Fix memory leak in mastertrainer_test (fixes issue #3215)

The issue was introduced in commit 6e9456415.

Partially reverting this commit fixes it.

Signed-off-by: Stefan Weil <sw@weilnetz.de>

src/training/classifier_tester.cpp

@@ -110,7 +110,8 @@ int main(int argc, char **argv) {
   tesseract::CheckSharedLibraryVersion();
   ParseArguments(&argc, &argv);
   STRING file_prefix;
-  auto [trainer,_] = tesseract::LoadTrainingData(argc, argv, false, false, &file_prefix);
+  auto trainer =
+    tesseract::LoadTrainingData(argc, argv, false, nullptr, &file_prefix);
   tesseract::TessBaseAPI* api;
   // Decode the classifier string.
   tesseract::ShapeClassifier* shape_classifier = InitializeClassifier(

src/training/common/commontraining.cpp

@@ -147,14 +147,16 @@ void ParseArguments(int* argc, char ***argv) {
 
 namespace tesseract {
 // Helper loads shape table from the given file.
-std::unique_ptr<ShapeTable> LoadShapeTable(const STRING& file_prefix) {
-  std::unique_ptr<ShapeTable> shape_table;
+ShapeTable* LoadShapeTable(const STRING& file_prefix) {
+  ShapeTable* shape_table = nullptr;
   STRING shape_table_file = file_prefix;
   shape_table_file += kShapeTableFileSuffix;
   TFile shape_fp;
   if (shape_fp.Open(shape_table_file.c_str(), nullptr)) {
-    shape_table = std::make_unique<ShapeTable>();
+    shape_table = new ShapeTable;
     if (!shape_table->DeSerialize(&shape_fp)) {
+      delete shape_table;
+      shape_table = nullptr;
       tprintf("Error: Failed to read shape table %s\n",
               shape_table_file.c_str());
     } else {
@@ -202,10 +204,9 @@ void WriteShapeTable(const STRING& file_prefix, const ShapeTable& shape_table) {
  * If shape_table is not nullptr, but failed to load, make a fake flat one,
  * as shape clustering was not run.
  */
-std::pair<std::unique_ptr<MasterTrainer>, std::unique_ptr<ShapeTable>>
-LoadTrainingData(int argc, const char* const * argv,
+std::unique_ptr<MasterTrainer> LoadTrainingData(int argc, const char* const * argv,
                                 bool replication,
-                                bool shape_analysis,
+                                ShapeTable** shape_table,
                                 STRING* file_prefix) {
   InitFeatureDefs(&feature_defs);
   InitIntegerFX();
@@ -218,9 +219,12 @@ LoadTrainingData(int argc, const char* const * argv,
   // a shape_table written by a previous shape clustering, then
   // shape_analysis will be true, meaning that the MasterTrainer will replace
   // some members of the unicharset with their fragments.
-  std::unique_ptr<ShapeTable> shape_table;
-  if (shape_analysis) {
-    shape_table = LoadShapeTable(*file_prefix);
+  bool shape_analysis = false;
+  if (shape_table != nullptr) {
+    *shape_table = LoadShapeTable(*file_prefix);
+    if (*shape_table != nullptr) shape_analysis = true;
+  } else {
+    shape_analysis = true;
   }
   auto trainer = std::make_unique<MasterTrainer>(NM_CHAR_ANISOTROPIC,
                                              shape_analysis,
@@ -283,18 +287,19 @@ LoadTrainingData(int argc, const char* const * argv,
     fprintf(stderr, "Failed to save unicharset to file %s\n", FLAGS_O.c_str());
     return {};
   }
-  if (shape_analysis) {
+
+  if (shape_table != nullptr) {
     // If we previously failed to load a shapetable, then shape clustering
     // wasn't run so make a flat one now.
-    if (!shape_table) {
-      shape_table = std::make_unique<ShapeTable>();
-      trainer->SetupFlatShapeTable(shape_table.get());
+    if (*shape_table == nullptr) {
+      *shape_table = new ShapeTable;
+      trainer->SetupFlatShapeTable(*shape_table);
       tprintf("Flat shape table summary: %s\n",
-              shape_table->SummaryStr().c_str());
+              (*shape_table)->SummaryStr().c_str());
     }
-    shape_table->set_unicharset(trainer->unicharset());
+    (*shape_table)->set_unicharset(trainer->unicharset());
   }
-  return { std::move(trainer), std::move(shape_table) };
+  return trainer;
 }
 
 }  // namespace tesseract.

src/training/common/commontraining.h

@@ -100,7 +100,7 @@ using MERGE_CLASS = MERGE_CLASS_NODE*;
 namespace tesseract {
 
 // Helper loads shape table from the given file.
-std::unique_ptr<ShapeTable> LoadShapeTable(const STRING& file_prefix);
+ShapeTable* LoadShapeTable(const STRING& file_prefix);
 // Helper to write the shape_table.
 TESS_COMMON_TRAINING_API
 void WriteShapeTable(const STRING& file_prefix, const ShapeTable& shape_table);
@@ -119,10 +119,9 @@ void WriteShapeTable(const STRING& file_prefix, const ShapeTable& shape_table);
 // If shape_table is not nullptr, but failed to load, make a fake flat one,
 // as shape clustering was not run.
 TESS_COMMON_TRAINING_API
-std::pair<std::unique_ptr<MasterTrainer>, std::unique_ptr<ShapeTable>>
-LoadTrainingData(int argc, const char* const * argv,
+std::unique_ptr<MasterTrainer> LoadTrainingData(int argc, const char* const * argv,
                                 bool replication,
-                                bool shape_analysis,
+                                ShapeTable** shape_table,
                                 STRING* file_prefix);
 
 }  // namespace tesseract.

src/training/mftraining.cpp

@@ -202,11 +202,12 @@ int main (int argc, char **argv) {
 
   ParseArguments(&argc, &argv);
 
+  ShapeTable* shape_table = nullptr;
   STRING file_prefix;
   // Load the training data.
-  auto [trainer,shape_table] = tesseract::LoadTrainingData(argc, argv,
+  auto trainer = tesseract::LoadTrainingData(argc, argv,
                                                        false,
-                                                       true,
+                                                       &shape_table,
                                                        &file_prefix);
   if (trainer == nullptr) return 1;  // Failed.
 
@@ -215,7 +216,7 @@ int main (int argc, char **argv) {
   // with the same list of unichars becomes a different class and the configs
   // represent the different combinations of fonts.
   IndexMapBiDi config_map;
-  SetupConfigMap(shape_table.get(), &config_map);
+  SetupConfigMap(shape_table, &config_map);
 
   WriteShapeTable(file_prefix, *shape_table);
   // If the shape_table is flat, then either we didn't run shape clustering, or
@@ -269,6 +270,7 @@ int main (int argc, char **argv) {
   }
   delete [] float_classes;
   FreeLabeledClassList(mf_classes);
+  delete shape_table;
   printf("Done!\n");
   if (!FLAGS_test_ch.empty()) {
     // If we are displaying debug window(s), wait for the user to look at them.

src/training/shapeclustering.cpp

@@ -49,7 +49,9 @@ int main(int argc, char **argv) {
   ParseArguments(&argc, &argv);
 
   STRING file_prefix;
-  auto [trainer,_] = tesseract::LoadTrainingData(argc, argv, false, false, &file_prefix);
+  auto trainer =
+      tesseract::LoadTrainingData(argc, argv, false, nullptr, &file_prefix);
+
   if (!trainer)
     return 1;
 

unittest/mastertrainer_test.cc

@@ -159,6 +159,14 @@ class MasterTrainerTest : public testing::Test {
     return file::JoinPath(FLAGS_test_tmpdir, name);
   }
 
+  MasterTrainerTest() {
+    shape_table_ = nullptr;
+    master_trainer_ = nullptr;
+  }
+  ~MasterTrainerTest() {
+    delete shape_table_;
+  }
+
   // Initializes the master_trainer_ and shape_table_.
   // if load_from_tmp, then reloads a master trainer that was saved by a
   // previous call in which it was false.
@@ -171,9 +179,10 @@ class MasterTrainerTest : public testing::Test {
     const char* argv[] = {tr_file_name.c_str()};
     int argc = 1;
     STRING file_prefix;
-    auto [m,s] = LoadTrainingData(argc, argv, false, true, &file_prefix);
-    master_trainer_ = std::move(m);
-    shape_table_ = std::move(s);
+    delete shape_table_;
+    shape_table_ = nullptr;
+    master_trainer_ =
+        LoadTrainingData(argc, argv, false, &shape_table_, &file_prefix);
     EXPECT_TRUE(master_trainer_ != nullptr);
     EXPECT_TRUE(shape_table_ != nullptr);
   }
@@ -226,7 +235,7 @@ class MasterTrainerTest : public testing::Test {
   }
 
   // Objects declared here can be used by all tests in the test case for Foo.
-  std::unique_ptr<ShapeTable> shape_table_;
+  ShapeTable* shape_table_;
   std::unique_ptr<MasterTrainer> master_trainer_;
 #endif
 };
@@ -257,7 +266,7 @@ TEST_F(MasterTrainerTest, ErrorCounterTest) {
   // count junk.
   if (shape_table_->FindShape(0, -1) < 0) shape_table_->AddShape(0, 0);
   // Make a mock classifier.
-  auto shape_classifier = std::make_unique<MockClassifier>(shape_table_.get());
+  auto shape_classifier = std::make_unique<MockClassifier>(shape_table_);
   // Get the accuracy report.
   STRING accuracy_report;
   master_trainer_->TestClassifierOnSamples(tesseract::CT_UNICHAR_TOP1_ERR, 0,