feat: allow analyze_page to accept pre-computed OCR lines

Co-Authored-By: Claude Opus 4.6 <[email protected]>

tests/test_analyze_page.py

@@ -315,3 +315,36 @@ async def test_analyze_page_font_caching():
     # But both redactions should share the same font
     assert result.redactions[0].font is font_match
     assert result.redactions[1].font is font_match
+
+
+# ---------------------------------------------------------------------------
+# test_analyze_page_uses_precomputed_ocr_lines
+# ---------------------------------------------------------------------------
+
+@pytest.mark.anyio
+async def test_analyze_page_uses_precomputed_ocr_lines():
+    """When ocr_lines is provided, analyze_page should use them directly
+    without calling ocr_page.  The returned lines must be the same object."""
+    line = _make_line_from_words(["Hello", "world"])
+    precomputed_lines = [line]
+    page_image = _dummy_page_image()
+
+    mock_ocr = patch(
+        "unredact.pipeline.analyze_page.ocr_page",
+        return_value=[],  # should never be called
+    )
+
+    with (
+        mock_ocr as m_ocr,
+        patch(
+            "unredact.pipeline.analyze_page.detect_redactions_llm",
+            new_callable=AsyncMock,
+            return_value=[],
+        ),
+    ):
+        result = await analyze_page(page_image, ocr_lines=precomputed_lines)
+
+    # ocr_page must not have been called
+    m_ocr.assert_not_called()
+    # The returned lines must be the exact same object (not a copy)
+    assert result.lines is precomputed_lines

unredact/pipeline/analyze_page.py

@@ -45,6 +45,7 @@ class PageAnalysis:
 async def analyze_page(
     page_image: Image.Image,
     on_progress: callable | None = None,
+    ocr_lines: list[OcrLine] | None = None,
 ) -> PageAnalysis:
     """Run the full per-page analysis pipeline.
 
@@ -61,12 +62,17 @@ async def analyze_page(
         page_image: PIL Image of the document page.
         on_progress: Optional callback ``(event_name, data_dict)`` for
             progress reporting.
+        ocr_lines: Optional pre-computed OCR lines.  When provided the
+            OCR step is skipped and these lines are used directly.
 
     Returns:
         PageAnalysis with OCR lines and redaction analysis results.
     """
-    # Step 1: OCR (blocking — run in thread)
-    lines: list[OcrLine] = await asyncio.to_thread(ocr_page, page_image)
+    # Step 1: OCR (blocking — run in thread), or use pre-computed lines
+    if ocr_lines is not None:
+        lines = ocr_lines
+    else:
+        lines: list[OcrLine] = await asyncio.to_thread(ocr_page, page_image)
 
     if on_progress:
         on_progress("ocr_done", {"line_count": len(lines)})