hunting heap memory leakages in & around tesseract; turns out in the end it's easier to dig them up via the MSVC memory profiler, but DO REMEMBER to run that one as follows, or it will all bee for naught:

- put breakpoint at start of app; preferably beyond the initialization of the global fz_context() as that one will linger until the very eend and thus show as leakage.
- run in debugger until first breakpoint.
- create heap memory snapshot in the right-side panel
- make sure the exit of the app has a breakpoint too!
- run until the very end: it doesn't matter if it's not symmetrical: best to place the end breakpoint as late as possible.
- create second memory snapshot once you get there in the debugger.
- click in right-side panel on second snapshot memory leakage amount to get the diagnostics window.
- once that one is visible, you can stop the debugger: it should remain, so you can click on lines and inspect leakage & reported callstacks / source code.

Author: GerHobbelt

include/tesseract/baseapi.h

@@ -826,8 +826,10 @@ class TESS_API TessBaseAPI {
 
 protected:
   Tesseract *tesseract_;          ///< The underlying data object.
+#if !DISABLED_LEGACY_ENGINE
   Tesseract *osd_tesseract_;      ///< For orientation & script detection.
   EquationDetect *equ_detect_;    ///< The equation detector.
+#endif
   FileReader reader_;             ///< Reads files from any filesystem.
   ImageThresholder *thresholder_; ///< Image thresholding module.
   std::vector<ParagraphModel *> *paragraph_models_;

src/api/baseapi.cpp

@@ -1483,10 +1483,10 @@ bool TessBaseAPI::ProcessPage(Pix *pix, int page_index, const char *filename,
     monitor.set_deadline_msecs(timeout_millisec);
 
     // Now run the main recognition.
-    failed = Recognize(&monitor) < 0;
+    failed = (Recognize(&monitor) < 0);
   } else {
     // Normal layout and character recognition with no timeout.
-    failed = Recognize(nullptr) < 0;
+    failed = (Recognize(nullptr) < 0);
   }
 
   if (tesseract_->tessedit_write_images) {
@@ -2401,6 +2401,9 @@ bool TessBaseAPI::Threshold(Pix **pix) {
         tesseract_->AddPixDebugPage(tesseract_->pix_thresholds(), "Otsu (tesseract) : Thresholds");
         tesseract_->AddPixDebugPage(pix_binary, "Otsu (tesseract) : Binary = post-image");
       }
+
+      if (!go)
+        pix_binary.destroy();
 	  } else {
 		  auto [ok, pix_grey, pix_binary, pix_thresholds] = thresholder_->Threshold(thresholding_method);
 
@@ -2421,7 +2424,10 @@ bool TessBaseAPI::Threshold(Pix **pix) {
         tesseract_->AddPixDebugPage(tesseract_->pix_thresholds(), (caption + " : Thresholds").c_str());
         tesseract_->AddPixDebugPage(pix_binary, (caption + " : Binary = post-image").c_str());
       }
-	  }
+
+      if (!go)
+        pix_binary.destroy();
+    }
   }
 
   thresholder_->GetImageSizes(&rect_left_, &rect_top_, &rect_width_, &rect_height_, &image_width_,

src/ccmain/pgedit.cpp

@@ -974,7 +974,8 @@ void Tesseract::display_current_page_result(PAGE_RES* page_res) {
   //boxaDestroy(BOXA * *pboxa);
   l_uint32 bordercolor;
   composeRGBAPixel(255, 32, 32, 255, &bordercolor);
-  pixDrawBoxa(pix, boxlist, 2, bordercolor);
+  pix = pixDrawBoxa(pix, boxlist, 2, bordercolor);
+  boxaDestroy(&boxlist);
 
   int block_count = 1;
 
@@ -1009,6 +1010,8 @@ void Tesseract::display_current_page_result(PAGE_RES* page_res) {
     }
     //image_win->Update();
   }
+
+  this->AddPixDebugPage(pix, "current page results", false);
 }
 
 #endif // !GRAPHICS_DISABLED

src/ccmain/tesseractclass.cpp

@@ -501,11 +501,14 @@ Tesseract::Tesseract()
 
 Tesseract::~Tesseract() {
   Clear();
-  pix_original_.destroy();
   end_tesseract();
   for (auto *lang : sub_langs_) {
     delete lang;
   }
+#if !DISABLED_LEGACY_ENGINE
+  delete equ_detect_;
+  equ_detect_ = nullptr;
+#endif // !DISABLED_LEGACY_ENGINE
   delete lstm_recognizer_;
   lstm_recognizer_ = nullptr;
 }
@@ -535,6 +538,7 @@ void Tesseract::Clear() {
 #endif
     pixa_debug__.WritePNGs(file_path.c_str());
   }
+  pix_original_.destroy();
   pixa_debug__.Clear();
   pix_binary_.destroy();
   pix_grey_.destroy();

src/ccstruct/debugpixa.h

@@ -40,7 +40,7 @@ namespace tesseract {
       int color = depth < 8 ? 1 : (depth > 8 ? 0x00ff0000 : 0x80);
       Image pix_debug =
         pixAddSingleTextblock(pix, fonts_, caption, color, L_ADD_BELOW, nullptr);
-      if (keep_a_copy)
+      if (!keep_a_copy)
         pix.destroy();
       pixaAddPix(pixa_, pix_debug, L_INSERT);
 #endif

src/ccstruct/image.cpp

@@ -39,6 +39,14 @@ bool Image::isZero() const {
   return r == 1;
 }
 
+Image& Image::operator =(Pix* pix) {
+  if (pix_ != nullptr) {
+    pixDestroy(&pix_);
+  }
+  pix_ = pix;
+  return *this;
+}
+
 Image Image::operator|(Image i) const {
   return pixOr(nullptr, pix_, i);
 }

src/ccstruct/image.h

@@ -37,12 +37,14 @@ class TESS_API Image {
   operator Pix *() const { return pix_; }
   explicit operator Pix **() { return &pix_; }
   Pix *operator->() const { return pix_; }
+  Image& operator =(Pix* pix);
 
   // api
   Image clone() const; // increases refcount
   Image copy() const;  // does full copy
   void destroy();
   bool isZero() const;
+  void replace(Pix* pix);
 
   // ops
   Image operator|(Image) const;

src/dict/dict.cpp

@@ -385,11 +385,13 @@ void Dict::End() {
       delete dawg;
     }
   }
-  dawg_cache_->FreeDawg(bigram_dawg_);
+  if (dawg_cache_ != nullptr) {
+    dawg_cache_->FreeDawg(bigram_dawg_);
+  }
   if (dawg_cache_is_ours_) {
     delete dawg_cache_;
-    dawg_cache_ = nullptr;
   }
+  dawg_cache_ = nullptr;
   for (auto successor : successors_) {
     delete successor;
   }

src/textord/colfind.cpp

@@ -189,7 +189,8 @@ void ColumnFinder::SetupAndFilterNoise(PageSegMode pageseg_mode, Image photo_mas
       //boxaDestroy(BOXA * *pboxa);
       l_uint32 bordercolor;
       composeRGBAPixel(255, 32, 32, 255, &bordercolor);
-      pixDrawBoxa(pix, boxlist, 2, bordercolor);
+      pix = pixDrawBoxa(pix, boxlist, 2, bordercolor);
+      boxaDestroy(&boxlist);
 
       input_block->plot_graded_blobs(pix);
 
@@ -429,7 +430,8 @@ int ColumnFinder::FindBlocks(PageSegMode pageseg_mode, Image scaled_color, int s
         //boxaDestroy(BOXA * *pboxa);
         l_uint32 bordercolor;
         composeRGBAPixel(255, 32, 32, 255, &bordercolor);
-        pixDrawBoxa(pix, boxlist, 2, bordercolor);
+        pix = pixDrawBoxa(pix, boxlist, 2, bordercolor);
+        boxaDestroy(&boxlist);
 
         input_block->plot_graded_blobs(pix);
 
@@ -479,7 +481,8 @@ int ColumnFinder::FindBlocks(PageSegMode pageseg_mode, Image scaled_color, int s
       //boxaDestroy(BOXA * *pboxa);
       l_uint32 bordercolor;
       composeRGBAPixel(255, 32, 32, 255, &bordercolor);
-      pixDrawBoxa(pix, boxlist, 2, bordercolor);
+      pix = pixDrawBoxa(pix, boxlist, 2, bordercolor);
+      boxaDestroy(&boxlist);
 
       part_grid_.DisplayBoxes(pix);
       DisplayTabVectors(pix);
@@ -544,7 +547,8 @@ int ColumnFinder::FindBlocks(PageSegMode pageseg_mode, Image scaled_color, int s
         //boxaDestroy(BOXA * *pboxa);
         l_uint32 bordercolor;
         composeRGBAPixel(255, 32, 32, 255, &bordercolor);
-        pixDrawBoxa(pix, boxlist, 2, bordercolor);
+        pix = pixDrawBoxa(pix, boxlist, 2, bordercolor);
+        boxaDestroy(&boxlist);
 
         part_grid_.DisplayBoxes(pix);
         if (!textord_debug_printable) {
@@ -660,7 +664,8 @@ void ColumnFinder::DisplayBlocks(BLOCK_LIST *blocks) {
     //boxaDestroy(BOXA * *pboxa);
     l_uint32 bordercolor;
     composeRGBAPixel(255, 32, 32, 255, &bordercolor);
-    pixDrawBoxa(pix, boxlist, 2, bordercolor);
+    pix = pixDrawBoxa(pix, boxlist, 2, bordercolor);
+    boxaDestroy(&boxlist);
 
     DisplayBoxes(pix);
     BLOCK_IT block_it(blocks);
@@ -704,7 +709,8 @@ void ColumnFinder::DisplayColumnBounds(PartSetVector *sets) {
     //boxaDestroy(BOXA * *pboxa);
     l_uint32 bordercolor;
     composeRGBAPixel(255, 32, 32, 255, &bordercolor);
-    pixDrawBoxa(pix, boxlist, 2, bordercolor);
+    pix = pixDrawBoxa(pix, boxlist, 2, bordercolor);
+    boxaDestroy(&boxlist);
 
     DisplayBoxes(pix);
     //col_win->Pen(textord_debug_printable ? ScrollView::BLUE : ScrollView::GREEN);

src/textord/tabfind.cpp

@@ -495,7 +495,8 @@ void TabFind::TidyBlobs(TO_BLOCK *block) {
       //boxaDestroy(BOXA * *pboxa);
       l_uint32 bordercolor;
       composeRGBAPixel(255, 32, 32, 255, &bordercolor);
-      pixDrawBoxa(pix, boxlist, 2, bordercolor);
+      pix = pixDrawBoxa(pix, boxlist, 2, bordercolor);
+      boxaDestroy(&boxlist);
 
       block->plot_graded_blobs(pix);
       block->plot_noise_blobs(pix);