Upcoming changes in pypdfium2 & suggestions to improve PDF rendering integration

[mara004]

Hello,

I'm the maintainer of pypdfium2 and found your interesting library using sourcegraph. It's amazing to see pypdfium2 in practical use!

The PDF backend code of doctr seems to be defined in io/pdf.py. I took a quick look and have some considerations:

• pypdfium2 does not support file paths as pathlib.Path anymore. I initially had support for pathlib but then switched back to os.path for various reasons. This means you should probably call the rendering function with str(file) if file is a path.
• At the time your PDF backend was implemented, the support model of pypdfium2 could only return PIL.Image.Image. I recently changed the API to allow getting the data as bytes: render_pdf() is replaced by render_pdf_topil() and render_pdf_tobytes().
• nit: There is a code comment that still assumes use of pymupdf: # Read pages with fitz and convert them to numpy ndarrays

Kind regards

[fg-mindee]

Hi @mara004 👋

Thanks a lot for the heads up!

• casting it to a string is a very viable option 👍
• glad the library keeps on improving! I suggest waiting for your next stable release on pypi to make this change :)
• good catch!

Happy to to discuss this further if needed!