Fixes batch 2

tests/parser/file/test_anydoc_parser.py

@@ -13,6 +13,7 @@ Three behaviours matter and are asserted separately:
 import importlib.machinery
 import sys
 import types
+from pathlib import Path
 
 import pytest
 
@@ -319,9 +320,7 @@ def test_init_parser_imports_for_real_not_just_find_spec(monkeypatch):
 
 # --- PDF trust check + tableize wiring (PR 3) -----------------------------------
 
-from pathlib import Path as _Path
-
-FIXTURES = _Path(__file__).parent / "fixtures"
+FIXTURES = Path(__file__).parent / "fixtures"
 CID_PDF = FIXTURES / "nda_en_zh_cid_font.pdf"
 
 

tests/parser/file/test_docling_parser.py

@@ -158,8 +158,12 @@ class TestOcrEngineSelection:
 
     def test_build_tesseract_reads_ocr_langs(self, settings, monkeypatch):
         pytest.importorskip("docling")
+        import application.parser.file.ocr_parser as op
         from application.parser.file.docling_parser import _build_ocr_options
 
+        # Pack inventory unknown: the resolved list is passed through untouched
+        # (a host with tesseract but no chi_sim pack would otherwise drop it).
+        monkeypatch.setattr(op, "tesseract_languages", lambda: None)
         monkeypatch.setattr(settings, "OCR_LANGS", "eng+chi_sim")
         options = _build_ocr_options("tesseract", None, True)
 

tests/parser/file/test_ocr_parser.py

@@ -964,3 +964,38 @@ def test_delegate_parse_lets_setup_errors_through():
 
     with pytest.raises(ImportError, match="docling is required"):
         delegate_parse(_NeedsLib(), Path("x.pdf"), "ignore")
+
+
+class TestTextLayerDelegate:
+    """``text_layer_delegate`` mirrors ``parse_file``'s delegation decision."""
+
+    def test_every_page_with_text_returns_the_text_parser(self, monkeypatch):
+        import pypdfium2 as pdfium
+
+        class _Pdf:
+            def close(self):
+                pass
+
+        monkeypatch.setattr(pdfium, "PdfDocument", lambda path: _Pdf())
+        text_parser = object()
+        parser = op.NativeOcrPdfParser(text_parser=text_parser, min_text_chars=20)
+        monkeypatch.setattr(parser, "_text_layer_counts", lambda pdf: [50, 60])
+        assert parser.text_layer_delegate(Path("x.pdf")) is text_parser
+
+    def test_a_scanned_page_keeps_the_native_path(self, monkeypatch):
+        import pypdfium2 as pdfium
+
+        class _Pdf:
+            def close(self):
+                pass
+
+        monkeypatch.setattr(pdfium, "PdfDocument", lambda path: _Pdf())
+        parser = op.NativeOcrPdfParser(text_parser=object(), min_text_chars=20)
+        monkeypatch.setattr(parser, "_text_layer_counts", lambda pdf: [50, 3])
+        assert parser.text_layer_delegate(Path("x.pdf")) is None
+
+    def test_no_text_parser_or_unreadable_file_is_none(self, tmp_path):
+        assert op.NativeOcrPdfParser().text_layer_delegate(Path("x.pdf")) is None
+        broken = tmp_path / "broken.pdf"
+        broken.write_bytes(b"not a pdf")
+        assert op.NativeOcrPdfParser(text_parser=object()).text_layer_delegate(broken) is None

tests/parser/file/test_pdf_trust.py

@@ -182,6 +182,36 @@ class TestStreamWalk:
         assert result["type0_no_tounicode"] == 1
 
 
+class TestObjectWalk:
+    def test_bodies_match_the_regex_reference(self):
+        """The forward walk yields exactly what ``\\d+\\s+\\d+\\s+obj(.*?)endobj`` did."""
+        import re
+
+        from application.parser.file.pdf_trust import _object_bodies
+
+        reference = re.compile(rb"\d+\s+\d+\s+obj(.*?)endobj", re.DOTALL)
+        data = (
+            TYPE0_BARE
+            + b"7 0 obj no terminator here "  # orphan header: body runs to the next endobj
+            + SIMPLE_FONT
+            + b"junk 12 0 obj<</A/B>>endobj"
+            + _stream(b"x")
+            + b"99 0 obj trailing orphan"
+        )
+        assert list(_object_bodies(data)) == [m.group(1) for m in reference.finditer(data)]
+
+    def test_orphan_headers_are_linear(self):
+        """Headers with no ``endobj`` after them: the old regex rescanned to EOF for
+        each one (2000 in 1 MB took 13.5 s), pinning an ingest worker per upload."""
+        import time
+
+        data = TYPE0_BARE + b"1 0 obj\n" * 20_000 + b"\x00" * 500_000
+        started = time.perf_counter()
+        result = check_pdf_fonts(data)
+        assert time.perf_counter() - started < 1.0
+        assert result["type0_no_tounicode"] == 1
+
+
 # --- the CJK cross-check is confirmed against pdfium's text layer --------------