ppisar commented on the pull-request: `Package tests` that you are following:
``
I also agree that packaging the tests is costly comparing to the fact that they are
already performed at build time. On the other hand the build-time tests are performed
against a code in the working directory which, e.g. by a packaging mistake, can differ
from a code shipped and installed by an RPM package. Hence the advantage of the packaged
tests is that it indeed validates the code Fedora delivers.
It also eases regression testing: You can install the tests subpackage without hassling
with building from sources. That's not so difficult in case of Perl packages, but
prominently soothing for complex C or Rust or Go packages.
The packaged tests also open a possibility to perform a regression testing across multiple
packages. This is currently achieved by Koschei, but Koschei can be quite slow, skips some
architectures and the gap between packages in a build root and a testing repository can be
quite large. There are prospects to execute the tests of the reverse dependencies on
submitting updates into Bodhi. E.g. as part of the gating. That would catch regressions in
the testing phase which is indeed the purpose of Bodhi. In contrast, Koschei sees the
error after the broken package becomes stable. That's too late.
And I also heard an opinion that %check phases could be completely skipped at build time
and replaced with the packaged tests. That would significantly improve bootstrapping
Fedora by simplifying a dependency graph for building. The tests would be performed
asynchronously later.
``
To reply, visit the link below
https://src.fedoraproject.org/rpms/perl-Sys-Virt/pull-request/1