On Sat, May 23, 2026 at 01:08:46PM +0200, Philipp Kern wrote: > For bookworm - but not bookworm-backports - this is correct. Technically the > use of the unshare backend is a per-buildd host setting, but all but one > (mipsel-conova-02.debian.org - which might have been an oversight) have it > set.
Thanks a lot. (The thing about bookworm-backports make sense). > > I ask because after an archive rebuild I found several packages in > > bookworm which FTBFS using the file backend but build ok using the > > unshare backend, at least these five: > > > > libbio-db-refseq-perl > > poetry > > qtnetworkauth-everywhere-src > > ruby-fastimage > > scikit-build > > > > (I have not reported any of them yet because I need to understand > > the issue first). > > > > My current theory so far is that those packages try to use the network > > and they build ok at the time, but now this only works ok if network > > access is disabled (which happens by default when using the unshare > > backend, which would explain why using unshare makes those weird build > > failures to disappear). > > Thanks for doing this, we should definitely get FTBFS bugs for them even if > they build fine on the buildds. Small update: I filed bugs for all the 5 packages above. At least four of them try Internet access as I suspected. This is not very surprising, after all: When the package relies on Internet to build, the build is successful or not depending on the mood of the other end of the connection, and this explains why we did not caught those packages before in the release cycle. Thanks.
