Latest version of pymupdf4llm.to_markdown returns empty text for some PDFs.

[bbfrog]

The old version 0.0.16 works, while the latest version doesn't (pymupdf4llm.to_markdown return empty text). Below is an example PDF, Please help to check and fix. Thanks!

Compass-Pathways-to-announce-fourth-quarter-and-year-end-financial-results-on-February-27th-2025-2025.pdf

[JorjMcKie]

The base library qualifies the text in this PDF as alpha=0 (i.e. invisible). Therefore it is being ignored in this package.
I am going to submit an issue in MuPDF's issue system.

Independently from this, We are going to add an option that allows extracting text even if transparent.

[JorjMcKie]

MuPDF issue link https://bugs.ghostscript.com/show_bug.cgi?id=708584

[JorjMcKie]

The MuPDF team has identified the reason: the use of "Type 3" fonts inevitably leads to alpha=0.
We will introduce an indicator that allows to detect this and react accordingly.

[JorjMcKie]

I have adjusted PyMuPDF4LLM to never ignore text that was written with a Type 3 font. This will resolve this issue.
Contained in version 0.0.25.

[JorjMcKie]

Fixed in version 0.0.25.

[azhurb]

It seems that 0.0.25 doesn't fix the problem for all cases. I have a PDF that works fine with 0.0.17, but I get an empty string on 0.0.24 and 0.0.25. Unfortunately, I can't share the file here, but it has images in the header, and the output of pdffonts is

name                                 type              encoding         emb sub uni object ID
------------------------------------ ----------------- ---------------- --- --- --- ---------
Courier                              Type 1            WinAnsi          no  no  no      22  0
Courier-Bold                         Type 1            WinAnsi          no  no  no      23  0

I can open a new issue if it helps.

[JorjMcKie]

@azhurb - obviously you are reporting a different issue. So, please open a new one.
However, providing a problem file is mandatory, we cannot accept a bug report otherwise.
Either prepare a file without confidentiality concerns that exhibits the problem, or send it to my e-mail address.