[Buildroot] [PATCH buildroot-test 1/1] scripts/autobuild-run: add support for --branch option

[James Hilliard]

On Thu, Feb 9, 2023 at 1:44 PM Arnout Vandecappelle <arnout at mind.be> wrote:
>
>
>
> On 09/02/2023 21:32, James Hilliard wrote:
> > On Thu, Feb 9, 2023 at 1:19 PM Arnout Vandecappelle <arnout at mind.be> wrote:
> >>
> >>
> >>
> >> On 08/02/2023 17:46, James Hilliard wrote:
> >>> On Wed, Feb 8, 2023 at 8:15 AM Thomas Petazzoni
> >>> <thomas.petazzoni at bootlin.com> wrote:
> >>>>
> >>>> On Tue,  7 Jun 2022 17:12:11 -0600
> >>>> James Hilliard <james.hilliard1 at gmail.com> wrote:
> >>>>
> >>>>> In some cases it only makes sense to run tests against a specific
> >>>>> branch, such as when testing new genrandconfig flags.
> >>>>>
> >>>>> Add a config option to override the branch selection.
> >>>>>
> >>>>> Signed-off-by: James Hilliard <james.hilliard1 at gmail.com>
> >>>>> ---
> >>>>>    scripts/autobuild-run | 11 ++++++++++-
> >>>>>    1 file changed, 10 insertions(+), 1 deletion(-)
> >>>>
> >>>> We looked at this with Arnout, and we don't really see what this would
> >>>> be needed, and actually it could be annoying to have people test
> >>>> "random" branches. If you need extensions to genrandconfig, then the
> >>>> autobuild-run script should be backward compatible with older versions
> >>>> of genrandconfig to be able to support older branches for a certain
> >>>> while.
> >>>
> >>> There are a number of other reasons this is useful, for example some
> >>> build hosts(for example my asahi-linux build host) have compatibility
> >>> issues in older branches so it doesn't make sense to run autobuilders
> >>
> >>    If there are compatibility issues, the fixes should be backported to the
> >> stable branches...
> >
> > Some of the changes would be rather difficult to backport, for example
> > the 2022.02.x branch doesn't have support for --no-toolchains-csv which
> > is effectively required for supporting non-x86_64 autobuilders as we
> > obviously can't use prebuilt toolchains that only run on x86_64 hosts on
> > an aarch64 host.
> >
> > There are many fixes related to various  --no-toolchains-csv build failures
> > that would also need to be backported as well if even feasible.
>
>   Even so, I'm much more inclined to rewrite the autobuild-run logic to
> dynamically test which options genrandconfig can take and adapt accordingly.

The genrandconfig issue is only one of a number of issues that I've
seen come up,
there's been a rather long tail of lingering aarch64 compatibility
issues that sometimes
require major package updates. For example some systemd fixes for aarch64 build
hosts aren't really feasible to backport due to requiring major refactoring.

What we probably eventually want is the ability to identify issues as build host
environment specific by rerunning failed builds on different hosts
automatically,
but that would probably require some infrastructure changes.

> That way you don't have to manually maintain those branches and update them on
> each autobuilder instance.

I'm pretty much just using this to restrict my autobuilders to master
branch when there
are known compatibility issues with older branches for a particular
host configuration.

>
>   Regards,
>   Arnout
>
> >
> >>
> >>    Regards,
> >>    Arnout
> >>
> >>> on anything other than say master.
> >>>
> >>> To prevent testing of unwanted branches I could modify this to restrict
> >>> the branch override to only allow selecting the branches listed in:
> >>> http://autobuild.buildroot.org/branches
> >>>
> >>>>
> >>>> Thanks!
> >>>>
> >>>> Thomas
> >>>> --
> >>>> Thomas Petazzoni, CTO, Bootlin
> >>>> Embedded Linux and Kernel engineering
> >>>> https://bootlin.com