Wrong characters displayed after font subsetting (w/ native method)

[Jeremy-Hibiki]

Description of the bug

We are working on a project to "inplace translate" academic papers and seeing some outputs weird.
Just load, subset and save the paper, and you will find texts with the sans font are unreadable.

The problem only occurs with native subsetting method, while using fonttools can produce normal outputs.

How to reproduce the bug

import fitz
import httpx

fstream = httpx.get("https://arxiv.org/pdf/2504.13181v1").read()
with open("original.pdf", "wb") as f:
    f.write(fstream)
doc = fitz.Document(stream=fstream)
doc.ez_save("before_subset.pdf")
doc.subset_fonts()
doc.ez_save("after_subset.pdf")

https://arxiv.org/pdf/2504.13180
https://arxiv.org/pdf/2504.13181

Both abnormal papers are from Meta, using the same template.

PyMuPDF version

1.25.5

Operating system

Linux

Python version

3.11

[JorjMcKie]

Thanks for the report. This looks like an upstream (MuPDF) problem.

[JorjMcKie]

For convenience, here is the downloaded test file:

original.pdf

[JorjMcKie]

Here is the link to the PDF bug: https://bugs.ghostscript.com/show_bug.cgi?id=708489.

BTW as I wrote there, the fallback logic simply determines that there is nothing to do because all used fonts already are subsets!

[Jeremy-Hibiki]

To clarify, we are embedding translated texts and CJK fonts (which are typically huge) into the PDF, and subsetting them to reduce file size.

We narrowed down the bug to the font subsetting operation, but it looks like it is more than that.

[JorjMcKie]

The MuPDF team is already preparing a fix.

[JorjMcKie]

Just confirmed correct functioning with the latest MuPDF version, specifically commit f35fb0d8c439b28579590547e4b8db3057d7bdd0

[julian-smith-artifex-com]

It looks like there are still some errors after Document.subset_fonts() with latest MuPDF, so this issue is not yet completely solved.

[julian-smith-artifex-com]

The partial fix is in the latest release, PyMuPDF-1.26.0.

[julian-smith-artifex-com]

Uploading test files for reference.
test_4457_b.pdf
test_4457_a.pdf

[JorjMcKie]

@julian-smith-artifex-com - update:

I have tested the two files test_4457_b.pdf/ test_4457_a.pdf again using version 1.26.3 of PyMuPDF (which is bound to MuPDF 1.26.3) and found that font subsets now are produced correctly.

[julian-smith-artifex-com]

Fixed in PyMuPDF-1.26.4.