"""OCR extraction tests — verify pymupdf-layout OCR works on image-only pages."""
from pathlib import Path
import pymupdf
from zotero_chunk_rag.pdf_processor import extract_document
def _create_scanned_pdf(output_path: Path) -> Path:
"""Create a PDF with one image-only page (text rendered as image).
This simulates a scanned document page. The text 'The quick brown fox
jumps over the lazy dog' is rendered as a raster image and embedded.
"""
doc = pymupdf.open()
# Page 1: normal text (control)
page1 = doc.new_page(width=612, height=792)
page1.insert_text((72, 100), "This is a normal text page.", fontsize=12)
# Page 2: text rendered as image (simulates scan)
page2 = doc.new_page(width=612, height=792)
# Create a temporary doc with text, render to pixmap, insert as image
tmp = pymupdf.open()
tmp_page = tmp.new_page(width=400, height=100)
tmp_page.insert_text(
(20, 50),
"The quick brown fox jumps over the lazy dog",
fontsize=14,
)
pix = tmp_page.get_pixmap(dpi=200)
tmp.close()
# Insert the pixmap as an image on page 2
img_rect = pymupdf.Rect(72, 100, 472, 300)
page2.insert_image(img_rect, pixmap=pix)
doc.save(str(output_path))
doc.close()
return output_path
def test_ocr_extracts_text_from_image_page(tmp_path):
"""The image-only page must produce some text via OCR."""
pdf_path = _create_scanned_pdf(tmp_path / "scanned.pdf")
ex = extract_document(pdf_path)
assert len(ex.pages) == 2
# Page 1 (normal text) should have text
assert len(ex.pages[0].markdown.strip()) > 10, (
"Page 1 (normal text) has no content"
)
# Page 2 (image-only) — if OCR is working, it should extract some text.
# At minimum, some words from "The quick brown fox..." should appear.
page2_text = ex.pages[1].markdown.lower()
# We check for at least 2 words from the phrase to account for OCR errors
ocr_words_found = sum(
1 for w in ["quick", "brown", "fox", "jumps", "lazy", "dog"]
if w in page2_text
)
assert ocr_words_found >= 2, (
f"OCR failed: only {ocr_words_found}/6 expected words found in page 2. "
f"Page 2 text: {ex.pages[1].markdown[:200]!r}"
)
def test_ocr_pages_counted_in_stats(tmp_path):
"""Stats must report ocr_pages > 0 when OCR was used."""
pdf_path = _create_scanned_pdf(tmp_path / "scanned.pdf")
ex = extract_document(pdf_path)
# OCR must work — Tesseract is a required dependency.
# If this fails, the environment is misconfigured, not the code.
assert ex.stats["ocr_pages"] >= 1, (
f"OCR page not counted in stats. ocr_pages={ex.stats['ocr_pages']}. "
f"Page 2 text: {ex.pages[1].markdown[:200]!r}. "
f"Stats: {ex.stats}"
)
