAX_MAPSIZE);
+  STATIC_ASSERT(MDBX_MIN_PAGESIZE > sizeof(page_t) + sizeof(meta_t));
+  ENSURE(env, is_powerof2(pagesize));
+  ENSURE(env, pagesize >= MDBX_MIN_PAGESIZE);
+  ENSURE(env, pagesize <= MDBX_MAX_PAGESIZE);
+  ENSURE(env, !env->page_auxbuf && env->ps != pagesize);
+  env->ps = (unsigned)pagesize;
+
+  STATIC_ASSERT(MAX_GC1OVPAGE(MDBX_MIN_PAGESIZE) > 4);
+  STATIC_ASSERT(MAX_GC1OVPAGE(MDBX_MAX_PAGESIZE) < PAGELIST_LIMIT);
+  const intptr_t maxgc_ov1page = (pagesize - PAGEHDRSZ) / sizeof(pgno_t) - 1;
+  ENSURE(env, maxgc_ov1page > 42 && maxgc_ov1page < (intptr_t)PAGELIST_LIMIT / 4);
+  env->maxgc_large1page = (unsigned)maxgc_ov1page;
+  env->maxgc_per_branch = (unsigned)((pagesize - PAGEHDRSZ) / (sizeof(indx_t) + sizeof(node_t) + sizeof(txnid_t)));
+
+  STATIC_ASSERT(LEAF_NODE_MAX(MDBX_MIN_PAGESIZE) > sizeof(tree_t) + NODESIZE + 42);
+  STATIC_ASSERT(LEAF_NODE_MAX(MDBX_MAX_PAGESIZE) < UINT16_MAX);
+  STATIC_ASSERT(LEAF_NODE_MAX(MDBX_MIN_PAGESIZE) >= BRANCH_NODE_MAX(MDBX_MIN_PAGESIZE));
+  STATIC_ASSERT(BRANCH_NODE_MAX(MDBX_MAX_PAGESIZE) > NODESIZE + 42);
+  STATIC_ASSERT(BRANCH_NODE_MAX(MDBX_MAX_PAGESIZE) < UINT16_MAX);
+  const intptr_t branch_nodemax = BRANCH_NODE_MAX(pagesize);
+  const intptr_t leaf_nodemax = LEAF_NODE_MAX(pagesize);
+  ENSURE(env, branch_nodemax > (intptr_t)(NODESIZE + 42) && branch_nodemax % 2 == 0 &&
+                  leaf_nodemax > (intptr_t)(sizeof(tree_t) + NODESIZE + 42) && leaf_nodemax >= branch_nodemax &&
+                  leaf_nodemax < (int)UINT16_MAX && leaf_nodemax % 2 == 0);
+  env->leaf_nodemax = (uint16_t)leaf_nodemax;
+  env->branch_nodemax = (uint16_t)branch_nodemax;
+  env->ps2ln = (uint8_t)log2n_powerof2(pagesize);
+  eASSERT(env, pgno2bytes(env, 1) == pagesize);
+  eASSERT(env, bytes2pgno(env, pagesize + pagesize) == 2);
+  recalculate_merge_thresholds(env);
+  recalculate_subpage_thresholds(env);
+  env_options_adjust_dp_limit(env);
+  return env->ps;
+}
+
+__cold int env_sync(MDBX_env *env, bool force, bool nonblock) {
+  if (unlikely(env->flags & MDBX_RDONLY))
+    return MDBX_EACCESS;
+
+  MDBX_txn *const txn_owned = env_owned_wrtxn(env);
+  bool should_unlock = false;
+  int rc = MDBX_RESULT_TRUE /* means "nothing to sync" */;
+
+retry:;
+  unsigned flags = env->flags & ~(MDBX_NOMETASYNC | txn_shrink_allowed);
+  if (unlikely((flags & (ENV_FATAL_ERROR | ENV_ACTIVE)) != ENV_ACTIVE)) {
+    rc = (flags & ENV_FATAL_ERROR) ? MDBX_PANIC : MDBX_EPERM;
+    goto bailout;
+  }
+
+  const troika_t troika = (txn_owned || should_unlock) ? env->basal_txn->wr.troika : meta_tap(env);
+  const meta_ptr_t head = meta_recent(env, &troika);
+  const uint64_t unsynced_pages = atomic_load64(&env->lck->unsynced_pages, mo_Relaxed);
+  if (unsynced_pages == 0) {
+    const uint32_t synched_meta_txnid_u32 = atomic_load32(&env->lck->meta_sync_txnid, mo_Relaxed);
+    if (synched_meta_txnid_u32 == (uint32_t)head.txnid && head.is_steady)
+      goto bailout;
+  }
+
+  if (should_unlock && (env->flags & MDBX_WRITEMAP) &&
+      unlikely(head.ptr_c->geometry.first_unallocated > bytes2pgno(env, env->dxb_mmap.current))) {
+
+    if (unlikely(env->stuck_meta >= 0) && troika.recent != (uint8_t)env->stuck_meta) {
+      NOTICE("skip %s since wagering meta-page (%u) is mispatch the recent "
+             "meta-page (%u)",
+             "sync datafile", env->stuck_meta, troika.recent);
+      rc = MDBX_RESULT_TRUE;
+    } else {
+      rc = dxb_resize(env, head.ptr_c->geometry.first_unallocated, head.ptr_c->geometry.now, head.ptr_c->geometry.upper,
+                      implicit_grow);
+      if (unlikely(rc != MDBX_SUCCESS))
+        goto bailout;
+    }
+  }
+
+  const size_t autosync_threshold = atomic_load32(&env->lck->autosync_threshold, mo_Relaxed);
+  const uint64_t autosync_period = atomic_load64(&env->lck->autosync_period, mo_Relaxed);
+  uint64_t eoos_timestamp;
+  if (force || (autosync_threshold && unsynced_pages >= autosync_threshold) ||
+      (autosync_period && (eoos_timestamp = atomic_load64(&env->lck->eoos_timestamp, mo_Relaxed)) &&
+       osal_monotime() - eoos_timestamp >= autosync_period))
+    flags &= MDBX_WRITEMAP /* clear flags for full steady sync */;
+
+  if (!txn_owned) {
+    if (!should_unlock) {
+#if MDBX_ENABLE_PGOP_STAT
+      unsigned wops = 0;
+#endif /* MDBX_ENABLE_PGOP_STAT */
+
+      int err;
+      /* pre-sync to avoid latency for writer */
+      if (unsynced_pages > /* FIXME: define threshold */ 42 && (flags & MDBX_SAFE_NOSYNC) == 0) {
+        eASSERT(env, ((flags ^ env->flags) & MDBX_WRITEMAP) == 0);
+        if (flags & MDBX_WRITEMAP) {
+          /* Acquire guard to avoid collision with remap */
+#if defined(_WIN32) || defined(_WIN64)
+          imports.srwl_AcquireShared(&env->remap_guard);
+#else
+          err = osal_fastmutex_acquire(&env->remap_guard);
+          if (unlikely(err != MDBX_SUCCESS))
+            return err;
+#endif
+          const size_t usedbytes = pgno_align2os_bytes(env, head.ptr_c->geometry.first_unallocated);
+          err = osal_msync(&env->dxb_mmap, 0, usedbytes, MDBX_SYNC_DATA);
+#if defined(_WIN32) || defined(_WIN64)
+          imports.srwl_ReleaseShared(&env->remap_guard);
+#else
+          int unlock_err = osal_fastmutex_release(&env->remap_guard);
+          if (unlikely(unlock_err != MDBX_SUCCESS) && err == MDBX_SUCCESS)
+            err = unlock_err;
+#endif
+        } else
+          err = osal_fsync(env->lazy_fd, MDBX_SYNC_DATA);
+
+        if (unlikely(err != MDBX_SUCCESS))
+          return err;
+
+#if MDBX_ENABLE_PGOP_STAT
+        wops = 1;
+#endif /* MDBX_ENABLE_PGOP_STAT */
+        /* pre-sync done */
+        rc = MDBX_SUCCESS /* means "some data was synced" */;
+      }
+
+      err = lck_txn_lock(env, nonblock);
+      if (unlikely(err != MDBX_SUCCESS))
+        return err;
+
+      should_unlock = true;
+#if MDBX_ENABLE_PGOP_STAT
+      env->lck->pgops.wops.weak += wops;
+#endif /* MDBX_ENABLE_PGOP_STAT */
+      env->basal_txn->wr.troika = meta_tap(env);
+      eASSERT(env, !env->txn && !env->basal_txn->nested);
+      goto retry;
+    }
+    eASSERT(env, head.txnid == recent_committed_txnid(env));
+    env->basal_txn->txnid = head.txnid;
+    txn_gc_detent(env->basal_txn);
+    flags |= txn_shrink_allowed;
+  }
+
+  eASSERT(env, txn_owned || should_unlock);
+  eASSERT(env, !txn_owned || (flags & txn_shrink_allowed) == 0);
+
+  if (!head.is_steady && unlikely(env->stuck_meta >= 0) && troika.recent != (uint8_t)env->stuck_meta) {
+    NOTICE("skip %s since wagering meta-page (%u) is mispatch the recent "
+           "meta-page (%u)",
+           "sync datafile", env->stuck_meta, troika.recent);
+    rc = MDBX_RESULT_TRUE;
+    goto bailout;
+  }
+  if (!head.is_steady || ((flags & MDBX_SAFE_NOSYNC) == 0 && unsynced_pages)) {
+    DEBUG("meta-head %" PRIaPGNO ", %s, sync_pending %" PRIu64, data_page(head.ptr_c)->pgno,
+          durable_caption(head.ptr_c), unsynced_pages);
+    meta_t meta = *head.ptr_c;
+    rc = dxb_sync_locked(env, flags, &meta, &env->basal_txn->wr.troika);
+    if (unlikely(rc != MDBX_SUCCESS))
+      goto bailout;
+  }
+
+  /* LY: sync meta-pages if MDBX_NOMETASYNC enabled
+   *     and someone was not synced above. */
+  if (atomic_load32(&env->lck->meta_sync_txnid, mo_Relaxed) != (uint32_t)head.txnid)
+    rc = meta_sync(env, head);
+
+bailout:
+  if (should_unlock)
+    lck_txn_unlock(env);
+  return rc;
+}
+
+__cold int env_open(MDBX_env *env, mdbx_mode_t mode) {
+  /* Использование O_DSYNC или FILE_FLAG_WRITE_THROUGH:
+   *
+   *   0) Если размер страниц БД меньше системной страницы ОЗУ, то ядру ОС
+   *      придется чаще обновлять страницы в unified page cache.
+   *
+   *      Однако, O_DSYNC не предполагает отключение unified page cache,
+   *      поэтому подобные затруднения будем считать проблемой ОС и/или
+   *      ожидаемым пенальти из-за использования мелких страниц БД.
+   *
+   *   1) В режиме MDBX_SYNC_DURABLE - O_DSYNC для записи как данных,
+   *      так и мета-страниц. Однако, на Linux отказ от O_DSYNC с последующим
+   *      fdatasync() может быть выгоднее при использовании HDD, так как
+   *      позволяет io-scheduler переупорядочить запись с учетом актуального
+   *      расположения файла БД на носителе.
+   *
+   *   2) В режиме MDBX_NOMETASYNC - O_DSYNC можно использовать для данных,
+   *      но в этом может не быть смысла, так как fdatasync() всё равно
+   *      требуется для гарантии фиксации мета после предыдущей транзакции.
+   *
+   *      В итоге на нормальных системах (не Windows) есть два варианта:
+   *       - при возможности O_DIRECT и/или io_ring для данных, скорее всего,
+   *         есть смысл вызвать fdatasync() перед записью данных, а затем
+   *         использовать O_DSYNC;
+   *       - не использовать O_DSYNC и вызывать fdatasync() после записи данных.
+   *
+   *      На Windows же следует минимизировать использование FlushFileBuffers()
+   *      из-за проблем с производительностью. Поэтому на Windows в режиме
+   *      MDBX_NOMETASYNC:
+   *       - мета обновляется через дескриптор без FILE_FLAG_WRITE_THROUGH;
+   *       - перед началом записи данных вызывается FlushFileBuffers(), если
+   *         meta_sync_txnid не совпадает с последней записанной мета;
+   *       - данные записываются через дескриптор с FILE_FLAG_WRITE_THROUGH.
+   *
+   *   3) В режиме MDBX_SAFE_NOSYNC - O_DSYNC нет смысла использовать, пока не
+   *      будет реализована возможность полностью асинхронной "догоняющей"
+   *      записи в выделенном процессе-сервере с io-ring очередями внутри.
+   *
+   * -----
+   *
+   * Использование O_DIRECT или FILE_FLAG_NO_BUFFERING:
+   *
+   *   Назначение этих флагов в отключении файлового дескриптора от
+   *   unified page cache, т.е. от отображенных в память данных в случае
+   *   libmdbx.
+   *
+   *   Поэтому, использование direct i/o в libmdbx без MDBX_WRITEMAP лишено
+   *   смысла и контр-продуктивно, ибо так мы провоцируем ядро ОС на
+   *   не-когерентность отображения в память с содержимым файла на носителе,
+   *   либо требуем дополнительных проверок и действий направленных на
+   *   фактическое отключение O_DIRECT для отображенных в память данных.
+   *
+   *   В режиме MDBX_WRITEMAP когерентность отображенных данных обеспечивается
+   *   физически. Поэтому использование direct i/o может иметь смысл, если у
+   *   ядра ОС есть какие-то проблемы с msync(), в том числе с
+   *   производительностью:
+   *    - использование io_ring или gather-write может быть дешевле, чем
+   *      просмотр PTE ядром и запись измененных/грязных;
+   *    - но проблема в том, что записываемые из user mode страницы либо не
+   *      будут помечены чистыми (и соответственно будут записаны ядром
+   *      еще раз), либо ядру необходимо искать и чистить PTE при получении
+   *      запроса на запись.
+   *
+   *   Поэтому O_DIRECT или FILE_FLAG_NO_BUFFERING используется:
+   *    - только в режиме MDBX_SYNC_DURABLE с MDBX_WRITEMAP;
+   *    - когда ps >= me_os_psize;
+   *    - опция сборки MDBX_AVOID_MSYNC != 0, которая по-умолчанию включена
+   *      только на Windows (см ниже).
+   *
+   * -----
+   *
+   * Использование FILE_FLAG_OVERLAPPED на Windows:
+   *
+   * У Windows очень плохо с I/O (за исключением прямых постраничных
+   * scatter/gather, которые работают в обход проблемного unified page
+   * cache и поэтому почти бесполезны в libmdbx).
+   *
+   * При этом всё еще хуже при использовании FlushFileBuffers(), что также
+   * требуется после FlushViewOfFile() в режиме MDBX_WRITEMAP. Поэтому
+   * на Windows вместо FlushViewOfFile() и FlushFileBuffers() следует
+   * использовать запись через дескриптор с FILE_FLAG_WRITE_THROUGH.
+   *
+   * В свою очередь, запись с FILE_FLAG_WRITE_THROUGH дешевле/быстрее
+   * при использовании FILE_FLAG_OVERLAPPED. В результате, на Windows
+   * в durable-режимах запись данных всегда в overlapped-режиме,
+   * при этом для записи мета требуется отдельный не-overlapped дескриптор.
+   */
+
+  env->pid = osal_getpid();
+  int rc = osal_openfile((env->flags & MDBX_RDONLY) ? MDBX_OPEN_DXB_READ : MDBX_OPEN_DXB_LAZY, env, env->pathname.dxb,
+                         &env->lazy_fd, mode);
+  if (unlikely(rc != MDBX_SUCCESS))
+    return rc;
+
+#if MDBX_LOCKING == MDBX_LOCKING_SYSV
+  env->me_sysv_ipc.key = ftok(env->pathname.dxb, 42);
+  if (unlikely(env->me_sysv_ipc.key == -1))
+    return errno;
+#endif /* MDBX_LOCKING */
+
+  /* Set the position in files outside of the data to avoid corruption
+   * due to erroneous use of file descriptors in the application code. */
+  const uint64_t safe_parking_lot_offset = UINT64_C(0x7fffFFFF80000000);
+  osal_fseek(env->lazy_fd, safe_parking_lot_offset);
+
+  env->fd4meta = env->lazy_fd;
+#if defined(_WIN32) || defined(_WIN64)
+  eASSERT(env, env->ioring.overlapped_fd == 0);
+  bool ior_direct = false;
+  if (!(env->flags & (MDBX_RDONLY | MDBX_SAFE_NOSYNC | MDBX_NOMETASYNC | MDBX_EXCLUSIVE))) {
+    if (MDBX_AVOID_MSYNC && (env->flags & MDBX_WRITEMAP)) {
+      /* Запрошен режим MDBX_SYNC_DURABLE | MDBX_WRITEMAP при активной опции
+       * MDBX_AVOID_MSYNC.
+       *
+       * 1) В этой комбинации наиболее выгодно использовать WriteFileGather(),
+       * но для этого необходимо открыть файл с флагом FILE_FLAG_NO_BUFFERING и
+       * после обеспечивать выравнивание адресов и размера данных на границу
+       * системной страницы, что в свою очередь возможно если размер страницы БД
+       * не меньше размера системной страницы ОЗУ. Поэтому для открытия файла в
+       * нужном режиме требуется знать размер страницы БД.
+       *
+       * 2) Кроме этого, в Windows запись в заблокированный регион файла
+       * возможно только через тот-же дескриптор. Поэтому изначальный захват
+       * блокировок посредством lck_seize(), захват/освобождение блокировок
+       * во время пишущих транзакций и запись данных должны выполнятся через
+       * один дескриптор.
+       *
+       * Таким образом, требуется прочитать волатильный заголовок БД, чтобы
+       * узнать размер страницы, чтобы открыть дескриптор файла в режиме нужном
+       * для записи данных, чтобы использовать именно этот дескриптор для
+       * изначального захвата блокировок. */
+      meta_t header;
+      uint64_t dxb_filesize;
+      int err = dxb_read_header(env, &header, MDBX_SUCCESS, true);
+      if ((err == MDBX_SUCCESS && header.pagesize >= globals.sys_pagesize) ||
+          (err == MDBX_ENODATA && mode && env->ps >= globals.sys_pagesize &&
+           osal_filesize(env->lazy_fd, &dxb_filesize) == MDBX_SUCCESS && dxb_filesize == 0))
+        /* Может быть коллизия, если два процесса пытаются одновременно создать
+         * БД с разным размером страницы, который у одного меньше системной
+         * страницы, а у другого НЕ меньше. Эта допустимая, но очень странная
+         * ситуация. Поэтому считаем её ошибочной и не пытаемся разрешить. */
+        ior_direct = true;
+    }
+
+    rc = osal_openfile(ior_direct ? MDBX_OPEN_DXB_OVERLAPPED_DIRECT : MDBX_OPEN_DXB_OVERLAPPED, env, env->pathname.dxb,
+                       &env->ioring.overlapped_fd, 0);
+    if (unlikely(rc != MDBX_SUCCESS))
+      return rc;
+    env->dxb_lock_event = CreateEventW(nullptr, true, false, nullptr);
+    if (unlikely(!env->dxb_lock_event))
+      return (int)GetLastError();
+    osal_fseek(env->ioring.overlapped_fd, safe_parking_lot_offset);
+  }
+#else
+  if (mode == 0) {
+    /* pickup mode for lck-file */
+    struct stat st;
+    if (unlikely(fstat(env->lazy_fd, &st)))
+      return errno;
+    mode = st.st_mode;
+  }
+  mode = (/* inherit read permissions for group and others */ mode & (S_IRGRP | S_IWGRP | S_IROTH | S_IWOTH)) |
+         /* always add read/write for owner */ S_IRUSR | S_IWUSR |
+         ((mode & S_IRGRP) ? /* +write if readable by group */ S_IWGRP : 0) |
+         ((mode & S_IROTH) ? /* +write if readable by others */ S_IWOTH : 0);
+#endif /* !Windows */
+  const int lck_rc = lck_setup(env, mode);
+  if (unlikely(MDBX_IS_ERROR(lck_rc)))
+    return lck_rc;
+  if (env->lck_mmap.fd != INVALID_HANDLE_VALUE)
+    osal_fseek(env->lck_mmap.fd, safe_parking_lot_offset);
+
+  eASSERT(env, env->dsync_fd == INVALID_HANDLE_VALUE);
+  if (!(env->flags & (MDBX_RDONLY | MDBX_SAFE_NOSYNC | DEPRECATED_MAPASYNC
+#if defined(_WIN32) || defined(_WIN64)
+                      | MDBX_EXCLUSIVE
+#endif /* !Windows */
+                      ))) {
+    rc = osal_openfile(MDBX_OPEN_DXB_DSYNC, env, env->pathname.dxb, &env->dsync_fd, 0);
+    if (unlikely(MDBX_IS_ERROR(rc)))
+      return rc;
+    if (env->dsync_fd != INVALID_HANDLE_VALUE) {
+      if ((env->flags & MDBX_NOMETASYNC) == 0)
+        env->fd4meta = env->dsync_fd;
+      osal_fseek(env->dsync_fd, safe_parking_lot_offset);
+    }
+  }
+
+  const MDBX_env_flags_t lazy_flags = MDBX_SAFE_NOSYNC | MDBX_UTTERLY_NOSYNC | MDBX_NOMETASYNC;
+  const MDBX_env_flags_t mode_flags = lazy_flags | MDBX_LIFORECLAIM | MDBX_NORDAHEAD | MDBX_RDONLY | MDBX_WRITEMAP;
+
+  lck_t *const lck = env->lck_mmap.lck;
+  if (lck && lck_rc != MDBX_RESULT_TRUE && (env->flags & MDBX_RDONLY) == 0) {
+    MDBX_env_flags_t snap_flags;
+    while ((snap_flags = atomic_load32(&lck->envmode, mo_AcquireRelease)) == MDBX_RDONLY) {
+      if (atomic_cas32(&lck->envmode, MDBX_RDONLY, (snap_flags = (env->flags & mode_flags)))) {
+        /* The case:
+         *  - let's assume that for some reason the DB file is smaller
+         *    than it should be according to the geometry,
+         *    but not smaller than the last page used;
+         *  - the first process that opens the database (lck_rc == RESULT_TRUE)
+         *    does this in readonly mode and therefore cannot bring
+         *    the file size back to normal;
+         *  - some next process (lck_rc != RESULT_TRUE) opens the DB in
+         *    read-write mode and now is here.
+         *
+         * FIXME: Should we re-check and set the size of DB-file right here? */
+        break;
+      }
+      atomic_yield();
+    }
+
+    if (env->flags & MDBX_ACCEDE) {
+      /* Pickup current mode-flags (MDBX_LIFORECLAIM, MDBX_NORDAHEAD, etc). */
+      const MDBX_env_flags_t diff =
+          (snap_flags ^ env->flags) & ((snap_flags & lazy_flags) ? mode_flags : mode_flags & ~MDBX_WRITEMAP);
+      env->flags ^= diff;
+      NOTICE("accede mode-flags: 0x%X, 0x%X -> 0x%X", diff, env->flags ^ diff, env->flags);
+    }
+
+    /* Ранее упущенный не очевидный момент: При работе БД в режимах
+     * не-синхронной/отложенной фиксации на диске, все процессы-писатели должны
+     * иметь одинаковый режим MDBX_WRITEMAP.
+     *
+     * В противном случае, сброс на диск следует выполнять дважды: сначала
+     * msync(), затем fdatasync(). При этом msync() не обязан отрабатывать
+     * в процессах без MDBX_WRITEMAP, так как файл в память отображен только
+     * для чтения. Поэтому, в общем случае, различия по MDBX_WRITEMAP не
+     * позволяют выполнить фиксацию данных на диск, после их изменения в другом
+     * процессе.
+     *
+     * В режиме MDBX_UTTERLY_NOSYNC позволять совместную работу с MDBX_WRITEMAP
+     * также не следует, поскольку никакой процесс (в том числе последний) не
+     * может гарантированно сбросить данные на диск, а следовательно не должен
+     * помечать какую-либо транзакцию как steady.
+     *
+     * В результате, требуется либо запретить совместную работу процессам с
+     * разным MDBX_WRITEMAP в режиме отложенной записи, либо отслеживать такое
+     * смешивание и блокировать steady-пометки - что контрпродуктивно. */
+    const MDBX_env_flags_t rigorous_flags = (snap_flags & lazy_flags)
+                                                ? MDBX_SAFE_NOSYNC | MDBX_UTTERLY_NOSYNC | MDBX_WRITEMAP
+                                                : MDBX_SAFE_NOSYNC | MDBX_UTTERLY_NOSYNC;
+    const MDBX_env_flags_t rigorous_diff = (snap_flags ^ env->flags) & rigorous_flags;
+    if (rigorous_diff) {
+      ERROR("current mode/flags 0x%X incompatible with requested 0x%X, "
+            "rigorous diff 0x%X",
+            env->flags, snap_flags, rigorous_diff);
+      return MDBX_INCOMPATIBLE;
+    }
+  }
+
+  mincore_clean_cache(env);
+  const int dxb_rc = dxb_setup(env, lck_rc, mode);
+  if (MDBX_IS_ERROR(dxb_rc))
+    return dxb_rc;
+
+  rc = osal_check_fs_incore(env->lazy_fd);
+  env->incore = false;
+  if (rc == MDBX_RESULT_TRUE) {
+    env->incore = true;
+    NOTICE("%s", "in-core database");
+    rc = MDBX_SUCCESS;
+  } else if (unlikely(rc != MDBX_SUCCESS)) {
+    ERROR("check_fs_incore(), err %d", rc);
+    return rc;
+  }
+
+  if (unlikely(/* recovery mode */ env->stuck_meta >= 0) &&
+      (lck_rc != /* exclusive */ MDBX_RESULT_TRUE || (env->flags & MDBX_EXCLUSIVE) == 0)) {
+    ERROR("%s", "recovery requires exclusive mode");
+    return MDBX_BUSY;
+  }
+
+  DEBUG("opened dbenv %p", (void *)env);
+  env->flags |= ENV_ACTIVE;
+  if (!lck || lck_rc == MDBX_RESULT_TRUE) {
+    env->lck->envmode.weak = env->flags & mode_flags;
+    env->lck->meta_sync_txnid.weak = (uint32_t)recent_committed_txnid(env);
+    env->lck->readers_check_timestamp.weak = osal_monotime();
+  }
+  if (lck) {
+    if (lck_rc == MDBX_RESULT_TRUE) {
+      rc = lck_downgrade(env);
+      DEBUG("lck-downgrade-%s: rc %i", (env->flags & MDBX_EXCLUSIVE) ? "partial" : "full", rc);
+      if (rc != MDBX_SUCCESS)
+        return rc;
+    } else {
+      rc = mvcc_cleanup_dead(env, false, nullptr);
+      if (MDBX_IS_ERROR(rc))
+        return rc;
+    }
+  }
+
+  rc = (env->flags & MDBX_RDONLY) ? MDBX_SUCCESS
+                                  : osal_ioring_create(&env->ioring
+#if defined(_WIN32) || defined(_WIN64)
+                                                       ,
+                                                       ior_direct, env->ioring.overlapped_fd
+#endif /* Windows */
+                                    );
+  return rc;
+}
+
+__cold int env_close(MDBX_env *env, bool resurrect_after_fork) {
+  const unsigned flags = env->flags;
+  env->flags &= ~ENV_INTERNAL_FLAGS;
+  if (flags & ENV_TXKEY) {
+    thread_key_delete(env->me_txkey);
+    env->me_txkey = 0;
+  }
+
+  if (env->lck)
+    munlock_all(env);
+
+  rthc_lock();
+  int rc = rthc_remove(env);
+  rthc_unlock();
+
+#if MDBX_ENABLE_DBI_LOCKFREE
+  for (defer_free_item_t *next, *ptr = env->defer_free; ptr; ptr = next) {
+    next = ptr->next;
+    osal_free(ptr);
+  }
+  env->defer_free = nullptr;
+#endif /* MDBX_ENABLE_DBI_LOCKFREE */
+
+  if ((env->flags & MDBX_RDONLY) == 0)
+    osal_ioring_destroy(&env->ioring);
+
+  env->lck = nullptr;
+  if (env->lck_mmap.lck)
+    osal_munmap(&env->lck_mmap);
+
+  if (env->dxb_mmap.base) {
+    osal_munmap(&env->dxb_mmap);
+#ifdef ENABLE_MEMCHECK
+    VALGRIND_DISCARD(env->valgrind_handle);
+    env->valgrind_handle = -1;
+#endif /* ENABLE_MEMCHECK */
+  }
+
+#if defined(_WIN32) || defined(_WIN64)
+  eASSERT(env, !env->ioring.overlapped_fd || env->ioring.overlapped_fd == INVALID_HANDLE_VALUE);
+  if (env->dxb_lock_event != INVALID_HANDLE_VALUE) {
+    CloseHandle(env->dxb_lock_event);
+    env->dxb_lock_event = INVALID_HANDLE_VALUE;
+  }
+  eASSERT(env, !resurrect_after_fork);
+  if (env->pathname_char) {
+    osal_free(env->pathname_char);
+    env->pathname_char = nullptr;
+  }
+#endif /* Windows */
+
+  if (env->dsync_fd != INVALID_HANDLE_VALUE) {
+    (void)osal_closefile(env->dsync_fd);
+    env->dsync_fd = INVALID_HANDLE_VALUE;
+  }
+
+  if (env->lazy_fd != INVALID_HANDLE_VALUE) {
+    (void)osal_closefile(env->lazy_fd);
+    env->lazy_fd = INVALID_HANDLE_VALUE;
+  }
+
+  if (env->lck_mmap.fd != INVALID_HANDLE_VALUE) {
+    (void)osal_closefile(env->lck_mmap.fd);
+    env->lck_mmap.fd = INVALID_HANDLE_VALUE;
+  }
+
+  if (!resurrect_after_fork) {
+    if (env->kvs) {
+      for (size_t i = CORE_DBS; i < env->n_dbi; ++i)
+        if (env->kvs[i].name.iov_len)
+          osal_free(env->kvs[i].name.iov_base);
+      osal_free(env->kvs);
+      env->n_dbi = CORE_DBS;
+      env->kvs = nullptr;
+    }
+    if (env->page_auxbuf) {
+      osal_memalign_free(env->page_auxbuf);
+      env->page_auxbuf = nullptr;
+    }
+    if (env->dbi_seqs) {
+      osal_free(env->dbi_seqs);
+      env->dbi_seqs = nullptr;
+    }
+    if (env->dbs_flags) {
+      osal_free(env->dbs_flags);
+      env->dbs_flags = nullptr;
+    }
+    if (env->pathname.buffer) {
+      osal_free(env->pathname.buffer);
+      env->pathname.buffer = nullptr;
+    }
+    if (env->basal_txn) {
+      txn_basal_destroy(env->basal_txn);
+      env->basal_txn = nullptr;
+    }
+  }
+  env->stuck_meta = -1;
+  return rc;
+}

src/essentials.h

@@ -0,0 +1,125 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#pragma once
+
+#define LIBMDBX_INTERNALS
+#define MDBX_DEPRECATED
+
+#ifdef MDBX_CONFIG_H
+#include MDBX_CONFIG_H
+#endif
+
+#include "preface.h"
+
+#ifdef xMDBX_ALLOY
+/* Amalgamated build */
+#define MDBX_INTERNAL static
+#else
+/* Non-amalgamated build */
+#define MDBX_INTERNAL
+#endif /* xMDBX_ALLOY */
+
+#include "../mdbx.h"
+
+/*----------------------------------------------------------------------------*/
+/* Basic constants and types */
+
+typedef struct iov_ctx iov_ctx_t;
+#include "osal.h"
+
+#include "options.h"
+
+#include "atomics-types.h"
+
+#include "layout-dxb.h"
+#include "layout-lck.h"
+
+#define MIN_MAPSIZE (MDBX_MIN_PAGESIZE * MIN_PAGENO)
+#if defined(_WIN32) || defined(_WIN64)
+#define MAX_MAPSIZE32 UINT32_C(0x38000000)
+#else
+#define MAX_MAPSIZE32 UINT32_C(0x7f000000)
+#endif
+#define MAX_MAPSIZE64 ((MAX_PAGENO + 1) * (uint64_t)MDBX_MAX_PAGESIZE)
+
+#if MDBX_WORDBITS >= 64
+#define MAX_MAPSIZE MAX_MAPSIZE64
+#define PAGELIST_LIMIT ((size_t)MAX_PAGENO)
+#else
+#define MAX_MAPSIZE MAX_MAPSIZE32
+#define PAGELIST_LIMIT (MAX_MAPSIZE32 / MDBX_MIN_PAGESIZE)
+#endif /* MDBX_WORDBITS */
+
+#define MDBX_GOLD_RATIO_DBL 1.6180339887498948482
+#define MEGABYTE ((size_t)1 << 20)
+
+/*----------------------------------------------------------------------------*/
+
+union logger_union {
+  void *ptr;
+  MDBX_debug_func *fmt;
+  MDBX_debug_func_nofmt *nofmt;
+};
+
+struct libmdbx_globals {
+  bin128_t bootid;
+  unsigned sys_pagesize, sys_allocation_granularity;
+  uint8_t sys_pagesize_ln2;
+  uint8_t runtime_flags;
+  uint8_t loglevel;
+#if defined(_WIN32) || defined(_WIN64)
+  bool running_under_Wine;
+#elif defined(__linux__) || defined(__gnu_linux__)
+  bool running_on_WSL1 /* Windows Subsystem 1 for Linux */;
+  uint32_t linux_kernel_version;
+#endif /* Linux */
+  union logger_union logger;
+  osal_fastmutex_t debug_lock;
+  size_t logger_buffer_size;
+  char *logger_buffer;
+};
+
+#ifdef __cplusplus
+extern "C" {
+#endif /* __cplusplus */
+
+extern struct libmdbx_globals globals;
+#if defined(_WIN32) || defined(_WIN64)
+extern struct libmdbx_imports imports;
+#endif /* Windows */
+
+#include "logging_and_debug.h"
+
+#include "utils.h"
+
+#include "pnl.h"
+
+#ifdef __cplusplus
+}
+#endif /* __cplusplus */
+
+#define mdbx_sourcery_anchor XCONCAT(mdbx_sourcery_, MDBX_BUILD_SOURCERY)
+#if defined(xMDBX_TOOLS)
+extern LIBMDBX_API const char *const mdbx_sourcery_anchor;
+#endif
+
+#define MDBX_IS_ERROR(rc) ((rc) != MDBX_RESULT_TRUE && (rc) != MDBX_RESULT_FALSE)
+
+/*----------------------------------------------------------------------------*/
+
+MDBX_NOTHROW_CONST_FUNCTION MDBX_MAYBE_UNUSED static inline pgno_t int64pgno(int64_t i64) {
+  if (likely(i64 >= (int64_t)MIN_PAGENO && i64 <= (int64_t)MAX_PAGENO + 1))
+    return (pgno_t)i64;
+  return (i64 < (int64_t)MIN_PAGENO) ? MIN_PAGENO : MAX_PAGENO;
+}
+
+MDBX_NOTHROW_CONST_FUNCTION MDBX_MAYBE_UNUSED static inline pgno_t pgno_add(size_t base, size_t augend) {
+  assert(base <= MAX_PAGENO + 1 && augend < MAX_PAGENO);
+  return int64pgno((int64_t)base + (int64_t)augend);
+}
+
+MDBX_NOTHROW_CONST_FUNCTION MDBX_MAYBE_UNUSED static inline pgno_t pgno_sub(size_t base, size_t subtrahend) {
+  assert(base >= MIN_PAGENO && base <= MAX_PAGENO + 1 && subtrahend < MAX_PAGENO);
+  return int64pgno((int64_t)base - (int64_t)subtrahend);
+}

src/gc.h

@@ -0,0 +1,80 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#pragma once
+
+#include "essentials.h"
+
+/* Гистограмма решения нарезки фрагментов для ситуации нехватки идентификаторов/слотов. */
+typedef struct gc_dense_histogram {
+  /* Размер массива одновременно задаёт максимальный размер последовательностей,
+   * с которыми решается задача распределения.
+   *
+   * Использование длинных последовательностей контрпродуктивно, так как такие последовательности будут
+   * создавать/воспроизводить/повторять аналогичные затруднения при последующей переработке. Однако,
+   * в редких ситуациях это может быть единственным выходом. */
+  unsigned end;
+  pgno_t array[31];
+} gc_dense_histogram_t;
+
+typedef struct gc_update_context {
+  unsigned loop;
+  unsigned goodchunk;
+  bool dense;
+  pgno_t prev_first_unallocated;
+  size_t retired_stored;
+  size_t return_reserved_lo, return_reserved_hi;
+  txnid_t gc_first;
+  intptr_t return_left;
+#ifndef MDBX_DEBUG_GCU
+#define MDBX_DEBUG_GCU 0
+#endif
+#if MDBX_DEBUG_GCU
+  struct {
+    txnid_t prev;
+    unsigned n;
+  } dbg;
+#endif /* MDBX_DEBUG_GCU */
+  rkl_t ready4reuse, sequel;
+#if MDBX_ENABLE_BIGFOOT
+  txnid_t bigfoot;
+#endif /* MDBX_ENABLE_BIGFOOT */
+  union {
+    MDBX_cursor cursor;
+    cursor_couple_t couple;
+  };
+  gc_dense_histogram_t dense_histogram;
+} gcu_t;
+
+MDBX_INTERNAL int gc_put_init(MDBX_txn *txn, gcu_t *ctx);
+MDBX_INTERNAL void gc_put_destroy(gcu_t *ctx);
+
+#define ALLOC_DEFAULT 0     /* штатное/обычное выделение страниц */
+#define ALLOC_UNIMPORTANT 1 /* запрос неважен, невозможность выделения не приведет к ошибке транзакции */
+#define ALLOC_RESERVE 2     /* подготовка резерва для обновления GC, без аллокации */
+#define ALLOC_COALESCE 4    /* внутреннее состояние/флажок */
+#define ALLOC_SHOULD_SCAN 8 /* внутреннее состояние/флажок */
+#define ALLOC_LIFO 16       /* внутреннее состояние/флажок */
+
+MDBX_INTERNAL pgr_t gc_alloc_ex(const MDBX_cursor *const mc, const size_t num, uint8_t flags);
+
+MDBX_INTERNAL pgr_t gc_alloc_single(const MDBX_cursor *const mc);
+MDBX_INTERNAL int gc_update(MDBX_txn *txn, gcu_t *ctx);
+
+MDBX_NOTHROW_PURE_FUNCTION static inline size_t gc_stockpile(const MDBX_txn *txn) {
+  return MDBX_PNL_GETSIZE(txn->wr.repnl) + txn->wr.loose_count;
+}
+
+MDBX_NOTHROW_PURE_FUNCTION static inline size_t gc_chunk_bytes(const size_t chunk) {
+  return (chunk + 1) * sizeof(pgno_t);
+}
+
+MDBX_INTERNAL bool gc_repnl_has_span(const MDBX_txn *txn, const size_t num);
+
+static inline bool gc_is_reclaimed(const MDBX_txn *txn, const txnid_t id) {
+  return rkl_contain(&txn->wr.gc.reclaimed, id) || rkl_contain(&txn->wr.gc.comeback, id);
+}
+
+static inline txnid_t txnid_min(txnid_t a, txnid_t b) { return (a < b) ? a : b; }
+
+static inline txnid_t txnid_max(txnid_t a, txnid_t b) { return (a > b) ? a : b; }

src/global.c

@@ -0,0 +1,474 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+static void mdbx_init(void);
+static void mdbx_fini(void);
+
+/*----------------------------------------------------------------------------*/
+/* mdbx constructor/destructor */
+
+#if defined(_WIN32) || defined(_WIN64)
+
+#if MDBX_BUILD_SHARED_LIBRARY
+#if MDBX_WITHOUT_MSVC_CRT && defined(NDEBUG)
+/* DEBUG/CHECKED builds still require MSVC's CRT for runtime checks.
+ *
+ * Define dll's entry point only for Release build when NDEBUG is defined and
+ * MDBX_WITHOUT_MSVC_CRT=ON. if the entry point isn't defined then MSVC's will
+ * automatically use DllMainCRTStartup() from CRT library, which also
+ * automatically call DllMain() from our mdbx.dll */
+#pragma comment(linker, "/ENTRY:DllMain")
+#endif /* MDBX_WITHOUT_MSVC_CRT */
+
+BOOL APIENTRY DllMain(HANDLE module, DWORD reason, LPVOID reserved)
+#else
+#if !MDBX_MANUAL_MODULE_HANDLER
+static
+#endif /* !MDBX_MANUAL_MODULE_HANDLER */
+    void NTAPI
+    mdbx_module_handler(PVOID module, DWORD reason, PVOID reserved)
+#endif /* MDBX_BUILD_SHARED_LIBRARY */
+{
+  (void)reserved;
+  switch (reason) {
+  case DLL_PROCESS_ATTACH:
+    windows_import();
+    mdbx_init();
+    break;
+  case DLL_PROCESS_DETACH:
+    mdbx_fini();
+    break;
+
+  case DLL_THREAD_ATTACH:
+    break;
+  case DLL_THREAD_DETACH:
+    rthc_thread_dtor(module);
+    break;
+  }
+#if MDBX_BUILD_SHARED_LIBRARY
+  return TRUE;
+#endif
+}
+
+#if !MDBX_BUILD_SHARED_LIBRARY && !MDBX_MANUAL_MODULE_HANDLER
+/* *INDENT-OFF* */
+/* clang-format off */
+#if defined(_MSC_VER)
+#  pragma const_seg(push)
+#  pragma data_seg(push)
+
+#  ifndef _M_IX86
+     /* kick a linker to create the TLS directory if not already done */
+#    pragma comment(linker, "/INCLUDE:_tls_used")
+     /* Force some symbol references. */
+#    pragma comment(linker, "/INCLUDE:mdbx_tls_anchor")
+     /* specific const-segment for WIN64 */
+#    pragma const_seg(".CRT$XLB")
+     const
+#  else
+     /* kick a linker to create the TLS directory if not already done */
+#    pragma comment(linker, "/INCLUDE:__tls_used")
+     /* Force some symbol references. */
+#    pragma comment(linker, "/INCLUDE:_mdbx_tls_anchor")
+     /* specific data-segment for WIN32 */
+#    pragma data_seg(".CRT$XLB")
+#  endif
+
+   __declspec(allocate(".CRT$XLB")) PIMAGE_TLS_CALLBACK mdbx_tls_anchor = mdbx_module_handler;
+#  pragma data_seg(pop)
+#  pragma const_seg(pop)
+
+#elif defined(__GNUC__)
+#  ifndef _M_IX86
+     const
+#  endif
+   PIMAGE_TLS_CALLBACK mdbx_tls_anchor __attribute__((__section__(".CRT$XLB"), used)) = mdbx_module_handler;
+#else
+#  error FIXME
+#endif
+/* *INDENT-ON* */
+/* clang-format on */
+#endif /* !MDBX_BUILD_SHARED_LIBRARY && !MDBX_MANUAL_MODULE_HANDLER */
+
+#else
+
+#if defined(__linux__) || defined(__gnu_linux__)
+#include <sys/utsname.h>
+
+MDBX_EXCLUDE_FOR_GPROF
+__cold static uint8_t probe_for_WSL(const char *tag) {
+  const char *const WSL = strstr(tag, "WSL");
+  if (WSL && WSL[3] >= '2' && WSL[3] <= '9')
+    return WSL[3] - '0';
+  const char *const wsl = strstr(tag, "wsl");
+  if (wsl && wsl[3] >= '2' && wsl[3] <= '9')
+    return wsl[3] - '0';
+  if (WSL || wsl || strcasestr(tag, "Microsoft"))
+    /* Expecting no new kernel within WSL1, either it will explicitly
+     * marked by an appropriate WSL-version hint. */
+    return (globals.linux_kernel_version < /* 4.19.x */ 0x04130000) ? 1 : 2;
+  return 0;
+}
+#endif /* Linux */
+
+#ifdef ENABLE_GPROF
+extern void _mcleanup(void);
+extern void monstartup(unsigned long, unsigned long);
+extern void _init(void);
+extern void _fini(void);
+extern void __gmon_start__(void) __attribute__((__weak__));
+#endif /* ENABLE_GPROF */
+
+MDBX_EXCLUDE_FOR_GPROF
+__cold static __attribute__((__constructor__)) void mdbx_global_constructor(void) {
+#ifdef ENABLE_GPROF
+  if (!&__gmon_start__)
+    monstartup((uintptr_t)&_init, (uintptr_t)&_fini);
+#endif /* ENABLE_GPROF */
+
+#if defined(__linux__) || defined(__gnu_linux__)
+  struct utsname buffer;
+  if (uname(&buffer) == 0) {
+    int i = 0;
+    char *p = buffer.release;
+    while (*p && i < 4) {
+      if (*p >= '0' && *p <= '9') {
+        long number = strtol(p, &p, 10);
+        if (number > 0) {
+          if (number > 255)
+            number = 255;
+          globals.linux_kernel_version += number << (24 - i * 8);
+        }
+        ++i;
+      } else {
+        ++p;
+      }
+    }
+    /* "Official" way of detecting WSL1 but not WSL2
+     * https://github.com/Microsoft/WSL/issues/423#issuecomment-221627364
+     *
+     * WARNING: False negative detection of WSL1 will result in DATA LOSS!
+     * So, the REQUIREMENTS for this code:
+     *  1. MUST detect WSL1 without false-negatives.
+     *  2. DESIRABLE detect WSL2 but without the risk of violating the first. */
+    globals.running_on_WSL1 =
+        probe_for_WSL(buffer.version) == 1 || probe_for_WSL(buffer.sysname) == 1 || probe_for_WSL(buffer.release) == 1;
+  }
+#endif /* Linux */
+
+  mdbx_init();
+}
+
+MDBX_EXCLUDE_FOR_GPROF
+__cold static __attribute__((__destructor__)) void mdbx_global_destructor(void) {
+  mdbx_fini();
+#ifdef ENABLE_GPROF
+  if (!&__gmon_start__)
+    _mcleanup();
+#endif /* ENABLE_GPROF */
+}
+
+#endif /* ! Windows */
+
+/******************************************************************************/
+
+struct libmdbx_globals globals;
+
+__cold static void mdbx_init(void) {
+  globals.runtime_flags = ((MDBX_DEBUG) > 0) * MDBX_DBG_ASSERT + ((MDBX_DEBUG) > 1) * MDBX_DBG_AUDIT;
+  globals.loglevel = MDBX_LOG_FATAL;
+  ENSURE(nullptr, osal_fastmutex_init(&globals.debug_lock) == 0);
+  osal_ctor();
+  assert(globals.sys_pagesize > 0 && (globals.sys_pagesize & (globals.sys_pagesize - 1)) == 0);
+  rthc_ctor();
+#if MDBX_DEBUG
+  ENSURE(nullptr, troika_verify_fsm());
+  ENSURE(nullptr, pv2pages_verify());
+#endif /* MDBX_DEBUG*/
+}
+
+MDBX_EXCLUDE_FOR_GPROF
+__cold static void mdbx_fini(void) {
+  const uint32_t current_pid = osal_getpid();
+  TRACE(">> pid %d", current_pid);
+  rthc_dtor(current_pid);
+  osal_dtor();
+  TRACE("<< pid %d\n", current_pid);
+  ENSURE(nullptr, osal_fastmutex_destroy(&globals.debug_lock) == 0);
+}
+
+/******************************************************************************/
+
+/* *INDENT-OFF* */
+/* clang-format off */
+
+__dll_export
+#ifdef __attribute_used__
+    __attribute_used__
+#elif defined(__GNUC__) || __has_attribute(__used__)
+    __attribute__((__used__))
+#endif
+#ifdef __attribute_externally_visible__
+        __attribute_externally_visible__
+#elif (defined(__GNUC__) && !defined(__clang__)) ||                            \
+    __has_attribute(__externally_visible__)
+    __attribute__((__externally_visible__))
+#endif
+    const struct MDBX_build_info mdbx_build = {
+#ifdef MDBX_BUILD_TIMESTAMP
+    MDBX_BUILD_TIMESTAMP
+#else
+    "\"" __DATE__ " " __TIME__ "\""
+#endif /* MDBX_BUILD_TIMESTAMP */
+
+    ,
+#ifdef MDBX_BUILD_TARGET
+    MDBX_BUILD_TARGET
+#else
+  #if defined(__ANDROID_API__)
+    "Android" MDBX_STRINGIFY(__ANDROID_API__)
+  #elif defined(__linux__) || defined(__gnu_linux__)
+    "Linux"
+  #elif defined(EMSCRIPTEN) || defined(__EMSCRIPTEN__)
+    "webassembly"
+  #elif defined(__CYGWIN__)
+    "CYGWIN"
+  #elif defined(_WIN64) || defined(_WIN32) || defined(__TOS_WIN__) \
+      || defined(__WINDOWS__)
+    "Windows"
+  #elif defined(__APPLE__)
+    #if (defined(TARGET_OS_IPHONE) && TARGET_OS_IPHONE) \
+      || (defined(TARGET_IPHONE_SIMULATOR) && TARGET_IPHONE_SIMULATOR)
+      "iOS"
+    #else
+      "MacOS"
+    #endif
+  #elif defined(__FreeBSD__)
+    "FreeBSD"
+  #elif defined(__DragonFly__)
+    "DragonFlyBSD"
+  #elif defined(__NetBSD__)
+    "NetBSD"
+  #elif defined(__OpenBSD__)
+    "OpenBSD"
+  #elif defined(__bsdi__)
+    "UnixBSDI"
+  #elif defined(__MACH__)
+    "MACH"
+  #elif (defined(_HPUX_SOURCE) || defined(__hpux) || defined(__HP_aCC))
+    "HPUX"
+  #elif defined(_AIX)
+    "AIX"
+  #elif defined(__sun) && defined(__SVR4)
+    "Solaris"
+  #elif defined(__BSD__) || defined(BSD)
+    "UnixBSD"
+  #elif defined(__unix__) || defined(UNIX) || defined(__unix) \
+      || defined(__UNIX) || defined(__UNIX__)
+    "UNIX"
+  #elif defined(_POSIX_VERSION)
+    "POSIX" MDBX_STRINGIFY(_POSIX_VERSION)
+  #else
+    "UnknownOS"
+  #endif /* Target OS */
+
+    "-"
+
+  #if defined(__amd64__)
+    "AMD64"
+  #elif defined(__ia32__)
+    "IA32"
+  #elif defined(__e2k__) || defined(__elbrus__)
+    "Elbrus"
+  #elif defined(__alpha__) || defined(__alpha) || defined(_M_ALPHA)
+    "Alpha"
+  #elif defined(__aarch64__) || defined(_M_ARM64)
+    "ARM64"
+  #elif defined(__arm__) || defined(__thumb__) || defined(__TARGET_ARCH_ARM) \
+      || defined(__TARGET_ARCH_THUMB) || defined(_ARM) || defined(_M_ARM) \
+      || defined(_M_ARMT) || defined(__arm)
+    "ARM"
+  #elif defined(__mips64) || defined(__mips64__) || (defined(__mips) && (__mips >= 64))
+    "MIPS64"
+  #elif defined(__mips__) || defined(__mips) || defined(_R4000) || defined(__MIPS__)
+    "MIPS"
+  #elif defined(__hppa64__) || defined(__HPPA64__) || defined(__hppa64)
+    "PARISC64"
+  #elif defined(__hppa__) || defined(__HPPA__) || defined(__hppa)
+    "PARISC"
+  #elif defined(__ia64__) || defined(__ia64) || defined(_IA64) \
+      || defined(__IA64__) || defined(_M_IA64) || defined(__itanium__)
+    "Itanium"
+  #elif defined(__powerpc64__) || defined(__ppc64__) || defined(__ppc64) \
+      || defined(__powerpc64) || defined(_ARCH_PPC64)
+    "PowerPC64"
+  #elif defined(__powerpc__) || defined(__ppc__) || defined(__powerpc) \
+      || defined(__ppc) || defined(_ARCH_PPC) || defined(__PPC__) || defined(__POWERPC__)
+    "PowerPC"
+  #elif defined(__sparc64__) || defined(__sparc64)
+    "SPARC64"
+  #elif defined(__sparc__) || defined(__sparc)
+    "SPARC"
+  #elif defined(__s390__) || defined(__s390) || defined(__zarch__) || defined(__zarch)
+    "S390"
+  #else
+    "UnknownARCH"
+  #endif
+#endif /* MDBX_BUILD_TARGET */
+
+#ifdef MDBX_BUILD_TYPE
+# if defined(_MSC_VER)
+#   pragma message("Configuration-depended MDBX_BUILD_TYPE: " MDBX_BUILD_TYPE)
+# endif
+    "-" MDBX_BUILD_TYPE
+#endif /* MDBX_BUILD_TYPE */
+    ,
+    "MDBX_DEBUG=" MDBX_STRINGIFY(MDBX_DEBUG)
+#ifdef ENABLE_GPROF
+    " ENABLE_GPROF"
+#endif /* ENABLE_GPROF */
+    " MDBX_WORDBITS=" MDBX_STRINGIFY(MDBX_WORDBITS)
+    " BYTE_ORDER="
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+    "LITTLE_ENDIAN"
+#elif __BYTE_ORDER__ == __ORDER_BIG_ENDIAN__
+    "BIG_ENDIAN"
+#else
+    #error "FIXME: Unsupported byte order"
+#endif /* __BYTE_ORDER__ */
+    " MDBX_ENABLE_BIGFOOT=" MDBX_STRINGIFY(MDBX_ENABLE_BIGFOOT)
+    " MDBX_ENV_CHECKPID=" MDBX_ENV_CHECKPID_CONFIG
+    " MDBX_TXN_CHECKOWNER=" MDBX_TXN_CHECKOWNER_CONFIG
+    " MDBX_64BIT_ATOMIC=" MDBX_64BIT_ATOMIC_CONFIG
+    " MDBX_64BIT_CAS=" MDBX_64BIT_CAS_CONFIG
+    " MDBX_TRUST_RTC=" MDBX_TRUST_RTC_CONFIG
+    " MDBX_AVOID_MSYNC=" MDBX_STRINGIFY(MDBX_AVOID_MSYNC)
+    " MDBX_ENABLE_REFUND=" MDBX_STRINGIFY(MDBX_ENABLE_REFUND)
+    " MDBX_USE_MINCORE=" MDBX_STRINGIFY(MDBX_USE_MINCORE)
+    " MDBX_ENABLE_PGOP_STAT=" MDBX_STRINGIFY(MDBX_ENABLE_PGOP_STAT)
+    " MDBX_ENABLE_PROFGC=" MDBX_STRINGIFY(MDBX_ENABLE_PROFGC)
+#if MDBX_DISABLE_VALIDATION
+    " MDBX_DISABLE_VALIDATION=YES"
+#endif /* MDBX_DISABLE_VALIDATION */
+#ifdef __SANITIZE_ADDRESS__
+    " SANITIZE_ADDRESS=YES"
+#endif /* __SANITIZE_ADDRESS__ */
+#ifdef ENABLE_MEMCHECK
+    " ENABLE_MEMCHECK=YES"
+#endif /* ENABLE_MEMCHECK */
+#if MDBX_FORCE_ASSERTIONS
+    " MDBX_FORCE_ASSERTIONS=YES"
+#endif /* MDBX_FORCE_ASSERTIONS */
+#ifdef _GNU_SOURCE
+    " _GNU_SOURCE=YES"
+#else
+    " _GNU_SOURCE=NO"
+#endif /* _GNU_SOURCE */
+#ifdef __APPLE__
+    " MDBX_APPLE_SPEED_INSTEADOF_DURABILITY=" MDBX_STRINGIFY(MDBX_APPLE_SPEED_INSTEADOF_DURABILITY)
+#endif /* MacOS */
+#if defined(_WIN32) || defined(_WIN64)
+    " MDBX_WITHOUT_MSVC_CRT=" MDBX_STRINGIFY(MDBX_WITHOUT_MSVC_CRT)
+    " MDBX_BUILD_SHARED_LIBRARY=" MDBX_STRINGIFY(MDBX_BUILD_SHARED_LIBRARY)
+#if !MDBX_BUILD_SHARED_LIBRARY
+    " MDBX_MANUAL_MODULE_HANDLER=" MDBX_STRINGIFY(MDBX_MANUAL_MODULE_HANDLER)
+#endif
+    " WINVER=" MDBX_STRINGIFY(WINVER)
+#else /* Windows */
+    " MDBX_LOCKING=" MDBX_LOCKING_CONFIG
+    " MDBX_USE_OFDLOCKS=" MDBX_USE_OFDLOCKS_CONFIG
+#endif /* !Windows */
+    " MDBX_CACHELINE_SIZE=" MDBX_STRINGIFY(MDBX_CACHELINE_SIZE)
+    " MDBX_CPU_WRITEBACK_INCOHERENT=" MDBX_STRINGIFY(MDBX_CPU_WRITEBACK_INCOHERENT)
+    " MDBX_MMAP_INCOHERENT_CPU_CACHE=" MDBX_STRINGIFY(MDBX_MMAP_INCOHERENT_CPU_CACHE)
+    " MDBX_MMAP_INCOHERENT_FILE_WRITE=" MDBX_STRINGIFY(MDBX_MMAP_INCOHERENT_FILE_WRITE)
+    " MDBX_UNALIGNED_OK=" MDBX_STRINGIFY(MDBX_UNALIGNED_OK)
+    " MDBX_PNL_ASCENDING=" MDBX_STRINGIFY(MDBX_PNL_ASCENDING)
+    ,
+#ifdef MDBX_BUILD_COMPILER
+    MDBX_BUILD_COMPILER
+#else
+  #ifdef __INTEL_COMPILER
+    "Intel C/C++ " MDBX_STRINGIFY(__INTEL_COMPILER)
+  #elif defined(__apple_build_version__)
+    "Apple clang " MDBX_STRINGIFY(__apple_build_version__)
+  #elif defined(__ibmxl__)
+    "IBM clang C " MDBX_STRINGIFY(__ibmxl_version__) "." MDBX_STRINGIFY(__ibmxl_release__)
+    "." MDBX_STRINGIFY(__ibmxl_modification__) "." MDBX_STRINGIFY(__ibmxl_ptf_fix_level__)
+  #elif defined(__clang__)
+    "clang " MDBX_STRINGIFY(__clang_version__)
+  #elif defined(__MINGW64__)
+    "MINGW-64 " MDBX_STRINGIFY(__MINGW64_MAJOR_VERSION) "." MDBX_STRINGIFY(__MINGW64_MINOR_VERSION)
+  #elif defined(__MINGW32__)
+    "MINGW-32 " MDBX_STRINGIFY(__MINGW32_MAJOR_VERSION) "." MDBX_STRINGIFY(__MINGW32_MINOR_VERSION)
+  #elif defined(__MINGW__)
+    "MINGW " MDBX_STRINGIFY(__MINGW_MAJOR_VERSION) "." MDBX_STRINGIFY(__MINGW_MINOR_VERSION)
+  #elif defined(__IBMC__)
+    "IBM C " MDBX_STRINGIFY(__IBMC__)
+  #elif defined(__GNUC__)
+    "GNU C/C++ "
+    #ifdef __VERSION__
+      __VERSION__
+    #else
+      MDBX_STRINGIFY(__GNUC__) "." MDBX_STRINGIFY(__GNUC_MINOR__) "." MDBX_STRINGIFY(__GNUC_PATCHLEVEL__)
+    #endif
+  #elif defined(_MSC_VER)
+    "MSVC " MDBX_STRINGIFY(_MSC_FULL_VER) "-" MDBX_STRINGIFY(_MSC_BUILD)
+  #else
+    "Unknown compiler"
+  #endif
+#endif /* MDBX_BUILD_COMPILER */
+    ,
+#ifdef MDBX_BUILD_FLAGS_CONFIG
+    MDBX_BUILD_FLAGS_CONFIG
+#endif /* MDBX_BUILD_FLAGS_CONFIG */
+#if defined(MDBX_BUILD_FLAGS_CONFIG) && defined(MDBX_BUILD_FLAGS)
+    " "
+#endif
+#ifdef MDBX_BUILD_FLAGS
+    MDBX_BUILD_FLAGS
+#endif /* MDBX_BUILD_FLAGS */
+#if !(defined(MDBX_BUILD_FLAGS_CONFIG) || defined(MDBX_BUILD_FLAGS))
+    "undefined (please use correct build script)"
+#ifdef _MSC_VER
+#pragma message("warning: Build flags undefined. Please use correct build script")
+#else
+#warning "Build flags undefined. Please use correct build script"
+#endif // _MSC_VER
+#endif
+  , MDBX_BUILD_METADATA
+};
+
+#ifdef __SANITIZE_ADDRESS__
+#if !defined(_MSC_VER) || __has_attribute(weak)
+LIBMDBX_API __attribute__((__weak__))
+#endif
+const char *__asan_default_options(void) {
+  return "symbolize=1:allow_addr2line=1:"
+#if MDBX_DEBUG
+         "debug=1:"
+         "verbosity=2:"
+#endif /* MDBX_DEBUG */
+         "log_threads=1:"
+         "report_globals=1:"
+         "replace_str=1:replace_intrin=1:"
+         "malloc_context_size=9:"
+#if !defined(__APPLE__)
+         "detect_leaks=1:"
+#endif
+         "check_printf=1:"
+         "detect_deadlocks=1:"
+#ifndef LTO_ENABLED
+         "check_initialization_order=1:"
+#endif
+         "detect_stack_use_after_return=1:"
+         "intercept_tls_get_addr=1:"
+         "decorate_proc_maps=1:"
+         "abort_on_error=1";
+}
+#endif /* __SANITIZE_ADDRESS__ */
+
+/* *INDENT-ON* */
+/* clang-format on */

src/internals.h

@@ -0,0 +1,587 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \note Please refer to the COPYRIGHT file for explanations license change,
+/// credits and acknowledgments.
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#pragma once
+
+/*----------------------------------------------------------------------------*/
+
+#include "essentials.h"
+
+typedef struct dp dp_t;
+typedef struct dpl dpl_t;
+typedef struct kvx kvx_t;
+typedef struct meta_ptr meta_ptr_t;
+typedef struct inner_cursor subcur_t;
+typedef struct cursor_couple cursor_couple_t;
+typedef struct defer_free_item defer_free_item_t;
+
+typedef struct troika {
+  uint8_t fsm, recent, prefer_steady, tail_and_flags;
+#if MDBX_WORDBITS > 32 /* Workaround for false-positives from Valgrind */
+  uint32_t unused_pad;
+#endif
+#define TROIKA_HAVE_STEADY(troika) ((troika)->fsm & 7u)
+#define TROIKA_STRICT_VALID(troika) ((troika)->tail_and_flags & 64u)
+#define TROIKA_VALID(troika) ((troika)->tail_and_flags & 128u)
+#define TROIKA_TAIL(troika) ((troika)->tail_and_flags & 3u)
+  txnid_t txnid[NUM_METAS];
+} troika_t;
+
+typedef struct page_get_result {
+  page_t *page;
+  int err;
+} pgr_t;
+
+typedef struct node_search_result {
+  node_t *node;
+  bool exact;
+} nsr_t;
+
+typedef struct bind_reader_slot_result {
+  int err;
+  reader_slot_t *slot;
+} bsr_t;
+
+#include "atomics-ops.h"
+#include "proto.h"
+#include "rkl.h"
+#include "txl.h"
+#include "unaligned.h"
+#if defined(_WIN32) || defined(_WIN64)
+#include "windows-import.h"
+#endif /* Windows */
+
+enum signatures {
+  env_signature = INT32_C(0x1A899641),
+  txn_signature = INT32_C(0x13D53A31),
+  cur_signature_live = INT32_C(0x7E05D5B1),
+  cur_signature_ready4dispose = INT32_C(0x2817A047),
+  cur_signature_wait4eot = INT32_C(0x10E297A7)
+};
+
+/*----------------------------------------------------------------------------*/
+
+/* An dirty-page list item is an pgno/pointer pair. */
+struct dp {
+  page_t *ptr;
+  pgno_t pgno, npages;
+};
+
+enum dpl_rules {
+  dpl_gap_edging = 2,
+  dpl_gap_mergesort = 16,
+  dpl_reserve_gap = dpl_gap_mergesort + dpl_gap_edging,
+  dpl_insertion_threshold = 42
+};
+
+/* An DPL (dirty-page list) is a lazy-sorted array of MDBX_DPs. */
+struct dpl {
+  size_t sorted;
+  size_t length;
+  /* number of pages, but not an entries. */
+  size_t pages_including_loose;
+  /* allocated size excluding the dpl_reserve_gap */
+  size_t detent;
+  /* dynamic size with holes at zero and after the last */
+  dp_t items[dpl_reserve_gap];
+};
+
+/*----------------------------------------------------------------------------*/
+/* Internal structures */
+
+/* Comparing/ordering and length constraints */
+typedef struct clc {
+  MDBX_cmp_func *cmp; /* comparator */
+  size_t lmin, lmax;  /* min/max length constraints */
+} clc_t;
+
+/* Вспомогательная информация о table.
+ *
+ * Совокупность потребностей:
+ * 1. Для транзакций и основного курсора нужны все поля.
+ * 2. Для вложенного dupsort-курсора нужен компаратор значений, который изнутри
+ *    курсора будет выглядеть как компаратор ключей. Плюс заглушка компаратора
+ *    значений, которая не должна использоваться в штатных ситуациях, но
+ *    требуется хотя-бы для отслеживания таких обращений.
+ * 3. Использование компараторов для курсора и вложенного dupsort-курсора
+ *    должно выглядеть одинаково.
+ * 4. Желательно минимизировать объём данных размещаемых внутри вложенного
+ *    dupsort-курсора.
+ * 5. Желательно чтобы объем всей структуры был степенью двойки.
+ *
+ * Решение:
+ *  - не храним в dupsort-курсоре ничего лишнего, а только tree;
+ *  - в курсоры помещаем только указатель на clc_t, который будет указывать
+ *    на соответствующее clc-поле в общей kvx-таблице привязанной к env;
+ *  - компаратор размещаем в начале clc_t, в kvx_t сначала размещаем clc
+ *    для ключей, потом для значений, а имя БД в конце kvx_t.
+ *  - тогда в курсоре clc[0] будет содержать информацию для ключей,
+ *    а clc[1] для значений, причем компаратор значений для dupsort-курсора
+ *    будет попадать на MDBX_val с именем, что приведет к SIGSEGV при попытке
+ *    использования такого компаратора.
+ *  - размер kvx_t становится равным 8 словам.
+ *
+ * Трюки и прочая экономия на спичках:
+ *  - не храним dbi внутри курсора, вместо этого вычисляем его как разницу между
+ *    dbi_state курсора и началом таблицы dbi_state в транзакции. Смысл тут в
+ *    экономии кол-ва полей при инициализации курсора. Затрат это не создает,
+ *    так как dbi требуется для последующего доступа к массивам в транзакции,
+ *    т.е. при вычислении dbi разыменовывается тот-же указатель на txn
+ *    и читается та же кэш-линия с указателями. */
+typedef struct clc2 {
+  clc_t k; /* для ключей */
+  clc_t v; /* для значений */
+} clc2_t;
+
+struct kvx {
+  clc2_t clc;
+  MDBX_val name; /* имя table */
+};
+
+/* Non-shared DBI state flags inside transaction */
+enum dbi_state {
+  DBI_DIRTY = 0x01 /* DB was written in this txn */,
+  DBI_STALE = 0x02 /* Named-DB record is older than txnID */,
+  DBI_FRESH = 0x04 /* Named-DB handle opened in this txn */,
+  DBI_CREAT = 0x08 /* Named-DB handle created in this txn */,
+  DBI_VALID = 0x10 /* Handle is valid, see also DB_VALID */,
+  DBI_OLDEN = 0x40 /* Handle was closed/reopened outside txn */,
+  DBI_LINDO = 0x80 /* Lazy initialization done for DBI-slot */,
+};
+
+enum txn_flags {
+  txn_ro_begin_flags = MDBX_TXN_RDONLY | MDBX_TXN_RDONLY_PREPARE,
+  txn_rw_begin_flags = MDBX_TXN_NOMETASYNC | MDBX_TXN_NOSYNC | MDBX_TXN_TRY,
+  txn_shrink_allowed = UINT32_C(0x40000000),
+  txn_parked = MDBX_TXN_PARKED,
+  txn_gc_drained = 0x80 /* GC was depleted up to oldest reader */,
+  txn_may_have_cursors = 0x100,
+  txn_state_flags = MDBX_TXN_FINISHED | MDBX_TXN_ERROR | MDBX_TXN_DIRTY | MDBX_TXN_SPILLS | MDBX_TXN_HAS_CHILD |
+                    MDBX_TXN_INVALID | txn_gc_drained
+};
+
+/* A database transaction.
+ * Every operation requires a transaction handle. */
+struct MDBX_txn {
+  int32_t signature;
+  uint32_t flags; /* Transaction Flags */
+  size_t n_dbi;
+  size_t owner; /* thread ID that owns this transaction */
+
+  MDBX_txn *parent; /* parent of a nested txn */
+  MDBX_txn *nested; /* nested txn under this txn,
+                       set together with MDBX_TXN_HAS_CHILD */
+  geo_t geo;
+
+  /* The ID of this transaction. IDs are integers incrementing from
+   * INITIAL_TXNID. Only committed write transactions increment the ID. If a
+   * transaction aborts, the ID may be re-used by the next writer. */
+  txnid_t txnid, front_txnid;
+
+  MDBX_env *env; /* the DB environment */
+  tree_t *dbs;   /* Array of tree_t records for each known DB */
+
+#if MDBX_ENABLE_DBI_SPARSE
+  unsigned *__restrict dbi_sparse;
+#endif /* MDBX_ENABLE_DBI_SPARSE */
+
+  /* Array of non-shared txn's flags of DBI.
+   * Модификатор __restrict тут полезен и безопасен в текущем понимании,
+   * так как пересечение возможно только с dbi_state курсоров,
+   * и происходит по-чтению до последующего изменения/записи. */
+  uint8_t *__restrict dbi_state;
+
+  /* Array of sequence numbers for each DB handle. */
+  uint32_t *__restrict dbi_seqs;
+
+  /* Массив с головами односвязных списков отслеживания курсоров. */
+  MDBX_cursor **cursors;
+
+  /* "Канареечные" маркеры/счетчики */
+  MDBX_canary canary;
+
+  /* User-settable context */
+  void *userctx;
+
+  union {
+    struct {
+      /* For read txns: This thread/txn's slot table slot, or nullptr. */
+      reader_slot_t *slot;
+    } ro;
+    struct {
+      troika_t troika;
+      pnl_t __restrict repnl; /* Reclaimed GC pages */
+      struct {
+        rkl_t reclaimed; /* The list of reclaimed txn-ids from GC */
+        uint64_t spent;  /* Time spent reading and searching GC */
+        rkl_t comeback;  /* The list of ids of records returned into GC during commit, etc */
+      } gc;
+      bool prefault_write_activated;
+#if MDBX_ENABLE_REFUND
+      pgno_t loose_refund_wl /* FIXME: describe */;
+#endif /* MDBX_ENABLE_REFUND */
+      /* a sequence to spilling dirty page with LRU policy */
+      unsigned dirtylru;
+      /* dirtylist room: Dirty array size - dirty pages visible to this txn.
+       * Includes ancestor txns' dirty pages not hidden by other txns'
+       * dirty/spilled pages. Thus commit(nested txn) has room to merge
+       * dirtylist into parent after freeing hidden parent pages. */
+      size_t dirtyroom;
+      /* For write txns: Modified pages. Sorted when not MDBX_WRITEMAP. */
+      dpl_t *__restrict dirtylist;
+      /* The list of pages that became unused during this transaction. */
+      pnl_t __restrict retired_pages;
+      /* The list of loose pages that became unused and may be reused
+       * in this transaction, linked through `page_next()`. */
+      page_t *__restrict loose_pages;
+      /* Number of loose pages (wr.loose_pages) */
+      size_t loose_count;
+      union {
+        struct {
+          size_t least_removed;
+          /* The sorted list of dirty pages we temporarily wrote to disk
+           * because the dirty list was full. page numbers in here are
+           * shifted left by 1, deleted slots have the LSB set. */
+          pnl_t __restrict list;
+        } spilled;
+        size_t writemap_dirty_npages;
+        size_t writemap_spilled_npages;
+      };
+      /* In write txns, next is located the array of cursors for each DB */
+    } wr;
+  };
+};
+
+#define CURSOR_STACK_SIZE (16 + MDBX_WORDBITS / 4)
+
+struct MDBX_cursor {
+  int32_t signature;
+  union {
+    /* Тут некоторые трюки/заморочки с тем чтобы во всех основных сценариях
+     * проверять состояние курсора одной простой операцией сравнения,
+     * и при этом ни на каплю не усложнять код итерации стека курсора.
+     *
+     * Поэтому решение такое:
+     *  - поля flags и top сделаны знаковыми, а их отрицательные значения
+     *    используются для обозначения не-установленного/не-инициализированного
+     *    состояния курсора;
+     *  - для инвалидации/сброса курсора достаточно записать отрицательное
+     *    значение в объединенное поле top_and_flags;
+     *  - все проверки состояния сводятся к сравнению одного из полей
+     *    flags/snum/snum_and_flags, которые в зависимости от сценария,
+     *    трактуются либо как знаковые, либо как безнаковые. */
+    __anonymous_struct_extension__ struct {
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+      int8_t flags;
+      /* индекс вершины стека, меньше нуля для не-инициализированного курсора */
+      int8_t top;
+#else
+      int8_t top;
+      int8_t flags;
+#endif
+    };
+    int16_t top_and_flags;
+  };
+  /* флаги проверки, в том числе биты для проверки типа листовых страниц. */
+  uint8_t checking;
+  uint8_t pad;
+
+  /* Указывает на txn->dbi_state[] для DBI этого курсора.
+   * Модификатор __restrict тут полезен и безопасен в текущем понимании,
+   * так как пересечение возможно только с dbi_state транзакции,
+   * и происходит по-чтению до последующего изменения/записи. */
+  uint8_t *__restrict dbi_state;
+  /* Связь списка отслеживания курсоров в транзакции. */
+  MDBX_txn *txn;
+  /* Указывает на tree->dbs[] для DBI этого курсора. */
+  tree_t *tree;
+  /* Указывает на env->kvs[] для DBI этого курсора. */
+  clc2_t *clc;
+  subcur_t *__restrict subcur;
+  page_t *pg[CURSOR_STACK_SIZE]; /* stack of pushed pages */
+  indx_t ki[CURSOR_STACK_SIZE];  /* stack of page indices */
+  MDBX_cursor *next;
+  /* Состояние на момент старта вложенной транзакции */
+  MDBX_cursor *backup;
+};
+
+struct inner_cursor {
+  MDBX_cursor cursor;
+  tree_t nested_tree;
+};
+
+struct cursor_couple {
+  MDBX_cursor outer;
+  void *userctx; /* User-settable context */
+  subcur_t inner;
+};
+
+enum env_flags {
+  /* Failed to update the meta page. Probably an I/O error. */
+  ENV_FATAL_ERROR = INT32_MIN /* 0x80000000 */,
+  /* Some fields are initialized. */
+  ENV_ACTIVE = UINT32_C(0x20000000),
+  /* me_txkey is set */
+  ENV_TXKEY = UINT32_C(0x10000000),
+  /* Legacy MDBX_MAPASYNC (prior v0.9) */
+  DEPRECATED_MAPASYNC = UINT32_C(0x100000),
+  /* Legacy MDBX_COALESCE (prior v0.12) */
+  DEPRECATED_COALESCE = UINT32_C(0x2000000),
+  ENV_INTERNAL_FLAGS = ENV_FATAL_ERROR | ENV_ACTIVE | ENV_TXKEY,
+  /* Only a subset of the mdbx_env flags can be changed
+   * at runtime. Changing other flags requires closing the
+   * environment and re-opening it with the new flags. */
+  ENV_CHANGEABLE_FLAGS = MDBX_SAFE_NOSYNC | MDBX_NOMETASYNC | DEPRECATED_MAPASYNC | MDBX_NOMEMINIT |
+                         DEPRECATED_COALESCE | MDBX_PAGEPERTURB | MDBX_ACCEDE | MDBX_VALIDATION,
+  ENV_CHANGELESS_FLAGS = MDBX_NOSUBDIR | MDBX_RDONLY | MDBX_WRITEMAP | MDBX_NOSTICKYTHREADS | MDBX_NORDAHEAD |
+                         MDBX_LIFORECLAIM | MDBX_EXCLUSIVE,
+  ENV_USABLE_FLAGS = ENV_CHANGEABLE_FLAGS | ENV_CHANGELESS_FLAGS
+};
+
+/* The database environment. */
+struct MDBX_env {
+  /* ----------------------------------------------------- mostly static part */
+  mdbx_atomic_uint32_t signature;
+  uint32_t flags;
+  unsigned ps;          /* DB page size, initialized from me_os_psize */
+  osal_mmap_t dxb_mmap; /* The main data file */
+#define lazy_fd dxb_mmap.fd
+  mdbx_filehandle_t dsync_fd, fd4meta;
+#if defined(_WIN32) || defined(_WIN64)
+  HANDLE dxb_lock_event;
+#endif                  /* Windows */
+  osal_mmap_t lck_mmap; /* The lock file */
+  lck_t *lck;
+
+  uint16_t leaf_nodemax;   /* max size of a leaf-node */
+  uint16_t branch_nodemax; /* max size of a branch-node */
+  uint16_t subpage_limit;
+  uint16_t subpage_room_threshold;
+  uint16_t subpage_reserve_prereq;
+  uint16_t subpage_reserve_limit;
+  atomic_pgno_t mlocked_pgno;
+  uint8_t ps2ln;              /* log2 of DB page size */
+  int8_t stuck_meta;          /* recovery-only: target meta page or less that zero */
+  uint16_t merge_threshold;   /* pages emptier than this are candidates for merging */
+  unsigned max_readers;       /* size of the reader table */
+  MDBX_dbi max_dbi;           /* size of the DB table */
+  uint32_t pid;               /* process ID of this env */
+  osal_thread_key_t me_txkey; /* thread-key for readers */
+  struct {                    /* path to the DB files */
+    pathchar_t *lck, *dxb, *specified;
+    void *buffer;
+  } pathname;
+  void *page_auxbuf;              /* scratch area for DUPSORT put() */
+  MDBX_txn *basal_txn;            /* preallocated write transaction */
+  kvx_t *kvs;                     /* array of auxiliary key-value properties */
+  uint8_t *__restrict dbs_flags;  /* array of flags from tree_t.flags */
+  mdbx_atomic_uint32_t *dbi_seqs; /* array of dbi sequence numbers */
+  unsigned maxgc_large1page;      /* Number of pgno_t fit in a single large page */
+  unsigned maxgc_per_branch;
+  uint32_t registered_reader_pid; /* have liveness lock in reader table */
+  void *userctx;                  /* User-settable context */
+  MDBX_hsr_func *hsr_callback;    /* Callback for kicking laggard readers */
+  size_t madv_threshold;
+
+  struct {
+    unsigned dp_reserve_limit;
+    unsigned rp_augment_limit;
+    unsigned dp_limit;
+    unsigned dp_initial;
+    uint64_t gc_time_limit;
+    uint8_t dp_loose_limit;
+    uint8_t spill_max_denominator;
+    uint8_t spill_min_denominator;
+    uint8_t spill_parent4child_denominator;
+    unsigned merge_threshold_16dot16_percent;
+#if !(defined(_WIN32) || defined(_WIN64))
+    unsigned writethrough_threshold;
+#endif /* Windows */
+    bool prefault_write;
+    bool prefer_waf_insteadof_balance; /* Strive to minimize WAF instead of
+                                          balancing pages fullment */
+    bool need_dp_limit_adjust;
+    struct {
+      uint16_t limit;
+      uint16_t room_threshold;
+      uint16_t reserve_prereq;
+      uint16_t reserve_limit;
+    } subpage;
+
+    union {
+      unsigned all;
+      /* tracks options with non-auto values but tuned by user */
+      struct {
+        unsigned dp_limit : 1;
+        unsigned rp_augment_limit : 1;
+        unsigned prefault_write : 1;
+      } non_auto;
+    } flags;
+  } options;
+
+  /* struct geo_in_bytes used for accepting db-geo params from user for the new
+   * database creation, i.e. when mdbx_env_set_geometry() was called before
+   * mdbx_env_open(). */
+  struct {
+    size_t lower;  /* minimal size of datafile */
+    size_t upper;  /* maximal size of datafile */
+    size_t now;    /* current size of datafile */
+    size_t grow;   /* step to grow datafile */
+    size_t shrink; /* threshold to shrink datafile */
+  } geo_in_bytes;
+
+#if MDBX_LOCKING == MDBX_LOCKING_SYSV
+  union {
+    key_t key;
+    int semid;
+  } me_sysv_ipc;
+#endif /* MDBX_LOCKING == MDBX_LOCKING_SYSV */
+  bool incore;
+
+#if MDBX_ENABLE_DBI_LOCKFREE
+  defer_free_item_t *defer_free;
+#endif /* MDBX_ENABLE_DBI_LOCKFREE */
+
+  /* -------------------------------------------------------------- debugging */
+
+#if MDBX_DEBUG
+  MDBX_assert_func *assert_func; /*  Callback for assertion failures */
+#endif
+#ifdef ENABLE_MEMCHECK
+  int valgrind_handle;
+#endif
+#if defined(ENABLE_MEMCHECK) || defined(__SANITIZE_ADDRESS__)
+  pgno_t poison_edge;
+#endif /* ENABLE_MEMCHECK || __SANITIZE_ADDRESS__ */
+
+#ifndef xMDBX_DEBUG_SPILLING
+#define xMDBX_DEBUG_SPILLING 0
+#endif
+#if xMDBX_DEBUG_SPILLING == 2
+  size_t debug_dirtied_est, debug_dirtied_act;
+#endif /* xMDBX_DEBUG_SPILLING */
+
+  /* --------------------------------------------------- mostly volatile part */
+
+  MDBX_txn *txn; /* current write transaction */
+  struct {
+    txnid_t detent;
+  } gc;
+  osal_fastmutex_t dbi_lock;
+  unsigned n_dbi; /* number of DBs opened */
+
+  unsigned shadow_reserve_len;
+  page_t *__restrict shadow_reserve; /* list of malloc'ed blocks for re-use */
+
+  osal_ioring_t ioring;
+
+#if defined(_WIN32) || defined(_WIN64)
+  osal_srwlock_t remap_guard;
+  /* Workaround for LockFileEx and WriteFile multithread bug */
+  CRITICAL_SECTION windowsbug_lock;
+  char *pathname_char; /* cache of multi-byte representation of pathname
+                             to the DB files */
+#else
+  osal_fastmutex_t remap_guard;
+#endif
+
+  /* ------------------------------------------------- stub for lck-less mode */
+  mdbx_atomic_uint64_t lckless_placeholder[(sizeof(lck_t) + MDBX_CACHELINE_SIZE - 1) / sizeof(mdbx_atomic_uint64_t)];
+};
+
+/*----------------------------------------------------------------------------*/
+
+/* pseudo-error code, not exposed outside libmdbx */
+#define MDBX_NO_ROOT (MDBX_LAST_ADDED_ERRCODE + 33)
+
+/* Number of slots in the reader table.
+ * This value was chosen somewhat arbitrarily. The 61 is a prime number,
+ * and such readers plus a couple mutexes fit into single 4KB page.
+ * Applications should set the table size using mdbx_env_set_maxreaders(). */
+#define DEFAULT_READERS 61
+
+enum db_flags {
+  DB_PERSISTENT_FLAGS =
+      MDBX_REVERSEKEY | MDBX_DUPSORT | MDBX_INTEGERKEY | MDBX_DUPFIXED | MDBX_INTEGERDUP | MDBX_REVERSEDUP,
+
+  /* mdbx_dbi_open() flags */
+  DB_USABLE_FLAGS = DB_PERSISTENT_FLAGS | MDBX_CREATE | MDBX_DB_ACCEDE,
+
+  DB_VALID = 0x80u /* DB handle is valid, for dbs_flags */,
+  DB_POISON = 0x7fu /* update pending */,
+  DB_INTERNAL_FLAGS = DB_VALID
+};
+
+#if !defined(__cplusplus) || CONSTEXPR_ENUM_FLAGS_OPERATIONS
+MDBX_MAYBE_UNUSED static void static_checks(void) {
+  STATIC_ASSERT(MDBX_WORDBITS == sizeof(void *) * CHAR_BIT);
+  STATIC_ASSERT(UINT64_C(0x80000000) == (uint32_t)ENV_FATAL_ERROR);
+  STATIC_ASSERT_MSG(INT16_MAX - CORE_DBS == MDBX_MAX_DBI, "Oops, MDBX_MAX_DBI or CORE_DBS?");
+  STATIC_ASSERT_MSG((unsigned)(MDBX_DB_ACCEDE | MDBX_CREATE) ==
+                        ((DB_USABLE_FLAGS | DB_INTERNAL_FLAGS) & (ENV_USABLE_FLAGS | ENV_INTERNAL_FLAGS)),
+                    "Oops, some flags overlapped or wrong");
+  STATIC_ASSERT_MSG((DB_INTERNAL_FLAGS & DB_USABLE_FLAGS) == 0, "Oops, some flags overlapped or wrong");
+  STATIC_ASSERT_MSG((DB_PERSISTENT_FLAGS & ~DB_USABLE_FLAGS) == 0, "Oops, some flags overlapped or wrong");
+  STATIC_ASSERT(DB_PERSISTENT_FLAGS <= UINT8_MAX);
+  STATIC_ASSERT_MSG((ENV_INTERNAL_FLAGS & ENV_USABLE_FLAGS) == 0, "Oops, some flags overlapped or wrong");
+
+  STATIC_ASSERT_MSG((txn_state_flags & (txn_rw_begin_flags | txn_ro_begin_flags)) == 0,
+                    "Oops, some txn flags overlapped or wrong");
+  STATIC_ASSERT_MSG(((txn_rw_begin_flags | txn_ro_begin_flags | txn_state_flags) & txn_shrink_allowed) == 0,
+                    "Oops, some txn flags overlapped or wrong");
+
+  STATIC_ASSERT(sizeof(reader_slot_t) == 32);
+#if MDBX_LOCKING > 0
+  STATIC_ASSERT(offsetof(lck_t, wrt_lock) % MDBX_CACHELINE_SIZE == 0);
+  STATIC_ASSERT(offsetof(lck_t, rdt_lock) % MDBX_CACHELINE_SIZE == 0);
+#else
+  STATIC_ASSERT(offsetof(lck_t, cached_oldest) % MDBX_CACHELINE_SIZE == 0);
+  STATIC_ASSERT(offsetof(lck_t, rdt_length) % MDBX_CACHELINE_SIZE == 0);
+#endif /* MDBX_LOCKING */
+#if FLEXIBLE_ARRAY_MEMBERS
+  STATIC_ASSERT(offsetof(lck_t, rdt) % MDBX_CACHELINE_SIZE == 0);
+#endif /* FLEXIBLE_ARRAY_MEMBERS */
+
+#if FLEXIBLE_ARRAY_MEMBERS
+  STATIC_ASSERT(NODESIZE == offsetof(node_t, payload));
+  STATIC_ASSERT(PAGEHDRSZ == offsetof(page_t, entries));
+#endif /* FLEXIBLE_ARRAY_MEMBERS */
+  STATIC_ASSERT(sizeof(clc_t) == 3 * sizeof(void *));
+  STATIC_ASSERT(sizeof(kvx_t) == 8 * sizeof(void *));
+
+#define KVX_SIZE_LN2 MDBX_WORDBITS_LN2
+  STATIC_ASSERT(sizeof(kvx_t) == (1u << KVX_SIZE_LN2));
+}
+#endif /* Disabled for MSVC 19.0 (VisualStudio 2015) */
+
+/******************************************************************************/
+
+#include "node.h"
+
+#include "dbi.h"
+
+#include "cogs.h"
+
+#include "cursor.h"
+
+#include "dpl.h"
+
+#include "gc.h"
+
+#include "lck.h"
+
+#include "meta.h"
+
+#include "page-iov.h"
+
+#include "spill.h"
+
+#include "page-ops.h"
+
+#include "tls.h"
+
+#include "walk.h"
+
+#include "sort.h"

src/layout-dxb.h

@@ -0,0 +1,288 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \note Please refer to the COPYRIGHT file for explanations license change,
+/// credits and acknowledgments.
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#pragma once
+
+#include "essentials.h"
+
+#pragma pack(push, 4)
+
+/* A stamp that identifies a file as an MDBX file.
+ * There's nothing special about this value other than that it is easily
+ * recognizable, and it will reflect any byte order mismatches. */
+#define MDBX_MAGIC UINT64_C(/* 56-bit prime */ 0x59659DBDEF4C11)
+
+/* FROZEN: The version number for a database's datafile format. */
+#define MDBX_DATA_VERSION 3
+
+#define MDBX_DATA_MAGIC ((MDBX_MAGIC << 8) + MDBX_PNL_ASCENDING * 64 + MDBX_DATA_VERSION)
+#define MDBX_DATA_MAGIC_LEGACY_COMPAT ((MDBX_MAGIC << 8) + MDBX_PNL_ASCENDING * 64 + 2)
+#define MDBX_DATA_MAGIC_LEGACY_DEVEL ((MDBX_MAGIC << 8) + 255)
+
+/* handle for the DB used to track free pages. */
+#define FREE_DBI 0
+/* handle for the default DB. */
+#define MAIN_DBI 1
+/* Number of DBs in metapage (free and main) - also hardcoded elsewhere */
+#define CORE_DBS 2
+
+/* Number of meta pages - also hardcoded elsewhere */
+#define NUM_METAS 3
+
+/* A page number in the database.
+ *
+ * MDBX uses 32 bit for page numbers. This limits database
+ * size up to 2^44 bytes, in case of 4K pages. */
+typedef uint32_t pgno_t;
+typedef mdbx_atomic_uint32_t atomic_pgno_t;
+#define PRIaPGNO PRIu32
+#define MAX_PAGENO UINT32_C(0x7FFFffff)
+#define MIN_PAGENO NUM_METAS
+
+/* An invalid page number.
+ * Mainly used to denote an empty tree. */
+#define P_INVALID (~(pgno_t)0)
+
+/* A transaction ID. */
+typedef uint64_t txnid_t;
+typedef mdbx_atomic_uint64_t atomic_txnid_t;
+#define PRIaTXN PRIi64
+#define MIN_TXNID UINT64_C(1)
+#define MAX_TXNID (SAFE64_INVALID_THRESHOLD - 1)
+#define INITIAL_TXNID (MIN_TXNID + NUM_METAS - 1)
+#define INVALID_TXNID UINT64_MAX
+
+/* Used for offsets within a single page. */
+typedef uint16_t indx_t;
+
+typedef struct tree {
+  uint16_t flags;       /* see mdbx_dbi_open */
+  uint16_t height;      /* height of this tree */
+  uint32_t dupfix_size; /* key-size for MDBX_DUPFIXED (DUPFIX pages) */
+  pgno_t root;          /* the root page of this tree */
+  pgno_t branch_pages;  /* number of branch pages */
+  pgno_t leaf_pages;    /* number of leaf pages */
+  pgno_t large_pages;   /* number of large pages */
+  uint64_t sequence;    /* table sequence counter */
+  uint64_t items;       /* number of data items */
+  uint64_t mod_txnid;   /* txnid of last committed modification */
+} tree_t;
+
+/* database size-related parameters */
+typedef struct geo {
+  uint16_t grow_pv;   /* datafile growth step as a 16-bit packed (exponential
+                           quantized) value */
+  uint16_t shrink_pv; /* datafile shrink threshold as a 16-bit packed
+                           (exponential quantized) value */
+  pgno_t lower;       /* minimal size of datafile in pages */
+  pgno_t upper;       /* maximal size of datafile in pages */
+  union {
+    pgno_t now; /* current size of datafile in pages */
+    pgno_t end_pgno;
+  };
+  union {
+    pgno_t first_unallocated; /* first unused page in the datafile,
+                         but actually the file may be shorter. */
+    pgno_t next_pgno;
+  };
+} geo_t;
+
+/* Meta page content.
+ * A meta page is the start point for accessing a database snapshot.
+ * Pages 0-2 are meta pages. */
+typedef struct meta {
+  /* Stamp identifying this as an MDBX file.
+   * It must be set to MDBX_MAGIC with MDBX_DATA_VERSION. */
+  uint32_t magic_and_version[2];
+
+  /* txnid that committed this meta, the first of a two-phase-update pair */
+  union {
+    mdbx_atomic_uint32_t txnid_a[2];
+    uint64_t unsafe_txnid;
+  };
+
+  uint16_t reserve16;   /* extra flags, zero (nothing) for now */
+  uint8_t validator_id; /* ID of checksum and page validation method,
+                         * zero (nothing) for now */
+  int8_t extra_pagehdr; /* extra bytes in the page header,
+                         * zero (nothing) for now */
+
+  geo_t geometry; /* database size-related parameters */
+
+  union {
+    struct {
+      tree_t gc, main;
+    } trees;
+    __anonymous_struct_extension__ struct {
+      uint16_t gc_flags;
+      uint16_t gc_height;
+      uint32_t pagesize;
+    };
+  };
+
+  MDBX_canary canary;
+
+#define DATASIGN_NONE 0u
+#define DATASIGN_WEAK 1u
+#define SIGN_IS_STEADY(sign) ((sign) > DATASIGN_WEAK)
+  union {
+    uint32_t sign[2];
+    uint64_t unsafe_sign;
+  };
+
+  /* txnid that committed this meta, the second of a two-phase-update pair */
+  mdbx_atomic_uint32_t txnid_b[2];
+
+  /* Number of non-meta pages which were put in GC after COW. May be 0 in case
+   * DB was previously handled by libmdbx without corresponding feature.
+   * This value in couple with reader.snapshot_pages_retired allows fast
+   * estimation of "how much reader is restraining GC recycling". */
+  uint32_t pages_retired[2];
+
+  /* The analogue /proc/sys/kernel/random/boot_id or similar to determine
+   * whether the system was rebooted after the last use of the database files.
+   * If there was no reboot, but there is no need to rollback to the last
+   * steady sync point. Zeros mean that no relevant information is available
+   * from the system. */
+  bin128_t bootid;
+
+  /* GUID базы данных, начиная с v0.13.1 */
+  bin128_t dxbid;
+} meta_t;
+
+#pragma pack(1)
+
+typedef enum page_type {
+  P_BRANCH = 0x01u /* branch page */,
+  P_LEAF = 0x02u /* leaf page */,
+  P_LARGE = 0x04u /* large/overflow page */,
+  P_META = 0x08u /* meta page */,
+  P_LEGACY_DIRTY = 0x10u /* legacy P_DIRTY flag prior to v0.10 958fd5b9 */,
+  P_BAD = P_LEGACY_DIRTY /* explicit flag for invalid/bad page */,
+  P_DUPFIX = 0x20u /* for MDBX_DUPFIXED records */,
+  P_SUBP = 0x40u /* for MDBX_DUPSORT sub-pages */,
+  P_SPILLED = 0x2000u /* spilled in parent txn */,
+  P_LOOSE = 0x4000u /* page was dirtied then freed, can be reused */,
+  P_FROZEN = 0x8000u /* used for retire page with known status */,
+  P_ILL_BITS = (uint16_t)~(P_BRANCH | P_LEAF | P_DUPFIX | P_LARGE | P_SPILLED),
+
+  page_broken = 0,
+  page_large = P_LARGE,
+  page_branch = P_BRANCH,
+  page_leaf = P_LEAF,
+  page_dupfix_leaf = P_DUPFIX,
+  page_sub_leaf = P_SUBP | P_LEAF,
+  page_sub_dupfix_leaf = P_SUBP | P_DUPFIX,
+  page_sub_broken = P_SUBP,
+} page_type_t;
+
+/* Common header for all page types. The page type depends on flags.
+ *
+ * P_BRANCH and P_LEAF pages have unsorted 'node_t's at the end, with
+ * sorted entries[] entries referring to them. Exception: P_DUPFIX pages
+ * omit entries and pack sorted MDBX_DUPFIXED values after the page header.
+ *
+ * P_LARGE records occupy one or more contiguous pages where only the
+ * first has a page header. They hold the real data of N_BIG nodes.
+ *
+ * P_SUBP sub-pages are small leaf "pages" with duplicate data.
+ * A node with flag N_DUP but not N_TREE contains a sub-page.
+ * (Duplicate data can also go in tables, which use normal pages.)
+ *
+ * P_META pages contain meta_t, the start point of an MDBX snapshot.
+ *
+ * Each non-metapage up to meta_t.mm_last_pg is reachable exactly once
+ * in the snapshot: Either used by a database or listed in a GC record. */
+typedef struct page {
+  uint64_t txnid;        /* txnid which created page, maybe zero in legacy DB */
+  uint16_t dupfix_ksize; /* key size if this is a DUPFIX page */
+  uint16_t flags;
+  union {
+    uint32_t pages; /* number of overflow pages */
+    __anonymous_struct_extension__ struct {
+      indx_t lower; /* lower bound of free space */
+      indx_t upper; /* upper bound of free space */
+    };
+  };
+  pgno_t pgno; /* page number */
+
+#if FLEXIBLE_ARRAY_MEMBERS
+  indx_t entries[] /* dynamic size */;
+#endif /* FLEXIBLE_ARRAY_MEMBERS */
+} page_t;
+
+/* Size of the page header, excluding dynamic data at the end */
+#define PAGEHDRSZ 20u
+
+/* Header for a single key/data pair within a page.
+ * Used in pages of type P_BRANCH and P_LEAF without P_DUPFIX.
+ * We guarantee 2-byte alignment for 'node_t's.
+ *
+ * Leaf node flags describe node contents.  N_BIG says the node's
+ * data part is the page number of an overflow page with actual data.
+ * N_DUP and N_TREE can be combined giving duplicate data in
+ * a sub-page/table, and named databases (just N_TREE). */
+typedef struct node {
+#if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
+  union {
+    uint32_t dsize;
+    uint32_t child_pgno;
+  };
+  uint8_t flags; /* see node_flags */
+  uint8_t extra;
+  uint16_t ksize; /* key size */
+#else
+  uint16_t ksize; /* key size */
+  uint8_t extra;
+  uint8_t flags; /* see node_flags */
+  union {
+    uint32_t child_pgno;
+    uint32_t dsize;
+  };
+#endif /* __BYTE_ORDER__ */
+
+#if FLEXIBLE_ARRAY_MEMBERS
+  uint8_t payload[] /* key and data are appended here */;
+#endif /* FLEXIBLE_ARRAY_MEMBERS */
+} node_t;
+
+/* Size of the node header, excluding dynamic data at the end */
+#define NODESIZE 8u
+
+typedef enum node_flags {
+  N_BIG = 0x01 /* data put on large page */,
+  N_TREE = 0x02 /* data is a b-tree */,
+  N_DUP = 0x04 /* data has duplicates */
+} node_flags_t;
+
+#pragma pack(pop)
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline uint8_t page_type(const page_t *mp) { return mp->flags; }
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline uint8_t page_type_compat(const page_t *mp) {
+  /* Drop legacy P_DIRTY flag for sub-pages for compatilibity,
+   * for assertions only. */
+  return unlikely(mp->flags & P_SUBP) ? mp->flags & ~(P_SUBP | P_LEGACY_DIRTY) : mp->flags;
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool is_leaf(const page_t *mp) {
+  return (mp->flags & P_LEAF) != 0;
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool is_dupfix_leaf(const page_t *mp) {
+  return (mp->flags & P_DUPFIX) != 0;
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool is_branch(const page_t *mp) {
+  return (mp->flags & P_BRANCH) != 0;
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool is_largepage(const page_t *mp) {
+  return (mp->flags & P_LARGE) != 0;
+}
+
+MDBX_MAYBE_UNUSED MDBX_NOTHROW_PURE_FUNCTION static inline bool is_subpage(const page_t *mp) {
+  return (mp->flags & P_SUBP) != 0;
+}

src/layout-lck.h

@@ -0,0 +1,289 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \note Please refer to the COPYRIGHT file for explanations license change,
+/// credits and acknowledgments.
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#pragma once
+
+#include "essentials.h"
+
+/* The version number for a database's lockfile format. */
+#define MDBX_LOCK_VERSION 6
+
+#if MDBX_LOCKING == MDBX_LOCKING_WIN32FILES
+
+#define MDBX_LCK_SIGN UINT32_C(0xF10C)
+typedef void osal_ipclock_t;
+#elif MDBX_LOCKING == MDBX_LOCKING_SYSV
+
+#define MDBX_LCK_SIGN UINT32_C(0xF18D)
+typedef mdbx_pid_t osal_ipclock_t;
+
+#elif MDBX_LOCKING == MDBX_LOCKING_POSIX2001 || MDBX_LOCKING == MDBX_LOCKING_POSIX2008
+
+#define MDBX_LCK_SIGN UINT32_C(0x8017)
+typedef pthread_mutex_t osal_ipclock_t;
+
+#elif MDBX_LOCKING == MDBX_LOCKING_POSIX1988
+
+#define MDBX_LCK_SIGN UINT32_C(0xFC29)
+typedef sem_t osal_ipclock_t;
+
+#else
+#error "FIXME"
+#endif /* MDBX_LOCKING */
+
+/* Статистика профилирования работы GC */
+typedef struct gc_prof_stat {
+  /* Монотонное время по "настенным часам"
+   * затраченное на чтение и поиск внутри GC */
+  uint64_t rtime_monotonic;
+  /* Процессорное время в режим пользователя
+   * на подготовку страниц извлекаемых из GC, включая подкачку с диска. */
+  uint64_t xtime_cpu;
+  /* Количество итераций чтения-поиска внутри GC при выделении страниц */
+  uint32_t rsteps;
+  /* Количество запросов на выделение последовательностей страниц,
+   * т.е. когда запрашивает выделение больше одной страницы */
+  uint32_t xpages;
+  /* Счетчик выполнения по медленному пути (slow path execution count) */
+  uint32_t spe_counter;
+  /* page faults (hard page faults) */
+  uint32_t majflt;
+  /* Для разборок с pnl_merge() */
+  struct {
+    uint64_t time;
+    uint64_t volume;
+    uint32_t calls;
+  } pnl_merge;
+} gc_prof_stat_t;
+
+/* Statistics of pages operations for all transactions,
+ * including incomplete and aborted. */
+typedef struct pgops {
+  mdbx_atomic_uint64_t newly;   /* Quantity of a new pages added */
+  mdbx_atomic_uint64_t cow;     /* Quantity of pages copied for update */
+  mdbx_atomic_uint64_t clone;   /* Quantity of parent's dirty pages clones
+                                   for nested transactions */
+  mdbx_atomic_uint64_t split;   /* Page splits */
+  mdbx_atomic_uint64_t merge;   /* Page merges */
+  mdbx_atomic_uint64_t spill;   /* Quantity of spilled dirty pages */
+  mdbx_atomic_uint64_t unspill; /* Quantity of unspilled/reloaded pages */
+  mdbx_atomic_uint64_t wops;    /* Number of explicit write operations (not a pages) to a disk */
+  mdbx_atomic_uint64_t msync;   /* Number of explicit msync/flush-to-disk operations */
+  mdbx_atomic_uint64_t fsync;   /* Number of explicit fsync/flush-to-disk operations */
+
+  mdbx_atomic_uint64_t prefault; /* Number of prefault write operations */
+  mdbx_atomic_uint64_t mincore;  /* Number of mincore() calls */
+
+  mdbx_atomic_uint32_t incoherence; /* number of https://libmdbx.dqdkfa.ru/dead-github/issues/269
+                                       caught */
+  mdbx_atomic_uint32_t reserved;
+
+  /* Статистика для профилирования GC.
+   * Логически эти данные, возможно, стоит вынести в другую структуру,
+   * но разница будет сугубо косметическая. */
+  struct {
+    /* Затраты на поддержку данных пользователя */
+    gc_prof_stat_t work;
+    /* Затраты на поддержку и обновления самой GC */
+    gc_prof_stat_t self;
+    /* Итераций обновления GC,
+     * больше 1 если были повторы/перезапуски */
+    uint32_t wloops;
+    /* Итерации слияния записей GC */
+    uint32_t coalescences;
+    /* Уничтожения steady-точек фиксации в MDBX_UTTERLY_NOSYNC */
+    uint32_t wipes;
+    /* Сбросы данные на диск вне MDBX_UTTERLY_NOSYNC */
+    uint32_t flushes;
+    /* Попытки пнуть тормозящих читателей */
+    uint32_t kicks;
+  } gc_prof;
+} pgop_stat_t;
+
+/* Reader Lock Table
+ *
+ * Readers don't acquire any locks for their data access. Instead, they
+ * simply record their transaction ID in the reader table. The reader
+ * mutex is needed just to find an empty slot in the reader table. The
+ * slot's address is saved in thread-specific data so that subsequent
+ * read transactions started by the same thread need no further locking to
+ * proceed.
+ *
+ * If MDBX_NOSTICKYTHREADS is set, the slot address is not saved in
+ * thread-specific data. No reader table is used if the database is on a
+ * read-only filesystem.
+ *
+ * Since the database uses multi-version concurrency control, readers don't
+ * actually need any locking. This table is used to keep track of which
+ * readers are using data from which old transactions, so that we'll know
+ * when a particular old transaction is no longer in use. Old transactions
+ * that have discarded any data pages can then have those pages reclaimed
+ * for use by a later write transaction.
+ *
+ * The lock table is constructed such that reader slots are aligned with the
+ * processor's cache line size. Any slot is only ever used by one thread.
+ * This alignment guarantees that there will be no contention or cache
+ * thrashing as threads update their own slot info, and also eliminates
+ * any need for locking when accessing a slot.
+ *
+ * A writer thread will scan every slot in the table to determine the oldest
+ * outstanding reader transaction. Any freed pages older than this will be
+ * reclaimed by the writer. The writer doesn't use any locks when scanning
+ * this table. This means that there's no guarantee that the writer will
+ * see the most up-to-date reader info, but that's not required for correct
+ * operation - all we need is to know the upper bound on the oldest reader,
+ * we don't care at all about the newest reader. So the only consequence of
+ * reading stale information here is that old pages might hang around a
+ * while longer before being reclaimed. That's actually good anyway, because
+ * the longer we delay reclaiming old pages, the more likely it is that a
+ * string of contiguous pages can be found after coalescing old pages from
+ * many old transactions together. */
+
+/* The actual reader record, with cacheline padding. */
+typedef struct reader_slot {
+  /* Current Transaction ID when this transaction began, or INVALID_TXNID.
+   * Multiple readers that start at the same time will probably have the
+   * same ID here. Again, it's not important to exclude them from
+   * anything; all we need to know is which version of the DB they
+   * started from so we can avoid overwriting any data used in that
+   * particular version. */
+  atomic_txnid_t txnid;
+
+  /* The information we store in a single slot of the reader table.
+   * In addition to a transaction ID, we also record the process and
+   * thread ID that owns a slot, so that we can detect stale information,
+   * e.g. threads or processes that went away without cleaning up.
+   *
+   * NOTE: We currently don't check for stale records.
+   * We simply re-init the table when we know that we're the only process
+   * opening the lock file. */
+
+  /* Псевдо thread_id для пометки вытесненных читающих транзакций. */
+#define MDBX_TID_TXN_OUSTED (UINT64_MAX - 1)
+
+  /* Псевдо thread_id для пометки припаркованных читающих транзакций. */
+#define MDBX_TID_TXN_PARKED UINT64_MAX
+
+  /* The thread ID of the thread owning this txn. */
+  mdbx_atomic_uint64_t tid;
+
+  /* The process ID of the process owning this reader txn. */
+  mdbx_atomic_uint32_t pid;
+
+  /* The number of pages used in the reader's MVCC snapshot,
+   * i.e. the value of meta->geometry.first_unallocated and
+   * txn->geo.first_unallocated */
+  atomic_pgno_t snapshot_pages_used;
+  /* Number of retired pages at the time this reader starts transaction. So,
+   * at any time the difference meta.pages_retired -
+   * reader.snapshot_pages_retired will give the number of pages which this
+   * reader restraining from reuse. */
+  mdbx_atomic_uint64_t snapshot_pages_retired;
+} reader_slot_t;
+
+/* The header for the reader table (a memory-mapped lock file). */
+typedef struct shared_lck {
+  /* Stamp identifying this as an MDBX file.
+   * It must be set to MDBX_MAGIC with MDBX_LOCK_VERSION. */
+  uint64_t magic_and_version;
+
+  /* Format of this lock file. Must be set to MDBX_LOCK_FORMAT. */
+  uint32_t os_and_format;
+
+  /* Flags which environment was opened. */
+  mdbx_atomic_uint32_t envmode;
+
+  /* Threshold of un-synced-with-disk pages for auto-sync feature,
+   * zero means no-threshold, i.e. auto-sync is disabled. */
+  atomic_pgno_t autosync_threshold;
+
+  /* Low 32-bit of txnid with which meta-pages was synced,
+   * i.e. for sync-polling in the MDBX_NOMETASYNC mode. */
+#define MDBX_NOMETASYNC_LAZY_UNK (UINT32_MAX / 3)
+#define MDBX_NOMETASYNC_LAZY_FD (MDBX_NOMETASYNC_LAZY_UNK + UINT32_MAX / 8)
+#define MDBX_NOMETASYNC_LAZY_WRITEMAP (MDBX_NOMETASYNC_LAZY_UNK - UINT32_MAX / 8)
+  mdbx_atomic_uint32_t meta_sync_txnid;
+
+  /* Period for timed auto-sync feature, i.e. at the every steady checkpoint
+   * the mti_unsynced_timeout sets to the current_time + autosync_period.
+   * The time value is represented in a suitable system-dependent form, for
+   * example clock_gettime(CLOCK_BOOTTIME) or clock_gettime(CLOCK_MONOTONIC).
+   * Zero means timed auto-sync is disabled. */
+  mdbx_atomic_uint64_t autosync_period;
+
+  /* Marker to distinguish uniqueness of DB/CLK. */
+  mdbx_atomic_uint64_t bait_uniqueness;
+
+  /* Paired counter of processes that have mlock()ed part of mmapped DB.
+   * The (mlcnt[0] - mlcnt[1]) > 0 means at least one process
+   * lock at least one page, so therefore madvise() could return EINVAL. */
+  mdbx_atomic_uint32_t mlcnt[2];
+
+  MDBX_ALIGNAS(MDBX_CACHELINE_SIZE) /* cacheline ----------------------------*/
+
+  /* Statistics of costly ops of all (running, completed and aborted)
+   * transactions */
+  pgop_stat_t pgops;
+
+  MDBX_ALIGNAS(MDBX_CACHELINE_SIZE) /* cacheline ----------------------------*/
+
+#if MDBX_LOCKING > 0
+  /* Write transaction lock. */
+  osal_ipclock_t wrt_lock;
+#endif /* MDBX_LOCKING > 0 */
+
+  atomic_txnid_t cached_oldest;
+
+  /* Timestamp of entering an out-of-sync state. Value is represented in a
+   * suitable system-dependent form, for example clock_gettime(CLOCK_BOOTTIME)
+   * or clock_gettime(CLOCK_MONOTONIC). */
+  mdbx_atomic_uint64_t eoos_timestamp;
+
+  /* Number un-synced-with-disk pages for auto-sync feature. */
+  mdbx_atomic_uint64_t unsynced_pages;
+
+  /* Timestamp of the last readers check. */
+  mdbx_atomic_uint64_t readers_check_timestamp;
+
+  /* Number of page which was discarded last time by madvise(DONTNEED). */
+  atomic_pgno_t discarded_tail;
+
+  /* Shared anchor for tracking readahead edge and enabled/disabled status. */
+  pgno_t readahead_anchor;
+
+  /* Shared cache for mincore() results */
+  struct {
+    pgno_t begin[4];
+    uint64_t mask[4];
+  } mincore_cache;
+
+  MDBX_ALIGNAS(MDBX_CACHELINE_SIZE) /* cacheline ----------------------------*/
+
+#if MDBX_LOCKING > 0
+  /* Readeaders table lock. */
+  osal_ipclock_t rdt_lock;
+#endif /* MDBX_LOCKING > 0 */
+
+  /* The number of slots that have been used in the reader table.
+   * This always records the maximum count, it is not decremented
+   * when readers release their slots. */
+  mdbx_atomic_uint32_t rdt_length;
+  mdbx_atomic_uint32_t rdt_refresh_flag;
+
+#if FLEXIBLE_ARRAY_MEMBERS
+  MDBX_ALIGNAS(MDBX_CACHELINE_SIZE) /* cacheline ----------------------------*/
+  reader_slot_t rdt[] /* dynamic size */;
+
+/* Lockfile format signature: version, features and field layout */
+#define MDBX_LOCK_FORMAT                                                                                               \
+  (MDBX_LCK_SIGN * 27733 + (unsigned)sizeof(reader_slot_t) * 13 +                                                      \
+   (unsigned)offsetof(reader_slot_t, snapshot_pages_used) * 251 + (unsigned)offsetof(lck_t, cached_oldest) * 83 +      \
+   (unsigned)offsetof(lck_t, rdt_length) * 37 + (unsigned)offsetof(lck_t, rdt) * 29)
+#endif /* FLEXIBLE_ARRAY_MEMBERS */
+} lck_t;
+
+#define MDBX_LOCK_MAGIC ((MDBX_MAGIC << 8) + MDBX_LOCK_VERSION)
+
+#define MDBX_READERS_LIMIT 32767

src/lck.c

@@ -0,0 +1,174 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+__cold static int lck_setup_locked(MDBX_env *env) {
+  int err = rthc_register(env);
+  if (unlikely(err != MDBX_SUCCESS))
+    return err;
+
+  int lck_seize_rc = lck_seize(env);
+  if (unlikely(MDBX_IS_ERROR(lck_seize_rc)))
+    return lck_seize_rc;
+
+  if (env->lck_mmap.fd == INVALID_HANDLE_VALUE) {
+    env->lck = lckless_stub(env);
+    env->max_readers = UINT_MAX;
+    DEBUG("lck-setup:%s%s%s", " lck-less", (env->flags & MDBX_RDONLY) ? " readonly" : "",
+          (lck_seize_rc == MDBX_RESULT_TRUE) ? " exclusive" : " cooperative");
+    return lck_seize_rc;
+  }
+
+  DEBUG("lck-setup:%s%s%s", " with-lck", (env->flags & MDBX_RDONLY) ? " readonly" : "",
+        (lck_seize_rc == MDBX_RESULT_TRUE) ? " exclusive" : " cooperative");
+
+  MDBX_env *inprocess_neighbor = nullptr;
+  err = rthc_uniq_check(&env->lck_mmap, &inprocess_neighbor);
+  if (unlikely(MDBX_IS_ERROR(err)))
+    return err;
+  if (inprocess_neighbor) {
+    if ((globals.runtime_flags & MDBX_DBG_LEGACY_MULTIOPEN) == 0 || (inprocess_neighbor->flags & MDBX_EXCLUSIVE) != 0)
+      return MDBX_BUSY;
+    if (lck_seize_rc == MDBX_RESULT_TRUE) {
+      err = lck_downgrade(env);
+      if (unlikely(err != MDBX_SUCCESS))
+        return err;
+      lck_seize_rc = MDBX_RESULT_FALSE;
+    }
+  }
+
+  uint64_t size = 0;
+  err = osal_filesize(env->lck_mmap.fd, &size);
+  if (unlikely(err != MDBX_SUCCESS))
+    return err;
+
+  if (lck_seize_rc == MDBX_RESULT_TRUE) {
+    size = ceil_powerof2(env->max_readers * sizeof(reader_slot_t) + sizeof(lck_t), globals.sys_pagesize);
+    jitter4testing(false);
+  } else {
+    if (env->flags & MDBX_EXCLUSIVE)
+      return MDBX_BUSY;
+    if (size > INT_MAX || (size & (globals.sys_pagesize - 1)) != 0 || size < globals.sys_pagesize) {
+      ERROR("lck-file has invalid size %" PRIu64 " bytes", size);
+      return MDBX_PROBLEM;
+    }
+  }
+
+  const size_t maxreaders = ((size_t)size - sizeof(lck_t)) / sizeof(reader_slot_t);
+  if (maxreaders < 4) {
+    ERROR("lck-size too small (up to %" PRIuPTR " readers)", maxreaders);
+    return MDBX_PROBLEM;
+  }
+  env->max_readers = (maxreaders <= MDBX_READERS_LIMIT) ? (unsigned)maxreaders : (unsigned)MDBX_READERS_LIMIT;
+
+  err =
+      osal_mmap((env->flags & MDBX_EXCLUSIVE) | MDBX_WRITEMAP, &env->lck_mmap, (size_t)size, (size_t)size,
+                lck_seize_rc ? MMAP_OPTION_TRUNCATE | MMAP_OPTION_SEMAPHORE : MMAP_OPTION_SEMAPHORE, env->pathname.lck);
+  if (unlikely(err != MDBX_SUCCESS))
+    return err;
+
+#ifdef MADV_DODUMP
+  err = madvise(env->lck_mmap.lck, size, MADV_DODUMP) ? ignore_enosys_and_eagain(errno) : MDBX_SUCCESS;
+  if (unlikely(MDBX_IS_ERROR(err)))
+    return err;
+#endif /* MADV_DODUMP */
+
+#ifdef MADV_WILLNEED
+  err = madvise(env->lck_mmap.lck, size, MADV_WILLNEED) ? ignore_enosys_and_eagain(errno) : MDBX_SUCCESS;
+  if (unlikely(MDBX_IS_ERROR(err)))
+    return err;
+#elif defined(POSIX_MADV_WILLNEED)
+  err = ignore_enosys(posix_madvise(env->lck_mmap.lck, size, POSIX_MADV_WILLNEED));
+  if (unlikely(MDBX_IS_ERROR(err)))
+    return err;
+#endif /* MADV_WILLNEED */
+
+  lck_t *lck = env->lck_mmap.lck;
+  if (lck_seize_rc == MDBX_RESULT_TRUE) {
+    /* If we succeed got exclusive lock, then nobody is using the lock region
+     * and we should initialize it. */
+    memset(lck, 0, (size_t)size);
+    jitter4testing(false);
+    lck->magic_and_version = MDBX_LOCK_MAGIC;
+    lck->os_and_format = MDBX_LOCK_FORMAT;
+#if MDBX_ENABLE_PGOP_STAT
+    lck->pgops.wops.weak = 1;
+#endif /* MDBX_ENABLE_PGOP_STAT */
+    err = osal_msync(&env->lck_mmap, 0, (size_t)size, MDBX_SYNC_DATA | MDBX_SYNC_SIZE);
+    if (unlikely(err != MDBX_SUCCESS)) {
+      ERROR("initial-%s for lck-file failed, err %d", "msync/fsync", err);
+      eASSERT(env, MDBX_IS_ERROR(err));
+      return err;
+    }
+  } else {
+    if (lck->magic_and_version != MDBX_LOCK_MAGIC) {
+      const bool invalid = (lck->magic_and_version >> 8) != MDBX_MAGIC;
+      ERROR("lock region has %s", invalid ? "invalid magic"
+                                          : "incompatible version (only applications with nearly or the "
+                                            "same versions of libmdbx can share the same database)");
+      return invalid ? MDBX_INVALID : MDBX_VERSION_MISMATCH;
+    }
+    if (lck->os_and_format != MDBX_LOCK_FORMAT) {
+      ERROR("lock region has os/format signature 0x%" PRIx32 ", expected 0x%" PRIx32, lck->os_and_format,
+            MDBX_LOCK_FORMAT);
+      return MDBX_VERSION_MISMATCH;
+    }
+  }
+
+  err = lck_init(env, inprocess_neighbor, lck_seize_rc);
+  if (unlikely(err != MDBX_SUCCESS)) {
+    eASSERT(env, MDBX_IS_ERROR(err));
+    return err;
+  }
+
+  env->lck = lck;
+  eASSERT(env, !MDBX_IS_ERROR(lck_seize_rc));
+  return lck_seize_rc;
+}
+
+__cold int lck_setup(MDBX_env *env, mdbx_mode_t mode) {
+  eASSERT(env, env->lazy_fd != INVALID_HANDLE_VALUE);
+  eASSERT(env, env->lck_mmap.fd == INVALID_HANDLE_VALUE);
+
+  int err = osal_openfile(MDBX_OPEN_LCK, env, env->pathname.lck, &env->lck_mmap.fd, mode);
+  if (err != MDBX_SUCCESS) {
+    switch (err) {
+    case MDBX_EACCESS:
+    case MDBX_EPERM:
+      if (F_ISSET(env->flags, MDBX_RDONLY | MDBX_EXCLUSIVE))
+        break;
+      __fallthrough /* fall through */;
+    case MDBX_ENOFILE:
+    case MDBX_EROFS:
+      if (env->flags & MDBX_RDONLY) {
+        /* ENSURE the file system is read-only */
+        int err_rofs = osal_check_fs_rdonly(env->lazy_fd, env->pathname.lck, err);
+        if (err_rofs == MDBX_SUCCESS ||
+            /* ignore ERROR_NOT_SUPPORTED for exclusive mode */
+            (err_rofs == MDBX_ENOSYS && (env->flags & MDBX_EXCLUSIVE)))
+          break;
+        if (err_rofs != MDBX_ENOSYS)
+          err = err_rofs;
+      }
+      __fallthrough /* fall through */;
+    default:
+      ERROR("unable to open lck-file %" MDBX_PRIsPATH ", env-flags 0x%X, err %d", env->pathname.lck, env->flags, err);
+      return err;
+    }
+
+    /* LY: without-lck mode (e.g. exclusive or on read-only filesystem) */
+    env->lck_mmap.fd = INVALID_HANDLE_VALUE;
+    NOTICE("continue %" MDBX_PRIsPATH " within without-lck mode, env-flags 0x%X, lck-error %d", env->pathname.dxb,
+           env->flags, err);
+  }
+
+  rthc_lock();
+  err = lck_setup_locked(env);
+  rthc_unlock();
+  return err;
+}
+
+void mincore_clean_cache(const MDBX_env *const env) {
+  memset(env->lck->mincore_cache.begin, -1, sizeof(env->lck->mincore_cache.begin));
+}

src/lck.h

@@ -0,0 +1,110 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#pragma once
+
+#include "essentials.h"
+
+MDBX_INTERNAL int lck_setup(MDBX_env *env, mdbx_mode_t mode);
+#if MDBX_LOCKING > MDBX_LOCKING_SYSV
+MDBX_INTERNAL int lck_ipclock_stubinit(osal_ipclock_t *ipc);
+MDBX_INTERNAL int lck_ipclock_destroy(osal_ipclock_t *ipc);
+#endif /* MDBX_LOCKING > MDBX_LOCKING_SYSV */
+
+/// \brief Initialization of synchronization primitives linked with MDBX_env
+///   instance both in LCK-file and within the current process.
+/// \param
+///   global_uniqueness_flag = true - denotes that there are no other processes
+///     working with DB and LCK-file. Thus the function MUST initialize
+///     shared synchronization objects in memory-mapped LCK-file.
+///   global_uniqueness_flag = false - denotes that at least one process is
+///     already working with DB and LCK-file, including the case when DB
+///     has already been opened in the current process. Thus the function
+///     MUST NOT initialize shared synchronization objects in memory-mapped
+///     LCK-file that are already in use.
+/// \return Error code or zero on success.
+MDBX_INTERNAL int lck_init(MDBX_env *env, MDBX_env *inprocess_neighbor, int global_uniqueness_flag);
+
+/// \brief Disconnects from shared interprocess objects and destructs
+///   synchronization objects linked with MDBX_env instance
+///   within the current process.
+/// \param
+///   inprocess_neighbor = nullptr - if the current process does not have other
+///     instances of MDBX_env linked with the DB being closed.
+///     Thus the function MUST check for other processes working with DB or
+///     LCK-file, and keep or destroy shared synchronization objects in
+///     memory-mapped LCK-file depending on the result.
+///   inprocess_neighbor = not-nullptr - pointer to another instance of MDBX_env
+///     (anyone of there is several) working with DB or LCK-file within the
+///     current process. Thus the function MUST NOT try to acquire exclusive
+///     lock and/or try to destruct shared synchronization objects linked with
+///     DB or LCK-file. Moreover, the implementation MUST ensure correct work
+///     of other instances of MDBX_env within the current process, e.g.
+///     restore POSIX-fcntl locks after the closing of file descriptors.
+/// \return Error code (MDBX_PANIC) or zero on success.
+MDBX_INTERNAL int lck_destroy(MDBX_env *env, MDBX_env *inprocess_neighbor, const uint32_t current_pid);
+
+/// \brief Connects to shared interprocess locking objects and tries to acquire
+///   the maximum lock level (shared if exclusive is not available)
+///   Depending on implementation or/and platform (Windows) this function may
+///   acquire the non-OS super-level lock (e.g. for shared synchronization
+///   objects initialization), which will be downgraded to OS-exclusive or
+///   shared via explicit calling of lck_downgrade().
+/// \return
+///   MDBX_RESULT_TRUE (-1) - if an exclusive lock was acquired and thus
+///     the current process is the first and only after the last use of DB.
+///   MDBX_RESULT_FALSE (0) - if a shared lock was acquired and thus
+///     DB has already been opened and now is used by other processes.
+///   Otherwise (not 0 and not -1) - error code.
+MDBX_INTERNAL int lck_seize(MDBX_env *env);
+
+/// \brief Downgrades the level of initially acquired lock to
+///   operational level specified by argument. The reason for such downgrade:
+///    - unblocking of other processes that are waiting for access, i.e.
+///      if (env->flags & MDBX_EXCLUSIVE) != 0, then other processes
+///      should be made aware that access is unavailable rather than
+///      wait for it.
+///    - freeing locks that interfere file operation (especially for Windows)
+///   (env->flags & MDBX_EXCLUSIVE) == 0 - downgrade to shared lock.
+///   (env->flags & MDBX_EXCLUSIVE) != 0 - downgrade to exclusive
+///   operational lock.
+/// \return Error code or zero on success
+MDBX_INTERNAL int lck_downgrade(MDBX_env *env);
+
+MDBX_MAYBE_UNUSED MDBX_INTERNAL int lck_upgrade(MDBX_env *env, bool dont_wait);
+
+/// \brief Locks LCK-file or/and table of readers for (de)registering.
+/// \return Error code or zero on success
+MDBX_INTERNAL int lck_rdt_lock(MDBX_env *env);
+
+/// \brief Unlocks LCK-file or/and table of readers after (de)registering.
+MDBX_INTERNAL void lck_rdt_unlock(MDBX_env *env);
+
+/// \brief Acquires write-transaction lock.
+/// \return Error code or zero on success
+MDBX_INTERNAL int lck_txn_lock(MDBX_env *env, bool dont_wait);
+
+/// \brief Releases write-transaction lock..
+MDBX_INTERNAL void lck_txn_unlock(MDBX_env *env);
+
+/// \brief Sets alive-flag of reader presence (indicative lock) for PID of
+///   the current process. The function does no more than needed for
+///   the correct working of lck_rpid_check() in other processes.
+/// \return Error code or zero on success
+MDBX_INTERNAL int lck_rpid_set(MDBX_env *env);
+
+/// \brief Resets alive-flag of reader presence (indicative lock)
+///   for PID of the current process. The function does no more than needed
+///   for the correct working of lck_rpid_check() in other processes.
+/// \return Error code or zero on success
+MDBX_INTERNAL int lck_rpid_clear(MDBX_env *env);
+
+/// \brief Checks for reading process status with the given pid with help of
+///   alive-flag of presence (indicative lock) or using another way.
+/// \return
+///   MDBX_RESULT_TRUE (-1) - if the reader process with the given PID is alive
+///     and working with DB (indicative lock is present).
+///   MDBX_RESULT_FALSE (0) - if the reader process with the given PID is absent
+///     or not working with DB (indicative lock is not present).
+///   Otherwise (not 0 and not -1) - error code.
+MDBX_INTERNAL int lck_rpid_check(MDBX_env *env, uint32_t pid);

src/logging_and_debug.c

@@ -0,0 +1,250 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#include "internals.h"
+
+__cold void debug_log_va(int level, const char *function, int line, const char *fmt, va_list args) {
+  ENSURE(nullptr, osal_fastmutex_acquire(&globals.debug_lock) == 0);
+  if (globals.logger.ptr) {
+    if (globals.logger_buffer == nullptr)
+      globals.logger.fmt(level, function, line, fmt, args);
+    else {
+      const int len = vsnprintf(globals.logger_buffer, globals.logger_buffer_size, fmt, args);
+      if (len > 0)
+        globals.logger.nofmt(level, function, line, globals.logger_buffer, len);
+    }
+  } else {
+#if defined(_WIN32) || defined(_WIN64)
+    if (IsDebuggerPresent()) {
+      int prefix_len = 0;
+      char *prefix = nullptr;
+      if (function && line > 0)
+        prefix_len = osal_asprintf(&prefix, "%s:%d ", function, line);
+      else if (function)
+        prefix_len = osal_asprintf(&prefix, "%s: ", function);
+      else if (line > 0)
+        prefix_len = osal_asprintf(&prefix, "%d: ", line);
+      if (prefix_len > 0 && prefix) {
+        OutputDebugStringA(prefix);
+        osal_free(prefix);
+      }
+      char *msg = nullptr;
+      int msg_len = osal_vasprintf(&msg, fmt, args);
+      if (msg_len > 0 && msg) {
+        OutputDebugStringA(msg);
+        osal_free(msg);
+      }
+    }
+#else
+    if (function && line > 0)
+      fprintf(stderr, "%s:%d ", function, line);
+    else if (function)
+      fprintf(stderr, "%s: ", function);
+    else if (line > 0)
+      fprintf(stderr, "%d: ", line);
+    vfprintf(stderr, fmt, args);
+    fflush(stderr);
+#endif
+  }
+  ENSURE(nullptr, osal_fastmutex_release(&globals.debug_lock) == 0);
+}
+
+__cold void debug_log(int level, const char *function, int line, const char *fmt, ...) {
+  va_list args;
+  va_start(args, fmt);
+  debug_log_va(level, function, line, fmt, args);
+  va_end(args);
+}
+
+__cold void log_error(const int err, const char *func, unsigned line) {
+  assert(err != MDBX_SUCCESS);
+  if (unlikely(globals.loglevel >= MDBX_LOG_DEBUG)) {
+    const bool is_error = err != MDBX_RESULT_TRUE && err != MDBX_NOTFOUND;
+    char buf[256];
+    debug_log(is_error ? MDBX_LOG_ERROR : MDBX_LOG_VERBOSE, func, line, "%s %d (%s)\n",
+              is_error ? "error" : "condition", err, mdbx_strerror_r(err, buf, sizeof(buf)));
+  }
+}
+
+/* Dump a val in ascii or hexadecimal. */
+__cold const char *mdbx_dump_val(const MDBX_val *val, char *const buf, const size_t bufsize) {
+  if (!val)
+    return "<null>";
+  if (!val->iov_len)
+    return "<empty>";
+  if (!buf || bufsize < 4)
+    return nullptr;
+
+  if (!val->iov_base) {
+    int len = snprintf(buf, bufsize, "<nullptr.%zu>", val->iov_len);
+    assert(len > 0 && (size_t)len < bufsize);
+    (void)len;
+    return buf;
+  }
+
+  bool is_ascii = true;
+  const uint8_t *const data = val->iov_base;
+  for (size_t i = 0; i < val->iov_len; i++)
+    if (data[i] < ' ' || data[i] > '~') {
+      is_ascii = false;
+      break;
+    }
+
+  if (is_ascii) {
+    int len = snprintf(buf, bufsize, "%.*s", (val->iov_len > INT_MAX) ? INT_MAX : (int)val->iov_len, data);
+    assert(len > 0 && (size_t)len < bufsize);
+    (void)len;
+  } else {
+    char *const detent = buf + bufsize - 2;
+    char *ptr = buf;
+    *ptr++ = '<';
+    for (size_t i = 0; i < val->iov_len && ptr < detent; i++) {
+      const char hex[16] = {'0', '1', '2', '3', '4', '5', '6', '7', '8', '9', 'a', 'b', 'c', 'd', 'e', 'f'};
+      *ptr++ = hex[data[i] >> 4];
+      *ptr++ = hex[data[i] & 15];
+    }
+    if (ptr < detent)
+      *ptr++ = '>';
+    *ptr = '\0';
+  }
+  return buf;
+}
+
+/*------------------------------------------------------------------------------
+ LY: debug stuff */
+
+__cold const char *pagetype_caption(const uint8_t type, char buf4unknown[16]) {
+  switch (type) {
+  case P_BRANCH:
+    return "branch";
+  case P_LEAF:
+    return "leaf";
+  case P_LEAF | P_SUBP:
+    return "subleaf";
+  case P_LEAF | P_DUPFIX:
+    return "dupfix-leaf";
+  case P_LEAF | P_DUPFIX | P_SUBP:
+    return "dupfix-subleaf";
+  case P_LEAF | P_DUPFIX | P_SUBP | P_LEGACY_DIRTY:
+    return "dupfix-subleaf.legacy-dirty";
+  case P_LARGE:
+    return "large";
+  default:
+    snprintf(buf4unknown, 16, "unknown_0x%x", type);
+    return buf4unknown;
+  }
+}
+
+__cold static const char *leafnode_type(node_t *n) {
+  static const char *const tp[2][2] = {{"", ": DB"}, {": sub-page", ": sub-DB"}};
+  return (node_flags(n) & N_BIG) ? ": large page" : tp[!!(node_flags(n) & N_DUP)][!!(node_flags(n) & N_TREE)];
+}
+
+/* Display all the keys in the page. */
+__cold void page_list(page_t *mp) {
+  pgno_t pgno = mp->pgno;
+  const char *type;
+  node_t *node;
+  size_t i, nkeys, nsize, total = 0;
+  MDBX_val key;
+  DKBUF;
+
+  switch (page_type(mp)) {
+  case P_BRANCH:
+    type = "Branch page";
+    break;
+  case P_LEAF:
+    type = "Leaf page";
+    break;
+  case P_LEAF | P_SUBP:
+    type = "Leaf sub-page";
+    break;
+  case P_LEAF | P_DUPFIX:
+    type = "Leaf2 page";
+    break;
+  case P_LEAF | P_DUPFIX | P_SUBP:
+    type = "Leaf2 sub-page";
+    break;
+  case P_LARGE:
+    VERBOSE("Overflow page %" PRIaPGNO " pages %u\n", pgno, mp->pages);
+    return;
+  case P_META:
+    VERBOSE("Meta-page %" PRIaPGNO " txnid %" PRIu64 "\n", pgno, unaligned_peek_u64(4, page_meta(mp)->txnid_a));
+    return;
+  default:
+    VERBOSE("Bad page %" PRIaPGNO " flags 0x%X\n", pgno, mp->flags);
+    return;
+  }
+
+  nkeys = page_numkeys(mp);
+  VERBOSE("%s %" PRIaPGNO " numkeys %zu\n", type, pgno, nkeys);
+
+  for (i = 0; i < nkeys; i++) {
+    if (is_dupfix_leaf(mp)) { /* DUPFIX pages have no entries[] or node headers */
+      key = page_dupfix_key(mp, i, nsize = mp->dupfix_ksize);
+      total += nsize;
+      VERBOSE("key %zu: nsize %zu, %s\n", i, nsize, DKEY(&key));
+      continue;
+    }
+    node = page_node(mp, i);
+    key.iov_len = node_ks(node);
+    key.iov_base = node->payload;
+    nsize = NODESIZE + key.iov_len;
+    if (is_branch(mp)) {
+      VERBOSE("key %zu: page %" PRIaPGNO ", %s\n", i, node_pgno(node), DKEY(&key));
+      total += nsize;
+    } else {
+      if (node_flags(node) & N_BIG)
+        nsize += sizeof(pgno_t);
+      else
+        nsize += node_ds(node);
+      total += nsize;
+      nsize += sizeof(indx_t);
+      VERBOSE("key %zu: nsize %zu, %s%s\n", i, nsize, DKEY(&key), leafnode_type(node));
+    }
+    total = EVEN_CEIL(total);
+  }
+  VERBOSE("Total: header %u + contents %zu + unused %zu\n", is_dupfix_leaf(mp) ? PAGEHDRSZ : PAGEHDRSZ + mp->lower,
+          total, page_room(mp));
+}
+
+__cold static int setup_debug(MDBX_log_level_t level, MDBX_debug_flags_t flags, union logger_union logger, char *buffer,
+                              size_t buffer_size) {
+  ENSURE(nullptr, osal_fastmutex_acquire(&globals.debug_lock) == 0);
+
+  const int rc = globals.runtime_flags | (globals.loglevel << 16);
+  if (level != MDBX_LOG_DONTCHANGE)
+    globals.loglevel = (uint8_t)level;
+
+  if (flags != MDBX_DBG_DONTCHANGE) {
+    flags &=
+#if MDBX_DEBUG
+        MDBX_DBG_ASSERT | MDBX_DBG_AUDIT | MDBX_DBG_JITTER |
+#endif
+        MDBX_DBG_DUMP | MDBX_DBG_LEGACY_MULTIOPEN | MDBX_DBG_LEGACY_OVERLAP | MDBX_DBG_DONT_UPGRADE;
+    globals.runtime_flags = (uint8_t)flags;
+  }
+
+  assert(MDBX_LOGGER_DONTCHANGE == ((MDBX_debug_func *)(intptr_t)-1));
+  if (logger.ptr != (void *)((intptr_t)-1)) {
+    globals.logger.ptr = logger.ptr;
+    globals.logger_buffer = buffer;
+    globals.logger_buffer_size = buffer_size;
+  }
+
+  ENSURE(nullptr, osal_fastmutex_release(&globals.debug_lock) == 0);
+  return rc;
+}
+
+__cold int mdbx_setup_debug_nofmt(MDBX_log_level_t level, MDBX_debug_flags_t flags, MDBX_debug_func_nofmt *logger,
+                                  char *buffer, size_t buffer_size) {
+  union logger_union thunk;
+  thunk.nofmt = (logger && buffer && buffer_size) ? logger : MDBX_LOGGER_NOFMT_DONTCHANGE;
+  return setup_debug(level, flags, thunk, buffer, buffer_size);
+}
+
+__cold int mdbx_setup_debug(MDBX_log_level_t level, MDBX_debug_flags_t flags, MDBX_debug_func *logger) {
+  union logger_union thunk;
+  thunk.fmt = logger;
+  return setup_debug(level, flags, thunk, nullptr, 0);
+}

src/logging_and_debug.h

@@ -0,0 +1,159 @@
+/// \copyright SPDX-License-Identifier: Apache-2.0
+/// \author Леонид Юрьев aka Leonid Yuriev <leo@yuriev.ru> \date 2015-2025
+
+#pragma once
+
+#include "essentials.h"
+
+#ifndef __Wpedantic_format_voidptr
+MDBX_MAYBE_UNUSED static inline const void *__Wpedantic_format_voidptr(const void *ptr) { return ptr; }
+#define __Wpedantic_format_voidptr(ARG) __Wpedantic_format_voidptr(ARG)
+#endif /* __Wpedantic_format_voidptr */
+
+MDBX_INTERNAL void MDBX_PRINTF_ARGS(4, 5) debug_log(int level, const char *function, int line, const char *fmt, ...)
+    MDBX_PRINTF_ARGS(4, 5);
+MDBX_INTERNAL void debug_log_va(int level, const char *function, int line, const char *fmt, va_list args);
+
+#if MDBX_DEBUG
+#define LOG_ENABLED(LVL) unlikely(LVL <= globals.loglevel)
+#define AUDIT_ENABLED() unlikely((globals.runtime_flags & (unsigned)MDBX_DBG_AUDIT))
+#else /* MDBX_DEBUG */
+#define LOG_ENABLED(LVL) (LVL < MDBX_LOG_VERBOSE && LVL <= globals.loglevel)
+#define AUDIT_ENABLED() (0)
+#endif /* LOG_ENABLED() & AUDIT_ENABLED() */
+
+#if MDBX_FORCE_ASSERTIONS
+#define ASSERT_ENABLED() (1)
+#elif MDBX_DEBUG
+#define ASSERT_ENABLED() likely((globals.runtime_flags & (unsigned)MDBX_DBG_ASSERT))
+#else
+#define ASSERT_ENABLED() (0)
+#endif /* ASSERT_ENABLED() */
+
+#define DEBUG_EXTRA(fmt, ...)                                                                                          \
+  do {                                                                                                                 \
+    if (LOG_ENABLED(MDBX_LOG_EXTRA))                                                                                   \
+      debug_log(MDBX_LOG_EXTRA, __func__, __LINE__, fmt, __VA_ARGS__);                                                 \
+  } while (0)
+
+#define DEBUG_EXTRA_PRINT(fmt, ...)                                                                                    \
+  do {                                                                                                                 \
+    if (LOG_ENABLED(MDBX_LOG_EXTRA))                                                                                   \
+      debug_log(MDBX_LOG_EXTRA, nullptr, 0, fmt, __VA_ARGS__);                                                         \
+  } while (0)
+
+#define TRACE(fmt, ...)                                                                                                \
+  do {                                                                                                                 \
+    if (LOG_ENABLED(MDBX_LOG_TRACE))                                                                                   \
+      debug_log(MDBX_LOG_TRACE, __func__, __LINE__, fmt "\n", __VA_ARGS__);                                            \
+  } while (0)
+
+#define DEBUG(fmt, ...)                                                                                                \
+  do {                                                                                                                 \
+    if (LOG_ENABLED(MDBX_LOG_DEBUG))                                                                                   \
+      debug_log(MDBX_LOG_DEBUG, __func__, __LINE__, fmt "\n", __VA_ARGS__);                                            \
+  } while (0)
+
+#define VERBOSE(fmt, ...)                                                                                              \
+  do {                                                                                                                 \
+    if (LOG_ENABLED(MDBX_LOG_VERBOSE))                                                                                 \
+      debug_log(MDBX_LOG_VERBOSE, __func__, __LINE__, fmt "\n", __VA_ARGS__);                                          \
+  } while (0)
+
+#define NOTICE(fmt, ...)                                                                                               \
+  do {                                                                                                                 \
+    if (LOG_ENABLED(MDBX_LOG_NOTICE))                                                                                  \
+      debug_log(MDBX_LOG_NOTICE, __func__, __LINE__, fmt "\n", __VA_ARGS__);                                           \
+  } while (0)
+
+#define WARNING(fmt, ...)                                                                                              \
+  do {                                                                                                                 \
+    if (LOG_ENABLED(MDBX_LOG_WARN))                                                                                    \
+      debug_log(MDBX_LOG_WARN, __func__, __LINE__, fmt "\n", __VA_ARGS__);                                             \
+  } while (0)
+
+#undef ERROR /* wingdi.h                                                                                               \
+  Yeah, morons from M$ put such definition to the public header. */
+
+#define ERROR(fmt, ...)                                                                                                \
+  do {                                                                                                                 \
+    if (LOG_ENABLED(MDBX_LOG_ERROR))                                                                                   \
+      debug_log(MDBX_LOG_ERROR, __func__, __LINE__, fmt "\n", __VA_ARGS__);                                            \
+  } while (0)
+
+#define FATAL(fmt, ...) debug_log(MDBX_LOG_FATAL, __func__, __LINE__, fmt "\n", __VA_ARGS__);
+
+#if MDBX_DEBUG
+#define ASSERT_FAIL(env, msg, func, line) mdbx_assert_fail(env, msg, func, line)
+#else /* MDBX_DEBUG */
+MDBX_NORETURN __cold void assert_fail(const char *msg, const char *func, unsigned line);
+#define ASSERT_FAIL(env, msg, func, line)                                                                              \
+  do {                                                                                                                 \
+    (void)(env);                                                                                                       \
+    assert_fail(msg, func, line);                                                                                      \
+  } while (0)
+#endif /* MDBX_DEBUG */
+
+#define ENSURE_MSG(env, expr, msg)                                                                                     \
+  do {                                                                                                                 \
+    if (unlikely(!(expr)))                                                                                             \
+      ASSERT_FAIL(env, msg, __func__, __LINE__);                                                                       \
+  } while (0)
+
+#define ENSURE(env, expr) ENSURE_MSG(env, expr, #expr)
+
+/* assert(3) variant in environment context */
+#define eASSERT(env, expr)                                                                                             \
+  do {                                                                                                                 \
+    if (ASSERT_ENABLED())                                                                                              \
+      ENSURE(env, expr);                                                                                               \
+  } while (0)
+
+/* assert(3) variant in cursor context */
+#define cASSERT(mc, expr) eASSERT((mc)->txn->env, expr)
+
+/* assert(3) variant in transaction context */
+#define tASSERT(txn, expr) eASSERT((txn)->env, expr)
+
+#ifndef xMDBX_TOOLS /* Avoid using internal eASSERT() */
+#undef assert
+#define assert(expr) eASSERT(nullptr, expr)
+#endif
+
+MDBX_MAYBE_UNUSED static inline void jitter4testing(bool tiny) {
+#if MDBX_DEBUG
+  if (globals.runtime_flags & (unsigned)MDBX_DBG_JITTER)
+    osal_jitter(tiny);
+#else
+  (void)tiny;
+#endif
+}
+
+MDBX_MAYBE_UNUSED MDBX_INTERNAL void page_list(page_t *mp);
+
+MDBX_INTERNAL const char *pagetype_caption(const uint8_t type, char buf4unknown[16]);
+/* Key size which fits in a DKBUF (debug key buffer). */
+#define DKBUF_MAX 127
+#define DKBUF char dbg_kbuf[DKBUF_MAX * 4 + 2]
+#define DKEY(x) mdbx_dump_val(x, dbg_kbuf, DKBUF_MAX * 2 + 1)
+#define DVAL(x) mdbx_dump_val(x, dbg_kbuf + DKBUF_MAX * 2 + 1, DKBUF_MAX * 2 + 1)
+
+#if MDBX_DEBUG
+#define DKBUF_DEBUG DKBUF
+#define DKEY_DEBUG(x) DKEY(x)
+#define DVAL_DEBUG(x) DVAL(x)
+#else
+#define DKBUF_DEBUG ((void)(0))
+#define DKEY_DEBUG(x) ("-")
+#define DVAL_DEBUG(x) ("-")
+#endif
+
+MDBX_INTERNAL void log_error(const int err, const char *func, unsigned line);
+
+MDBX_MAYBE_UNUSED static inline int log_if_error(const int err, const char *func, unsigned line) {
+  if (unlikely(err != MDBX_SUCCESS))
+    log_error(err, func, line);
+  return err;
+}
+
+#define LOG_IFERR(err) log_if_error((err), __func__, __LINE__)

src/man1/mdbx_chk.1

@@ -0,0 +1,106 @@
+.\" Copyright 2015-2025 Leonid Yuriev <leo@yuriev.ru>.
+.\" Copying restrictions apply.  See COPYRIGHT/LICENSE.
+.TH MDBX_CHK 1 "2025-01-14" "MDBX 0.14"
+.SH NAME
+mdbx_chk \- MDBX checking tool
+.SH SYNOPSIS
+.B mdbx_chk
+[\c
+.BR \-V ]
+[\c
+.BR \-v [ v [ v ]]]
+[\c
+.BR \-n ]
+[\c
+.BR \-q ]
+[\c
+.BR \-c ]
+[\c
+.BR \-w ]
+[\c
+.BR \-d ]
+[\c
+.BR \-i ]
+[\c
+.BI \-s \ table\fR]
+.BR \ dbpath
+.SH DESCRIPTION
+The
+.B mdbx_chk
+utility is intended to check an MDBX database file.
+.SH OPTIONS
+.TP
+.BR \-V
+Write the library version number to the standard output, and exit.
+.TP
+.BR \-v
+Produce verbose output, including summarize space and page usage statistics.
+If \fB\-vv\fP is given, be more verbose, show summarized B-tree info
+and space allocation.
+If \fB\-vvv\fP is given, be more verbose, include summarized statistics
+of leaf B-tree pages.
+If \fB\-vvvv\fP is given, be even more verbose, show info of each page
+during B-tree traversal and basic info of each GC record.
+If \fB\-vvvvv\fP is given, turn maximal verbosity, display the full list
+of page IDs in the GC records and size of each key-value pair of database(s).
+.TP
+.BR \-q
+Be quiet; do not output anything even if an error was detected.
+.TP
+.BR \-c
+Force using cooperative mode while opening environment, i.e. don't try to open
+in exclusive/monopolistic mode. Only exclusive/monopolistic mode allow complete
+check, including full check of all meta-pages and actual size of database file.
+.TP
+.BR \-w
+Open environment in read-write mode and lock for writing while checking.
+This could be impossible if environment already used by another process(s)
+in an incompatible read-write mode. This allows rollback to last steady commit
+(in case environment was not closed properly) and then check transaction IDs
+of meta-pages. Otherwise, without \fB\-w\fP option environment will be
+opened in read-only mode.
+.TP
+.BR \-d
+Disable page-by-page traversal of B-tree. In this case, without B-tree
+traversal, it is unable to check for lost-unused pages nor for double-used
+pages.
+.TP
+.BR \-i
+Ignore wrong order errors, which will likely false-positive if custom
+comparator(s) was used.
+.TP
+.BR \-s \ table
+Verify and show info only for a specific table.
+.TP
+.BR \-0 | \-1 | \-2
+Using specific meta-page 0, or 2 for checking.
+.TP
+.BR \-t
+Turn to a specified meta-page on successful check.
+.TP
+.BR \-T
+Turn to a specified meta-page EVEN ON UNSUCCESSFUL CHECK!
+.TP
+.BR \-u
+Warms up the DB before checking via notifying OS kernel of subsequent access to the database pages.
+.TP
+.BR \-U
+Warms up the DB before checking, notifying the OS kernel of subsequent access to the database pages,
+then forcibly loads ones by sequential access and tries to lock database pages in memory.
+.TP
+.BR \-n
+Open MDBX environment(s) which do not use subdirectories.
+This is a legacy option. For now MDBX handles this automatically.
+
+.SH DIAGNOSTICS
+Exit status is zero if no errors occur. Errors result in a non-zero exit status
+and a diagnostic message being written to standard error
+if no quiet mode was requested.
+.SH "SEE ALSO"
+.BR mdbx_stat (1),
+.BR mdbx_copy (1),
+.BR mdbx_dump (1),
+.BR mdbx_load (1)
+.BR mdbx_drop (1)
+.SH AUTHOR
+Leonid Yuriev <https://gitflic.ru/user/erthink>

src/man1/mdbx_copy.1

@@ -0,0 +1,95 @@
+.\" Copyright 2015-2025 Leonid Yuriev <leo@yuriev.ru>.
+.\" Copyright 2015,2016 Peter-Service R&D LLC <http://billing.ru/>.
+.\" Copyright 2012-2015 Howard Chu, Symas Corp. All Rights Reserved.
+.\" Copying restrictions apply.  See COPYRIGHT/LICENSE.
+.TH MDBX_COPY 1 "2025-01-14" "MDBX 0.14"
+.SH NAME
+mdbx_copy \- MDBX environment copy tool
+.SH SYNOPSIS
+.B mdbx_copy
+[\c
+.BR \-V ]
+[\c
+.BR \-q ]
+[\c
+.BR \-c ]
+[\c
+.BR \-d ]
+[\c
+.BR \-p ]
+[\c
+.BR \-n ]
+.B src_path
+[\c
+.BR dest_path ]
+.SH DESCRIPTION
+The
+.B mdbx_copy
+utility copies an MDBX environment. The environment can
+be copied regardless of whether it is currently in use.
+No lockfile is created, since it gets recreated at need.
+
+If
+.I dest_path
+is specified it must be the path of an empty directory
+for storing the backup. Otherwise, the backup will be
+written to stdout.
+
+.SH OPTIONS
+.TP
+.BR \-V
+Write the library version number to the standard output, and exit.
+.TP
+.BR \-q
+Be quiet.
+.TP
+.BR \-c
+Compact while copying. Only current data pages will be copied; freed
+or unused pages will be omitted from the copy. This option will
+slow down the backup process as it is more CPU-intensive.
+Currently it fails if the environment has suffered a page leak.
+.TP
+.BR \-d
+Alters geometry to enforce the copy to be a dynamic size DB,
+which could be growth and shrink by reasonable steps on the fly.
+.TP
+.BR \-p
+Use read transaction parking/ousting during copying MVCC-snapshot.
+This allows the writing transaction to oust the read
+transaction used to copy the database if copying takes so long
+that it will interfere with the recycling old MVCC snapshots
+and may lead to an overflow of the database.
+However, if the reading transaction is ousted the copy will
+be aborted until successful completion. Thus, this option
+allows copy the database without interfering with write
+transactions and a threat of database overflow, but at the cost
+that copying will be aborted to prevent such conditions.
+.TP
+.BR \-u
+Warms up the DB before copying via notifying OS kernel of subsequent access to the database pages.
+.TP
+.BR \-U
+Warms up the DB before copying, notifying the OS kernel of subsequent access to the database pages,
+then forcibly loads ones by sequential access and tries to lock database pages in memory.
+.TP
+.BR \-n
+Open MDBX environment(s) which do not use subdirectories.
+This is legacy option. For now MDBX handles this automatically.
+
+.SH DIAGNOSTICS
+Exit status is zero if no errors occur.
+Errors result in a non-zero exit status and
+a diagnostic message being written to standard error.
+.SH CAVEATS
+This utility can trigger significant file size growth if run
+in parallel with write transactions, because pages which they
+free during copying cannot be reused until the copy is done.
+.SH "SEE ALSO"
+.BR mdbx_dump (1),
+.BR mdbx_chk (1),
+.BR mdbx_stat (1),
+.BR mdbx_load (1)
+.BR mdbx_drop (1)
+.SH AUTHOR
+Howard Chu of Symas Corporation <http://www.symas.com>,
+Leonid Yuriev <https://gitflic.ru/user/erthink>

src/man1/mdbx_drop.1

@@ -0,0 +1,48 @@
+.\" Copyright 2021-2025 Leonid Yuriev <leo@yuriev.ru>.
+.\" Copyright 2014-2021 Howard Chu, Symas Corp. All Rights Reserved.
+.\" Copying restrictions apply.  See COPYRIGHT/LICENSE.
+.TH MDBX_DROP 1 "2025-01-14" "MDBX 0.14"
+.SH NAME
+mdbx_drop \- MDBX database delete tool
+.SH SYNOPSIS
+.B mdbx_drop
+[\c
+.BR \-V ]
+[\c
+.BR \-d ]
+[\c
+.BI \-s \ table\fR]
+[\c
+.BR \-n ]
+.BR \ dbpath
+.SH DESCRIPTION
+The
+.B mdbx_drop
+utility empties or deletes a database in the specified
+environment.
+.SH OPTIONS
+.TP
+.BR \-V
+Write the library version number to the standard output, and exit.
+.TP
+.BR \-d
+Delete the specified database, don't just empty it.
+.TP
+.BR \-s \ table
+Operate on a specific table. If no table is specified, only the main table is dropped.
+.TP
+.BR \-n
+Dump an MDBX database which does not use subdirectories.
+This is legacy option. For now MDBX handles this automatically.
+
+.SH DIAGNOSTICS
+Exit status is zero if no errors occur.
+Errors result in a non-zero exit status and
+a diagnostic message being written to standard error.
+.SH "SEE ALSO"
+.BR mdbx_load (1),
+.BR mdbx_copy (1),
+.BR mdbx_chk (1),
+.BR mdbx_stat (1)
+.SH AUTHOR
+Howard Chu of Symas Corporation <http://www.symas.com>