On 9/28/26 12:18 PM, Michael J Gruber wrote:
Panu Matilainen via devel venit, vidit, dixit 2026-09-28 10:08:45:
On 9/22/26 6:32 PM, Neal Gompa wrote:
On Tue, Sep 22, 2026 at 11:08 AM Pavel Raiskup via buildsys
<[email protected]> wrote:
Hi everyone,
We are planning to change Mock's default behavior when building
packages. Specifically, we intend to unconditionally pass --noclean to
rpmbuild during builds. This means Mock will no longer execute the
%clean section of spec files (or the implicit rmbuild cleanup phase).
So far, Mock has been using --noclean only when
`cleanup_on_success=False` (opt-in behavior).
Modern RPM packaging guidelines consider the %clean section obsolete.
Mock already manages the chroot and build root cleanups at the top
level before and after builds. Running %clean inside the chroot seems
redundant, and is hard to cover correctly (especially with the new
"delayed %check" feature).
More context and the background discussion can be found in PR #1809:
https://github.com/rpm-software-management/mock/pull/1809
Request for Feedback:
Before we make this change default in an upcoming release, we would
like to hear from the community:
Do you have any legitimate use cases where executing the %clean
section or rmbuild phase inside Mock is still strictly required?
Are you aware of any workflows or tooling built on top of Mock that
would break if Mock always passes --noclean?
If you have any objections, edge cases, or concerns, or process
opinions - please let us know.
The %clean section is obsolete because we already define a default
%clean stage in RPM itself since rpm 4.8 in 2009:
https://github.com/rpm-software-management/rpm/commit/3fc58248d23d6f720942e5cbf4f92db246a802f0
And modern default %clean handles all the weird multi-buildroot stuff
that is now supported since RPM 4.20. Using --noclean turns all that
stuff off, and can result in broken build results.
Huh?
--noclean only really affects finished builds because in the other
modes, clean steps are not to be executed anyhow.
Please be more specific if you have a concrete example where --noclean
(inside mock specifically) causes problems. --noclean like any other
option of rpmbuild, and if it breaks something then that sounds like a
bug to me.
Maybe there's some confusion between the mock option and the rpmbuild
option here. Also, rpmbuild's man page seems to suggest that `--noclean`
is not the negation of `--clean`.
If the rpmbuild options are really about executing `%clean` or not after
the build, then they affect whatever is configured to run there - for
example, moving something from the build tree to somewhere else that has
been saved during a multi-stage build, or running whatever over the
created builds. This may or may not be a "creative" use of %clean, but
apparently necessary for multilib.
What on earth? :D
I get a feeling we mean some entirely different things with multilib
here. Can you point me to an example package that does this mysterious
(to me) multi-stage thing?
- Panu -
Michael
--
_______________________________________________
devel mailing list -- [email protected]
To unsubscribe send an email to [email protected]
Fedora Code of Conduct:
https://docs.fedoraproject.org/en-US/project/code-of-conduct/
List Guidelines: https://fedoraproject.org/wiki/Mailing_list_guidelines
List Archives:
https://lists.fedoraproject.org/archives/list/[email protected]
Do not reply to spam, report it:
https://forge.fedoraproject.org/infra/tickets/issues/new
