| Age | Commit message (Collapse) | Author |
|---|
|
It is a long-standing disagreement with lintian. Until now, the hinter
emitted hints when some package could plausibly be marked M-A:same even
when dependencies couldn't. lintian complains about that when such a
dependency comes from the same source package. We now follow lintian.
This deletes around 2000 hints and will cause people to have to apply
hints multiple times. For instance your libfoo-dev package will only be
flagged after your libfooN was flagged. So be it.
|
|
binNMUed packages are not currently reproducible, because buildds don't
pass --binNMU-timestamp to sbuild. Thus they use varying
SOURCE_DATE_EPOCH and produce faulty packages. As much as this is a real
bug, it is not actionable by maintainers. Hide such issues for now.
Link: https://salsa.debian.org/perl-team/modules/packages/libtie-hash-indexed-perl/-/merge_requests/1
Link: https://bugs.debian.org/843773
|
|
|
|
When an arch:any package ships a .so file in a public library search
path (e.g. a symlink as many lib*-dev packages do) it most likely
shouldn't be M-A:foreign. A common exception is plugins loaded into
programs, so exclude that case.
Many thanks to Johannes Schauer and Guillem Jover for helping discover
this pattern of Multi-Arch: foreign abuse.
|
|
Since all users of archdepcandidate run the results through "exists()"
or "group by", "union" vs "union all" does not make any difference to
the results.
On the performance side however, it avoids a b-tree merge getting the
maforeign_candidate query down from hours to seconds.
|
|
It builds on the core functionality of dedup, but uses a different
database schema. Unlike dedup, it aborts downloading Arch:all packages
early and consumes any other architecture in its entirety instead.
|
