feat(thumbnail): render PDF thumbnails with pypdfium2 instead of pdf2image

[Austin-s-h]

Summary

Replaces the pdf2image Python dependency in the thumbnail lambda with pypdfium2 native bindings, extracted into a dedicated pdf_thumbnail.py helper.

• Drops the pdf2image library — the only dependency change is swapping pdf2image for pypdfium2>=4,<5 in pyproject.toml; all other pins and the t4-lambda-shared source are unchanged.
• Poppler is now optional, not required. When pdftoppm/pdfinfo are present on PATH they're still used as a fast path; when they aren't, pypdfium2 renders natively. This is a compatibility improvement (native fallback), not a removal of the Poppler code path — the subprocess route is retained.
• Fixes a file-handle/mmap leak: the PdfDocument and page are closed in try/finally on every render and page-count.
• Configurable render DPI via PDF_PREVIEW_DPI (default 300, clamped to 72–600); pages are resized with LANCZOS, and pdf_render_dpi / pdf_resize_filter are reported in the response info.
• Guards pptx_to_pdf against a missing libreoffice binary with a clean PDFThumbError.

Adds tests/test_pdf_thumbnail.py (11 tests, all passing locally via uv run pytest).

Note

Touches lambdas/thumbnail/src/t4_lambda_thumbnail/__init__.py alongside sibling PR #5107 (raise PDF preview resolution to 2048x1536). A trivial merge-order conflict is expected and resolvable by taking both hunks.

🤖 Generated with Claude Code

Greptile Summary

This PR replaces the pdf2image/Poppler subprocess dependency in the thumbnail lambda with pypdfium2 native bindings, extracted into a new pdf_thumbnail.py helper. A pdftoppm/pdfinfo fast path is preserved when those binaries are present on PATH, with pypdfium2 as the fallback.

• Dependency swap: pdf2image removed from pyproject.toml and uv.lock; pypdfium2>=4,<5 (locked to 4.30.0) added — no Poppler system binary required in the base Lambda image.
• Resource management: PdfDocument and PdfPage are now closed in try/finally blocks on every render and page-count, fixing the per-render file-handle/mmap leak mentioned in the PR description. The intermediate PdfBitmap from render() is not yet wrapped in its own try/finally.
• New features: Configurable render DPI via PDF_PREVIEW_DPI (clamped 72–600, default 300); pdf_render_dpi and pdf_resize_filter added to response info. pptx_to_pdf now raises PDFThumbError on missing libreoffice rather than letting FileNotFoundError propagate.

Confidence Score: 4/5

Safe to merge; the core rendering logic is sound, error handling is improved over the previous implementation, and the 11 new tests cover all key branches.

The migration from pdf2image to pypdfium2 is clean and well-tested. The only items worth addressing are the PdfBitmap not being explicitly closed after to_pil(), and shutil.which() being probed on every request instead of once at module load. Neither affects correctness in CPython reference-counting GC, but worth tightening before shipping to production.

lambdas/thumbnail/src/t4_lambda_thumbnail/pdf_thumbnail.py — bitmap lifecycle and repeated shutil.which() calls

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 49.55%. Comparing base (3d20021) to head (d4ce2f5).
⚠️ Report is 2 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #5108      +/-   ##
==========================================
+ Coverage   49.46%   49.55%   +0.09%     
==========================================
  Files         843      844       +1     
  Lines       34411    34476      +65     
  Branches     5826     5826              
==========================================
+ Hits        17020    17085      +65     
  Misses      15502    15502              
  Partials     1889     1889              

Flag	Coverage Δ
api-python	93.14% <ø> (ø)
catalog	25.08% <ø> (ø)
lambda	97.15% <100.00%> (+0.06%)	⬆️
py-shared	98.02% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[greptile-apps]

The PdfBitmap returned by pdf_page.render() is not closed before the function returns. While CPython's reference counting usually frees the native FPDF_BITMAP allocation immediately, it's better to mirror the pattern already used for pdf_page and document — especially under memory pressure in Lambda where GC timing matters.

Suggested change

	pdf_page = document[page_index]
	try:
	bitmap = pdf_page.render(scale=dpi / 72)
	return bitmap.to_pil()
	finally:
	pdf_page.close()
	pdf_page = document[page_index]
	try:
	bitmap = pdf_page.render(scale=dpi / 72)
	try:
	return bitmap.to_pil()
	finally:
	bitmap.close()
	finally:
	pdf_page.close()

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[greptile-apps]

shutil.which() is called on every invocation of render_pdf_page and count_pdf_pages. In a warm Lambda the binary layout never changes between requests, so re-probing PATH on every call is wasted work. Caching at module load time (using None as a sentinel for "binary absent") avoids the redundant filesystem lookups.

Suggested change

	DEFAULT_PDF_RENDER_DPI = 300
	MAX_PDF_RENDER_DPI = 600
	DEFAULT_PDF_RENDER_DPI = 300
	MAX_PDF_RENDER_DPI = 600
	_PDFTOPPM = shutil.which("pdftoppm") # None when Poppler is not installed
	_PDFINFO = shutil.which("pdfinfo")

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!