Merge from main

Signed-off-by: Christoph Auer <cau@zurich.ibm.com>

docling/backend/docling_parse_v4_backend.py

@@ -187,7 +187,17 @@ class DoclingParseV4DocumentBackend(PdfDocumentBackend):
 
     def unload(self):
         super().unload()
-        self.dp_doc.unload()
+        # Unload docling-parse document first
-        with pypdfium2_lock:
+        if self.dp_doc is not None:
-            self._pdoc.close()
+            self.dp_doc.unload()
-        self._pdoc = None
+            self.dp_doc = None
+
+        # Then close pypdfium2 document with proper locking
+        if self._pdoc is not None:
+            with pypdfium2_lock:
+                try:
+                    self._pdoc.close()
+                except Exception:
+                    # Ignore cleanup errors
+                    pass
+            self._pdoc = None

docling/models/tesseract_ocr_model.py

@@ -144,7 +144,10 @@ class TesseractOcrModel(BaseOcrModel):
 
                         local_reader = self.reader
                         self.osd_reader.SetImage(high_res_image)
+
+                        doc_orientation = 0
                         osd = self.osd_reader.DetectOrientationScript()
+
                         # No text, or Orientation and Script detection failure
                         if osd is None:
                             _log.error(
@@ -158,11 +161,14 @@ class TesseractOcrModel(BaseOcrModel):
                             # to OCR in the hope OCR will succeed while OSD failed
                             if self._is_auto:
                                 continue
-                        doc_orientation = parse_tesseract_orientation(osd["orient_deg"])
+                        else:
-                        if doc_orientation != 0:
+                            doc_orientation = parse_tesseract_orientation(
-                            high_res_image = high_res_image.rotate(
+                                osd["orient_deg"]
-                                -doc_orientation, expand=True
                             )
+                            if doc_orientation != 0:
+                                high_res_image = high_res_image.rotate(
+                                    -doc_orientation, expand=True
+                                )
                         if self._is_auto:
                             script = osd["script_name"]
                             script = map_tesseract_script(script)

tests/test_backend_docling_parse_v4.py

@@ -46,6 +46,12 @@ def test_text_cell_counts():
                 )
             last_cell_count = len(cells)
 
+            # Clean up page backend after each iteration
+            page_backend.unload()
+
+    # Explicitly clean up document backend to prevent race conditions in CI
+    doc_backend.unload()
+
 
 def test_get_text_from_rect(test_doc_path):
     doc_backend = _get_backend(test_doc_path)
@@ -59,6 +65,10 @@ def test_get_text_from_rect(test_doc_path):
 
     assert textpiece.strip() == ref
 
+    # Explicitly clean up resources
+    page_backend.unload()
+    doc_backend.unload()
+
 
 def test_crop_page_image(test_doc_path):
     doc_backend = _get_backend(test_doc_path)
@@ -70,7 +80,14 @@ def test_crop_page_image(test_doc_path):
     )
     # im.show()
 
+    # Explicitly clean up resources
+    page_backend.unload()
+    doc_backend.unload()
+
 
 def test_num_pages(test_doc_path):
     doc_backend = _get_backend(test_doc_path)
     doc_backend.page_count() == 9
+
+    # Explicitly clean up resources to prevent race conditions in CI
+    doc_backend.unload()