Tags that name a version their commit does not declare

Luiz F Nunes da Silva lfnsilva.invest at gmail.com
Wed Nov 11 00:34:21 UTC 2026


Hi Chris,

Thanks. I looked at how distributions handle this, and have a follow-up
with one correction to my own first count.

nixpkgs does check. For Python modules, pythonMetadataCheckHook is on by
default and fails the build when .dist-info/METADATA disagrees with the
derivation version; versionCheckHook does the same for a program's
--version. The escape hatch, pyprojectVersionPatchHook, refuses to run when
the versions already agree, so its 195 uses are all cases where they
differed. At the pinned GitHub tag, 51 declare a different real version.
But 29 of those 51 have a publish job that writes the version from the tag
(uv version "${GITHUB_REF_NAME#v}" and similar): the file at the tag is
stale on purpose and the PyPI artifact is right. That leaves 22 with no
such step visible, as a floor still to be confirmed one by one. The rest
are placeholders (0.0.0), versions from VCS, or cases the data cannot
decide.

The other distributions I looked at, with the same three-way split, do not
compare the declared version: Debian (lintian compares debian/watch with
the changelog), Arch (devtools compares pkgver with nvchecker), Fedora
(only a zero-version guard), Gentoo and Guix. Homebrew leaves it to an
optional assert_match version in about 28% of formulae. Some upstreams are
patched by more than one distribution independently (caio in Debian and
Arch, cross-web in nixpkgs and Debian).

Everything is public with the scripts and raw data, so it can be reproduced
without me. I'd welcome an independent reproduction, from you or anyone on
the list, especially on the Debian side, where my sample was the first 19
patches in alphabetical order, not a random draw:
https://github.com/luizfnsilva/distro-version-survey-2026-11

Luiz

Em seg., 9 de nov. de 2026 às 20:05, Chris Lamb <lamby at debian.org> escreveu:

> Hey Luiz,
>
> > 1. Do rebuilders or distributions that build from upstream tags check
> >    that the tag and the declared version agree, or is that left to the
> >    upstream?
>
> I'd be curious about the situation for the rolling/functional Linux
> distributions.
> No doubt in Debian (and other distributions with the "source tarball"
> model) there
> are a handful of binary packages of version X that are ~incorrectly
> declaring
> themselves as X+1 internally by some mechanism.
>
> Funnily enough, our own reprotest has a related check to ensure that the
> version number in debian/changelog cannot get out of sync with the Python
> package's version [0].
>
>  [0]
> https://salsa.debian.org/reproducible-builds/reprotest/-/blob/main/debian/rules#L23
>
>
> Regards,
>
> --
>       ,''`.
>      : :'  :     Chris Lamb
>      `. `'`      lamby at debian.org 🍥 chris-lamb.co.uk
>        `-
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.reproducible-builds.org/pipermail/rb-general/attachments/20261110/6d31a4cf/attachment.htm>


More information about the rb-general mailing list