On Fri, Jun 4, 2021 at 4:42 PM Ævar Arnfjörð Bjarmason <avarab@xxxxxxxxx> wrote: > On Thu, Jun 03 2021, Jeff King wrote: > I have some incomplete work somewhere to slurp up all the package > sources I could find in the wild (SRPM's, Debian recipies etc.) and > their patches, the aim was to submit it into contrib/ so we could see > what monkeypatches to git.git existed in the wild. > > Last time I looked at those (and this is from memory, and it was a while > ago) many of those patches / build recipies were simply blindly skipping > or otherwise working around test failures. > > So we can't assume that failures in the wild are reported to us, and I > think many packagers are not running any of our tests at all. If it > compiles and seems to work they're probably just shipping it. Arch Linux boasts of following upstream as closely as possible, I looked at their build instructions for the umpteenth time to see if I could find anything interesting. While they don't have patches, they do have NO_SVN_TESTS=y (plus a bunch of `cp contrib/*`). So I don't think any packager can just blindly trust the Git project to do the sensible thing here. Who uses subversion anymore? And who that still uses subversion wants or needs git-svn? And why would failing any of these break the release? Maybe we need some "test-packager" target that limits the tests to something sensible. Ideally it should be the other way around: packagers run `make test`, and developers `make test-developer`. But that's a bigger discussion. Cheers. -- Felipe Contreras
