docling separation

tests/parser/file/test_docling_parser.py

@@ -910,6 +910,12 @@ class TestApplyInferenceSettings:
     and buys nothing for one-shot parses.
     """
 
+    @pytest.fixture(autouse=True)
+    def _requires_docling(self):
+        # These tests reach into docling.datamodel.settings for real; a base
+        # install (docling is an optional extra) skips them.
+        pytest.importorskip("docling")
+
     def test_disables_torch_compile_by_default(self, monkeypatch):
         from application.parser.file.docling_parser import _apply_inference_settings
 
@@ -1412,6 +1418,12 @@ class TestOCRDropoutHelpers:
 class TestForceFullPageOCRWiring:
     """``force_full_page_ocr`` must reach the pipeline for any OCR engine."""
 
+    @pytest.fixture(autouse=True)
+    def _requires_docling(self):
+        # Builds real ``docling.datamodel.pipeline_options``; a base install
+        # (docling is an optional extra) skips these.
+        pytest.importorskip("docling")
+
     def _pipeline_options(self, monkeypatch, **kwargs):
         import docling.datamodel.pipeline_options as dpo
 

tests/parser/file/test_pdfium_parser.py

@@ -158,6 +158,10 @@ def test_threshold_sends_sparse_text_to_fallback(tmp_path, parser_cls):
 
 @pytest.fixture
 def docling_engine(monkeypatch):
+    # The wiring assertions below expect the docling map; on a base install
+    # (docling is an optional extra) the engine degrades to the legacy
+    # parsers and these tests do not apply.
+    pytest.importorskip("docling")
     from application.core.settings import settings
 
     monkeypatch.setattr(settings, "DOC_PARSER_ENGINE", "docling")

tests/requirements.txt

@@ -12,3 +12,6 @@ jupyter-client>=8.0
 # (python-pptx and openpyxl already ship in application/requirements.txt).
 python-docx>=1.1
 reportlab>=4.0,<5
+# The optional docling engine stays installed in the test env so its parser
+# paths keep CI coverage even though production installs it on demand.
+-r ../application/requirements-docling.txt