Re: [PATCH 1/3] kci-gitlab: Introducing GitLab-CI Pipeline for Kernel Testing

From: Guenter Roeck
Date: Mon Mar 04 2024 - 11:18:08 EST


On Mon, Mar 4, 2024 at 8:05 AM Maxime Ripard <mripard@xxxxxxxxxx> wrote:
>
> On Mon, Mar 04, 2024 at 07:46:34AM -0800, Guenter Roeck wrote:
> > On Mon, Mar 4, 2024 at 1:24 AM Maxime Ripard <mripard@xxxxxxxxxx> wrote:
> > [ ... ]
> > >
> > > If anything, it's more of a side-effect to the push for COMPILE_TEST
> > > than anything.
> > >
> >
> > If the drm subsystem maintainers don't want people to build it with
> > COMPILE_TEST while at the same time not limiting it to platforms where
> > it doesn't even build, I'd suggest making it dependent on
> > !COMPILE_TEST.
>
> I don't think we want anything. My point was that you can't have an
> option that is meant to explore for bad practices and expose drivers
> that don't go through the proper abstraction, and at the same time
> complain that things gets broken. It's the whole point of it.
>
Can we get back to the original problem, please ?

Build errors such as failed 32-bit builds are a nuisance for those
running build tests. It seems to me that an automated infrastructure
to prevent such build errors from making it into the kernel should be
desirable. You seem to disagree, and at least it looked like you
complained about the existence of COMPILE_TEST, or that people are
doing COMPILE_TEST builds. What is your suggested alternative ?
Disabling build tests on drm doesn't seem to be it, and it seems you
don't like the idea of a basic generic CI either, but what is it ?

> > The same applies to all other subsystems where maintainers don't want
> > build tests to run but also don't want to add restrictions such as
> > "64-bit only". After all, this was just one example.
>
> We have drivers for some 32 bits platforms.
>
I said "such as". Again, that was an example. In this case it would
obviously only apply to parts of drm which are not supported on 32-bit
systems (and, presumably, the parts of drm which are supposed to be
supported on 32-bit systems should build on those).

Thanks,
Guenter