On Wed, Nov 19, 2025, 9:15 AM Thomas Huth <[email protected]> wrote:

> On 19/11/2025 09.05, Paolo Bonzini wrote:
> >
> >
> > Il mar 18 nov 2025, 22:50 John Snow <[email protected]
> > <mailto:[email protected]>> ha scritto:
> >
> >
> >
> >     On Mon, Nov 17, 2025, 3:47 PM Paolo Bonzini <[email protected]
> >     <mailto:[email protected]>> wrote:
> >
> >
> >
> >         Il lun 17 nov 2025, 20:07 Daniel P. Berrangé <
> [email protected]
> >         <mailto:[email protected]>> ha scritto:
> >
> >             On Mon, Nov 17, 2025 at 01:51:25PM -0500, John Snow wrote:
> >              > In order to convert the existing Makefile target from a
> manual
> >              > invocation of mkvenv to one that uses the meson dependency
> >             system, we
> >              > need to not suppress ninja here.
> >              >
> >              > I'm not sure if this creates problems I am not aware of;
> but
> >             invoking
> >              > ninja here is no longer spurious but will become
> necessary.
> >
> >             Yes, this will likely create problems. From the commit
> message that
> >             introduced NINJA=":"...
> >
> >               Avoid it with the same
> >                  trick that we are using in buildtest-template.yml
> already
> >             by disabling
> >                  the up-to-date check via NINJA=":".
> >
> >
> >         Move the check-venv call to buildtest-template.yml, right after
> >         configure? It's cheap enough.
> >
> >         Paolo
> >
> >
> >     What's the root issue here? That there's enough of a time delay
> between
> >     the actual configure and the test running that it re-runs configure?
> >
> >
> > That the git repo is cloned again and is therefore newer than the
> artifacts.
> > That triggers a full rebuild.
>
> Right. At least for the other "functional" and "check-system" jobs in our
> CI, this was a big issue, calling "make check-functional/qtest/iotest"
> there
> triggered a rebuild of all binaries, increasing our CI runtime
> significantly.
>
> For the "crash test" jobs, it was mainly about avoiding to rerun the
> "configure" step for cosmetical reasons each time the job runs. The
> rebuild
> of the qemu binaries is triggered by "make check-iotests/qtests/function"
> in
> the other jobs, and not by "make check-venv" alone, so it's not as bad as
> in
> the other jobs here:
> Without your patch, the jobs take about 16 minutes (crash-test-debian) and
> 10 minutes (crash-test-fedora):
>
>   https://gitlab.com/thuth/qemu/-/jobs/12138792667
>   https://gitlab.com/thuth/qemu/-/jobs/12138792674
>
> With your patch applied:
>
>   https://gitlab.com/thuth/qemu/-/jobs/12139438583
>   https://gitlab.com/thuth/qemu/-/jobs/12139438599
>
> the jobs take approx. the same amount of time, so I think it's OK here to
> drop the NINJA=":" again.
>
> But I think we should give Paolo's idea a try and move the "make
> check-venv"
> to buildtest-template.yml instead. In fact, that might even help to
> slightly
> decrease the total runtime of the jobs since all the other "check" jobs
> don't have to do this again (for example the check-functional jobs are
> running it, too, see:
> https://gitlab.com/qemu-project/qemu/-/jobs/12139584312#L1051).
>
>   Thomas
>

Ah, I see now, thanks! I can think of some remedies for this. No problem.

Reply via email to