PyPI Now Rejects V7-Format Source Tarballs, and Plain GNU Tar Is the Next Target
Software / news
PyPI Now Rejects V7-Format Source Tarballs, and Plain GNU Tar Is the Next Target
uv 0.13 broke on a 2016 PyPDF2 release the same day the PyPI change merged, and a maintainer puts the non-pax long tail at roughly 10% of the top 15,000.
PyPI now refuses source distributions packed as V7 tar archives, and a follow-up filed on Nov. 18 would extend the ban to every tarball that is not in pax format. The first rule landed in pypi/warehouse pull request #20829 on Oct. 27. The second is issue #21012, still open and without comments.
The change is described in a PyPI blog post dated Nov. 18. It says rejections began "over the last few weeks" and that the "overwhelming majority" of sdists are unaffected because they already follow the pax requirement. It gives no count of rejected uploads.
What PyPI rejects as of Nov. 18, 2026
Two kinds of upload now fail. The first is an sdist whose tar archive is V7 style, a pre-POSIX format from UNIX V7 that Python's own tarfile module never produces. The second is a wheel carrying an archive-level ZIP comment that a parser could mistake for a ZIP control structure.
The blog post says both checks build on ZIP differential work from August 2025 by Seth Larson. Downstream users see no change: PyPI says existing dependencies installed with pip install or uv add keep working.
| Upload | PyPI today | Where it is tracked |
|---|---|---|
| V7-style tar sdist | Rejected | PR #20829, merged Oct. 27 |
| GNU-style tar sdist | Accepted, counted in an internal metric | Issue #21012, opened Nov. 18 |
| Wheel with archive-level ZIP comment | Mostly rejected, a small subset allowed | Issue #20934 |
| Pax tar sdist, valid ZIP wheel | Accepted | None |
Issue #21012 and the 10% estimate
The issue was opened by William Woodruff, who also wrote the V7 pull request. He argues that "we should probably reject anything that isn't a pax tar, rather than specifically sussing out V7."
He gives no exact count. Citing outside statistics, he estimates that the long tail is "roughly ~10% of the top 15k" projects and says the top 360 look "pretty good." The issue does not say which statistics, so treat the figure as an estimate from one maintainer.
He expects a long warning and deprecation period, with emails to affected uploaders, using the machinery from the V7 pull request. The sdist specification already forbids GNU-style archives. PyPI has kept accepting them because some projects still produce them.
The uv break on PyPDF2 1.26.0
The same day PR #20829 merged, a user opened astral-sh/uv issue #22955. On uv 0.13.6, Linux x86_64 and Python 3.11, installing pypdf2==1.26.0 failed with "invalid tar identity: found [0, 0, 0, 0, 0, 0, 0, 0]". It worked on uv 0.12.0 and fails from 0.13.0 through 0.13.6.
Woodruff replied in that thread that the sdist is invalid, because the packaging spec calls for pax tarballs and "V7 tar doesn't really have a specification." He added that PyPDF2 1.26.0 dates from 2016. The issue is open and labelled needs-decision. Setting UV_LEGACY_TAR_BACKEND=1 restores extraction in the meantime.
That is the practical edge of the policy: PyPI blocks new uploads, but an old release already on the index can still trip an installer that has tightened its own parser. The sources describe upload rules and do not say that any existing release is removed.
What maintainers should change
PyPI's advice is to build with a Python build backend instead of packing archives by hand. Popular backends produce conforming files. Anyone packing manually has four rules from the post:
- Sdists must be POSIX pax tar archives; with GNU
tar, pass--format=pax. - Sdists must avoid sparse members and pax vendor extensions.
- Wheels must be valid ZIP archives.
- Member filenames in both formats must be UTF-8.
PyPI's archives documentation holds the current requirements.
Related reading on this site: the ONNX Runtime repository that returned 404 on GitHub while 1.32.0 stayed downloadable from PyPI, and the Optimizely CDN incident, another case of a distribution channel carrying something its owners did not intend.
Nothing in #21012 has a date. The next signal is a maintainer decision on the issue, and the first user-facing warning email for GNU tar uploads.
Sources
More in Software
- 01OpenAI's WebRTC Voice Stack Draws a 30 ms Objection That Its Own Engineers Never AddressedLuke Curley says WebRTC drops audio older than 30 ms. OpenAI's engineers wrote about port ranges and routing, which is a different problem.
- 02Superpowers Passes 322,000 GitHub Stars as a Set of 15 Prompted Skills for 16 Coding AgentsJesse Vincent's MIT-licensed framework tells agents to brainstorm, plan and test first. Its clone count is large, and its evidence for better results is one author's view.
- 03GitHub Disables 144 Picofly Repositories After a Nintendo Notice Names OneThe November 6 notice is the first on GitHub aimed at Switch modchip firmware, and it asked for forks to be taken down without listing a single one.
- 04Chrome's Gemini Nano Download: What Google's Docs Say, and the Numbers They Leave OutGoogle's page on built-in model management gives triggers, a 30-day rule and full-download updates, but no file size, no disk threshold and no consumer off switch.