Rework the dependencies between excuses.
Dependencies are specified based on packages (source or binary). Based
on these packages, dependencies on other excuses (that have these
packages) are calculated.
As with package dependencies, dependencies between excuses can contain
alternatives.
Each alternative can satisfy the dependency, so the excuse only becomes
invalid if all of the alternatives of a specific dependency are
invalidated.
Tracking of the alternatives and their validity is moved to separate
objects.
Adding a source parameter to the SourcePackage allow finding the source
the object is referring to, so that parameter doesn't have to be passed
along separately.
Signed-off-by: Ivo De Decker <ivodd@debian.org>
Because autopkgtest failure in a package is now a serious bug (RC), packages
that don't opt-in should not be tested. Due to the way autopkgtest and autodep8
work together, package that don't declare they have an autopkgtest can still be
tested. However, maintainers should not be force to say their package doesn't
work with autodep8 by introducing a fake test.
When there is no test in testing, and a failing test in unstable, don't
allow the package to migrate. Doing so would make it instantly RC-buggy.
Signed-off-by: Ivo De Decker <ivodd@debian.org>
If a new pseudo-essential package was added to testing *and* it is
mutually-exclusive with an existing member, britney would incorrectly
flag it as uninstallable (unless another change triggered a cache
invalidation of the pseudo-essential set).
With this commit, the cache invalidation is now done based on whether
we add something that *might* be in the (effective) pseudo-essential
set. It is an overapproximation as there are cases where the
invalidation is unnecessary, but at the moment it is not a performance
concern.
Closes: https://bugs.debian.org/944190
Signed-off-by: Niels Thykier <niels@thykier.net>
The optimizations relies on assumptions that are not valid with the
allow-uninst that has not been corrected. When these assumptions, we
end up with invalid nuninst counters.
Signed-off-by: Niels Thykier <niels@thykier.net>
The spec for Release files says that (at least) one of "Suite" and
"Codename" will be defined. This implies that "Suite" is optional
in some cases and SuiteLoader now correctly falls back to using
"Codename" in that case.
Closes: https://bugs.debian.org/942104
Signed-off-by: Niels Thykier <niels@thykier.net>
When an allow-uninst hint is added for an unversioned binary package, items
are allowed to migrate, even if they make that binary package uninstallable
(on the architecture specified in the hint, if one was specified, or on all
architectures otherwise).
Using this hint should avoid using force-hint to allow specific breakage.
Signed-off-by: Ivo De Decker <ivodd@debian.org>
When do_all fails, and there is no transaction to rollback, we will probably
leave a broken state behind. This codepath probably isn't triggered unless
something else is broken.
Signed-off-by: Ivo De Decker <ivodd@debian.org>
The code that was supposed to check uninstallability of existing binaries was
broken:
- it only checked arch: all on nobreakall archs, not arch any
- it checked if the new binary was in testing (this never happens)
- it doesn't work when binaries from both unstable and tpu are needed
Note that _excuse_unsat_deps() now handles nobreakall correctly.
Ivo De Decker
Paul Gevers
Niels Thykier
Matthias Klumpp
Julien Cristau