Re: [PATCH] gitlab-ci.yml: Use the whole tree as artifacts to speed up the CI

[Thomas Huth]

On 22/01/2021 11.18, Daniel P. Berrangé wrote:
> On Fri, Jan 22, 2021 at 11:07:22AM +0100, Thomas Huth wrote:
> > Currently, our check-system-* jobs are recompiling the whole sources
> > again. This happens due to the fact that the jobs are checking out
> > the whole source tree and required submodules again, and only try
> > to use the "build" directory with the binaries and object files
> > as an artifact from the previous stage - which simply does not work
> > anymore (with the current version of meson). Due to some changed
> > time stamps, meson is always trying to rebuild the whole tree.
>
> This used to work in the past didn't it ? Did something change in
> meson to break this, or have we just not noticed before.

As mentioned by Philippe already, there seems to be a problem in current 
meson which triggers a rebuild of the whole tree in some cases. See this 
thread here:

https://lists.gnu.org/archive/html/qemu-devel/2021-01/msg05409.html

> > So instead of trying to marry a freshly checked out source tree
> > with the pre-built binaries in these jobs, let's simply pass the
> > whole source including the submodules and the build tree as artifact
> > to the test jobs. That way timestamps get preserved and there is
> > no rebuild of the sources anymore. This saves ca. 15 - 20 minutes
> > of precious CI cycles in each run.
>
> I'm a little worried we might end up hitting the artifact size
> limit which is supposedly 1GB on gitlab.com.  Im guessing this
> must be measuring the compressed size though, as a src checkout
> with build dir  and .git dir is already way over 1GB.

I've added a "du -sh" in one of the jobs and the size of the directory was 
indeed 2 GB already. But I did not see any problems due to that size. So 
either the limit must be bigger, or it's about the compressed size. Either 
way, it's working at least right now.

We could maybe also simply exclude the .git directory of the source 
checkout, I think that's the biggest chunk there and likely not needed for 
the tests.

> For to ask, could we address it by using  'meson test
> --no-rebuild' perhaps ?

Interesting idea! I'll give it a try, and if it works, I'll send a patch for 
that instead.

 Thomas