Const members in structs

native_client/client.cc

@@ -45,7 +45,7 @@ struct meta_word {
 };
 
 char*
-CandidateTranscriptToString(CandidateTranscript* transcript)
+CandidateTranscriptToString(const CandidateTranscript* transcript)
 {
   std::string retval = "";
   for (int i = 0; i < transcript->num_tokens; i++) {
@@ -56,7 +56,7 @@ CandidateTranscriptToString(CandidateTranscript* transcript)
 }
 
 std::vector<meta_word>
-CandidateTranscriptToWords(CandidateTranscript* transcript)
+CandidateTranscriptToWords(const CandidateTranscript* transcript)
 {
   std::vector<meta_word> word_list;
 
@@ -101,7 +101,7 @@ CandidateTranscriptToWords(CandidateTranscript* transcript)
 }
 
 std::string
-CandidateTranscriptToJSON(CandidateTranscript *transcript)
+CandidateTranscriptToJSON(const CandidateTranscript *transcript)
 {
   std::ostringstream out_string;
 
@@ -130,7 +130,7 @@ MetadataToJSON(Metadata* result)
   out_string << "{\n";
 
   for (int j=0; j < result->num_transcripts; ++j) {
-    CandidateTranscript *transcript = &result->transcripts[j];
+    const CandidateTranscript *transcript = &result->transcripts[j];
 
     if (j == 0) {
       out_string << CandidateTranscriptToJSON(transcript);

native_client/deepspeech.cc

@@ -478,14 +478,14 @@ DS_FreeMetadata(Metadata* m)
   if (m) {
     for (int i = 0; i < m->num_transcripts; ++i) {
       for (int j = 0; j < m->transcripts[i].num_tokens; ++j) {
-        free(m->transcripts[i].tokens[j].text);
+        free((void*)m->transcripts[i].tokens[j].text);
       }
 
-      delete[] m->transcripts[i].tokens;
+      free((void*)m->transcripts[i].tokens);
     }
 
-    delete[] m->transcripts;
-    delete m;
+    free((void*)m->transcripts);
+    free(m);
   }
 }
 

native_client/deepspeech.h

@@ -24,13 +24,13 @@ typedef struct StreamingState StreamingState;
  */
 typedef struct TokenMetadata {
   /** The text corresponding to this token */
-  char* text;
+  const char* const text;
 
   /** Position of the token in units of 20ms */
-  unsigned int timestep;
+  const unsigned int timestep;
 
   /** Position of the token in seconds */
-  float start_time;
+  const float start_time;
 } TokenMetadata;
 
 /**
@@ -39,14 +39,14 @@ typedef struct TokenMetadata {
  */
 typedef struct CandidateTranscript {
   /** Array of TokenMetadata objects */
-  TokenMetadata* tokens;
+  const TokenMetadata* const tokens;
   /** Size of the tokens array */
-  unsigned int num_tokens;
+  const unsigned int num_tokens;
   /** Approximated confidence value for this transcript. This is roughly the
    * sum of the acoustic model logit values for each timestep/character that
    * contributed to the creation of this transcript.
    */
-  double confidence;
+  const double confidence;
 } CandidateTranscript;
 
 /**
@@ -54,9 +54,9 @@ typedef struct CandidateTranscript {
  */
 typedef struct Metadata {
   /** Array of CandidateTranscript objects */
-  CandidateTranscript* transcripts;
+  const CandidateTranscript* const transcripts;
   /** Size of the transcripts array */
-  unsigned int num_transcripts;
+  const unsigned int num_transcripts;
 } Metadata;
 
 enum DeepSpeech_Error_Codes

native_client/javascript/deepspeech.i

@@ -85,10 +85,6 @@ using namespace node;
 %ignore Metadata::num_transcripts;
 %ignore CandidateTranscript::num_tokens;
 
-%immutable Metadata::transcripts;
-%immutable CandidateTranscripts::tokens;
-%immutable TokenMetadata::text;
-
 %nodefaultctor Metadata;
 %nodefaultdtor Metadata;
 %nodefaultctor CandidateTranscript;

native_client/modelstate.cc

@@ -41,33 +41,35 @@ ModelState::decode_metadata(const DecoderState& state,
                             size_t num_results)
 {
   vector<Output> out = state.decode(num_results);
-  size_t num_returned = out.size();
+  unsigned int num_returned = out.size();
 
-  std::unique_ptr<Metadata> metadata(new Metadata);
-  metadata->num_transcripts = num_returned;
-
-  std::unique_ptr<CandidateTranscript[]> transcripts(new CandidateTranscript[num_returned]);
+  CandidateTranscript* transcripts = (CandidateTranscript*)malloc(sizeof(CandidateTranscript)*num_returned);
 
   for (int i = 0; i < num_returned; ++i) {
-    transcripts[i].num_tokens = out[i].tokens.size();
-    transcripts[i].confidence = out[i].confidence;
+    TokenMetadata* tokens = (TokenMetadata*)malloc(sizeof(TokenMetadata)*out[i].tokens.size());
 
-    std::unique_ptr<TokenMetadata[]> tokens(new TokenMetadata[transcripts[i].num_tokens]);
-
-    // Loop through each token
     for (int j = 0; j < out[i].tokens.size(); ++j) {
-      tokens[j].text = strdup(alphabet_.StringFromLabel(out[i].tokens[j]).c_str());
-      tokens[j].timestep = out[i].timesteps[j];
-      tokens[j].start_time = out[i].timesteps[j] * ((float)audio_win_step_ / sample_rate_);
-
-      if (tokens[j].start_time < 0) {
-        tokens[j].start_time = 0;
-      }
+      TokenMetadata token {
+        strdup(alphabet_.StringFromLabel(out[i].tokens[j]).c_str()),   // text
+        static_cast<unsigned int>(out[i].timesteps[j]),                // timestep
+        out[i].timesteps[j] * ((float)audio_win_step_ / sample_rate_), // start_time
+      };
+      memcpy(&tokens[j], &token, sizeof(TokenMetadata));
     }
 
-    transcripts[i].tokens = tokens.release();
+    CandidateTranscript transcript {
+      tokens,                                          // tokens
+      static_cast<unsigned int>(out[i].tokens.size()), // num_tokens
+      out[i].confidence,                               // confidence
+    };
+    memcpy(&transcripts[i], &transcript, sizeof(CandidateTranscript));
   }
 
-  metadata->transcripts = transcripts.release();
-  return metadata.release();
+  Metadata* ret = (Metadata*)malloc(sizeof(Metadata));
+  Metadata metadata {
+    transcripts,  // transcripts
+    num_returned, // num_transcripts
+  };
+  memcpy(ret, &metadata, sizeof(Metadata));
+  return ret;
 }

native_client/python/impl.i

@@ -108,10 +108,6 @@ static PyObject *parent_reference() {
   }
 }
 
-%immutable Metadata::transcripts;
-%immutable CandidateTranscript::tokens;
-%immutable TokenMetadata::text;
-
 %nodefaultctor Metadata;
 %nodefaultdtor Metadata;
 %nodefaultctor CandidateTranscript;