Merge pull request #399 from christianlouis/copilot/fix-ocr-language-data-issue

fix(ocr): ensure Tesseract language data is available for embed_text_layer regardless of active OCR provider

app/utils/ocr_language_manager.py

@@ -279,6 +279,12 @@ def ensure_ocr_languages_from_settings() -> dict[str, list[str]]:
     present.  Missing Tesseract tessdata files are downloaded automatically;
     missing EasyOCR models are downloaded via the library's built-in mechanism.
 
+    Tesseract language data is always ensured when ``ocrmypdf`` is available on
+    the system, regardless of which OCR providers are active.  This is required
+    because :func:`~app.utils.ocr_provider.embed_text_layer` uses ``ocrmypdf``
+    (and therefore Tesseract) as a post-processing fallback for **all** OCR
+    providers – not only the ``tesseract`` provider.
+
     This function is idempotent – calling it multiple times is safe.
 
     Returns:
@@ -297,7 +303,12 @@ def ensure_ocr_languages_from_settings() -> dict[str, list[str]]:
     providers_raw = getattr(settings, "ocr_providers", None) or "azure"
     active_providers = {p.strip().lower() for p in providers_raw.split(",") if p.strip()}
 
-    if "tesseract" in active_providers:
+    # Always ensure Tesseract language data when ocrmypdf is on the system PATH.
+    # embed_text_layer() calls ocrmypdf as a text-layer post-processor for every
+    # OCR provider that does not natively produce a searchable PDF (azure,
+    # easyocr, mistral, google_docai, aws_textract).  If the tessdata files are
+    # absent, ocrmypdf exits with code 3 and the text layer is silently skipped.
+    if "tesseract" in active_providers or shutil.which("ocrmypdf") is not None:
         lang_str = getattr(settings, "tesseract_language", None) or "eng"
         logger.info("Ensuring Tesseract language data for configured languages: %s", lang_str)
         result["tesseract_missing"] = ensure_tesseract_languages(lang_str)

app/utils/ocr_provider.py

@@ -93,6 +93,19 @@ def embed_text_layer(input_pdf_path: str, output_pdf_path: str, *, language: str
         )
         return False
 
+    # Ensure Tesseract language data is present before invoking ocrmypdf.
+    # ocrmypdf uses Tesseract internally regardless of which OCR provider is
+    # active, so we must guarantee the tessdata files exist here.
+    from app.utils.ocr_language_manager import ensure_tesseract_languages  # noqa: PLC0415
+
+    missing_langs = ensure_tesseract_languages(language)
+    if missing_langs:
+        logger.warning(
+            "[embed_text_layer] Missing Tesseract language data for: %s – "
+            "text-layer embedding may fail or produce degraded results.",
+            ", ".join(missing_langs),
+        )
+
     in_place = input_pdf_path == output_pdf_path
     if in_place:
         import tempfile

tests/test_ocr_language_manager.py

@@ -344,7 +344,10 @@ class TestEnsureOCRLanguagesFromSettings:
         assert result == {"tesseract_missing": [], "easyocr_failed": []}
 
     def test_runs_easyocr_check_when_provider_active(self):
-        """Calls ensure_easyocr_models when 'easyocr' is in ocr_providers."""
+        """Calls ensure_easyocr_models when 'easyocr' is in ocr_providers.
+
+        When ocrmypdf is absent, only easyocr language checks run.
+        """
         from app.utils.ocr_language_manager import ensure_ocr_languages_from_settings
 
         mock_settings = self._mock_settings(providers="easyocr", easyocr_langs="en,de")
@@ -352,6 +355,7 @@ class TestEnsureOCRLanguagesFromSettings:
             patch("app.utils.ocr_language_manager.ensure_tesseract_languages", return_value=[]) as mock_tess,
             patch("app.utils.ocr_language_manager.ensure_easyocr_models", return_value=[]) as mock_easy,
             patch("app.config.settings", mock_settings),
+            patch("shutil.which", return_value=None),
         ):
             result = ensure_ocr_languages_from_settings()
 
@@ -359,8 +363,8 @@ class TestEnsureOCRLanguagesFromSettings:
         mock_easy.assert_called_once_with(["en", "de"])
         assert result["easyocr_failed"] == []
 
-    def test_skips_both_when_only_azure(self):
-        """Neither Tesseract nor EasyOCR checks run when only Azure is configured."""
+    def test_skips_both_when_only_azure_and_no_ocrmypdf(self):
+        """Neither Tesseract nor EasyOCR checks run when only Azure is configured and ocrmypdf is absent."""
         from app.utils.ocr_language_manager import ensure_ocr_languages_from_settings
 
         mock_settings = self._mock_settings(providers="azure")
@@ -368,6 +372,7 @@ class TestEnsureOCRLanguagesFromSettings:
             patch("app.utils.ocr_language_manager.ensure_tesseract_languages") as mock_tess,
             patch("app.utils.ocr_language_manager.ensure_easyocr_models") as mock_easy,
             patch("app.config.settings", mock_settings),
+            patch("shutil.which", return_value=None),
         ):
             result = ensure_ocr_languages_from_settings()
 
@@ -375,6 +380,27 @@ class TestEnsureOCRLanguagesFromSettings:
         mock_easy.assert_not_called()
         assert result == {"tesseract_missing": [], "easyocr_failed": []}
 
+    def test_runs_tesseract_when_ocrmypdf_available_with_non_tesseract_provider(self):
+        """Tesseract language check runs when ocrmypdf is on PATH even without the tesseract provider.
+
+        embed_text_layer() uses ocrmypdf internally regardless of the active OCR
+        provider.  Pre-downloading language data on startup prevents code-3 failures.
+        """
+        from app.utils.ocr_language_manager import ensure_ocr_languages_from_settings
+
+        mock_settings = self._mock_settings(providers="azure", tesseract_lang="eng+deu+fra")
+        with (
+            patch("app.utils.ocr_language_manager.ensure_tesseract_languages", return_value=[]) as mock_tess,
+            patch("app.utils.ocr_language_manager.ensure_easyocr_models") as mock_easy,
+            patch("app.config.settings", mock_settings),
+            patch("shutil.which", return_value="/usr/local/bin/ocrmypdf"),
+        ):
+            result = ensure_ocr_languages_from_settings()
+
+        mock_tess.assert_called_once_with("eng+deu+fra")
+        mock_easy.assert_not_called()
+        assert result == {"tesseract_missing": [], "easyocr_failed": []}
+
     def test_runs_both_when_both_providers_active(self):
         """Both checks run when both 'tesseract' and 'easyocr' are in ocr_providers."""
         from app.utils.ocr_language_manager import ensure_ocr_languages_from_settings

tests/test_ocr_processing.py

@@ -1075,6 +1075,7 @@ class TestEmbedTextLayer:
 
         with (
             patch("shutil.which", return_value="/usr/bin/ocrmypdf"),
+            patch("app.utils.ocr_language_manager.ensure_tesseract_languages", return_value=[]),
             patch("subprocess.run", return_value=mock_proc) as mock_run,
         ):
             result = embed_text_layer(pdf, output, language="eng")
@@ -1097,6 +1098,7 @@ class TestEmbedTextLayer:
 
         with (
             patch("shutil.which", return_value="/usr/bin/ocrmypdf"),
+            patch("app.utils.ocr_language_manager.ensure_tesseract_languages", return_value=[]),
             patch("subprocess.run", return_value=mock_proc),
         ):
             result = embed_text_layer(pdf, str(tmp_path / "out.pdf"))
@@ -1113,6 +1115,7 @@ class TestEmbedTextLayer:
 
         with (
             patch("shutil.which", return_value="/usr/bin/ocrmypdf"),
+            patch("app.utils.ocr_language_manager.ensure_tesseract_languages", return_value=[]),
             patch("subprocess.run", side_effect=subprocess.TimeoutExpired(cmd="ocrmypdf", timeout=600)),
         ):
             result = embed_text_layer(pdf, str(tmp_path / "out.pdf"))
@@ -1140,6 +1143,7 @@ class TestEmbedTextLayer:
 
         with (
             patch("shutil.which", return_value="/usr/bin/ocrmypdf"),
+            patch("app.utils.ocr_language_manager.ensure_tesseract_languages", return_value=[]),
             patch("subprocess.run", side_effect=fake_run),
         ):
             result = embed_text_layer(pdf, pdf)
@@ -1172,6 +1176,7 @@ class TestEmbedTextLayer:
 
         with (
             patch("shutil.which", return_value="/usr/bin/ocrmypdf"),
+            patch("app.utils.ocr_language_manager.ensure_tesseract_languages", return_value=[]),
             patch("subprocess.run", return_value=mock_proc),
             patch("tempfile.mkstemp", side_effect=fake_mkstemp),
         ):