Docs: Appendix 4 (Performance Comparison) benchmark files can't be downloaded — table renders no links, and all 8 artifex.com/samples URLs return 404

[fstandhartinger]

Description of the bug

Documentation bug, not a library bug. Appendix 4: Performance Comparison Methodology, section "Files used", says each file comes with its "download link". Nobody can actually download the eight benchmark files, for two separate reasons:

1. The table renders no links. In docs/app4.rst the names are written as `adobe.pdf` (interpreted text) rather than `adobe.pdf`_, so the eight targets defined at the bottom of the file (lines 470–477) are never used. The rendered page has zero hrefs to artifex.com/samples/pdf/.

2. The targets are gone anyway. All eight return 404 today, and so do https://artifex.com/samples/pdf/ and https://artifex.com/samples/:

   https://artifex.com/samples/pdf/adobe.pdf                  404
   https://artifex.com/samples/pdf/artifex-website.pdf        404
   https://artifex.com/samples/pdf/db-systems.pdf             404
   https://artifex.com/samples/pdf/fontforge.pdf              404
   https://artifex.com/samples/pdf/pandas.pdf                 404
   https://artifex.com/samples/pdf/pymupdf.pdf                404
   https://artifex.com/samples/pdf/pythonbook.pdf             404
   https://artifex.com/samples/pdf/sample-50-MB-pdf-file.pdf  404
   

So the published timings can't currently be reproduced from the docs.

How to reproduce the bug

curl -sL https://pymupdf.readthedocs.io/en/latest/app4.html | grep -c 'artifex.com/samples'   # 0
curl -s -o /dev/null -w '%{http_code}\n' https://artifex.com/samples/pdf/adobe.pdf           # 404

The files do still exist in the Wayback Machine, byte for byte. I checked each archived copy's Content-Length against the "Size (bytes)" column of the table, and all eight match exactly:

File	Size in table	Archived copy (raw, id_)
adobe.pdf	32,472,771	20230327230046 → 32,472,771
artifex-website.pdf	31,570,732	20230517230115 → 31,570,732
db-systems.pdf	29,326,355	20240421152039 → 29,326,355
fontforge.pdf	8,222,384	20230517230115 → 8,222,384
pandas.pdf	10,585,962	20240423040027 → 10,585,962
pymupdf.pdf	6,805,176	20240421152031 → 6,805,176
pythonbook.pdf	9,983,856	20230517230115 → 9,983,856
sample-50-MB-pdf-file.pdf	52,521,850	20240421183811 → 52,521,850

Possible fixes, your call which:

• re-upload the eight files wherever Artifex now hosts samples, and add the missing trailing _ in the table so the links render; or
• point the eight targets at the archived copies above (and add the _).

I didn't open a PR because the first option needs someone with access to artifex.com. If you'd rather have the second, I'm happy to send one.

PyMuPDF version

Not version-specific: this is in the docs. Checked against main at a7ff9a3 and the live /en/latest/ page on 2026-09-19.

Operating system

n/a (checked from Linux)

Python version

n/a

---

Disclosure: I build a hosted HTML-to-PDF API, which has nothing to do with this issue. I found the dead links while running a link check over the docs of several PDF libraries. An AI assistant (Claude Opus 5, Anthropic) ran that check and helped me write this issue. Every status code and byte count above was measured before posting.