On Mon, Jan 25, 2021 at 4:37 PM Stephen Gallagher <sgallagh(a)redhat.com> wrote:
On Mon, Jan 25, 2021 at 10:31 AM Miro HronĨok <mhroncok(a)redhat.com> wrote:
>
> On 25. 01. 21 16:19, Stephen Gallagher wrote:
> > I'm fully in favor of this and I'd really like to see us add some
> > degree of CI gating to support it.
>
> Note that unfortunately CI gating happens too late. It has no capability to
> block commits that fail to build, because it only tests successful builds and
> because it only tests already pushed changes. CI on Pull Requests can solve
> this, but many packagers seem to be very much agianst the idea of sending PRs to
> packages they maintain themselves :(
>
There are still ways around this; we could disallow direct pushes to
release branches.
Yes, there are always going to be people who reject any new approach
we might take, but we should carefully consider whether "some/many
packagers are mildly annoyed" exceeds the potential gains inherent in
fewer broken builds and composes.
For those few of us who maintain hundreds-thousands of packages,
having everything go via a Pull Request is just not scalable as things
are right now.
If packaging tools can "hide the details" around this in a reliable
way so that packagers don't see what happens behind the scenes, then
we can talk.
But that would involve at least six new steps that would've to be
automated: 1) Creating a fork on src.fp.o (plus error handling around
already existing forks), 2) Cloning the fork instead of the main repo,
3) Automatically creating a PR after a git push, 4) Automatically
merging the PR if CI passes, 5) Deleting the fork, 6) Automatically
building the package in koji ...
Fabio