cmake projects build twice

[aleixpol]

As seen in the project below, we build the project whole on the make step and on the make install step every single object gets rebuilt.

• locally: d56457de-build.20251008-120541.log
• on the CI: https://gitlab.com/freedesktop-sdk/flatpak-kde-runtime/-/jobs/11643361049/artifacts/browse/cache/buildstream/

It could make sense to skip the "build" step altogether as install will do it all anyway.

[abderrahim]

I tracked down a similar issue, and I think it's a bug in buildbox-fuse. You can try to re-run your build with BUILDBOX_STAGER=copy-or-link and see if you can reproduce the issue (make sure to delete the artifact and so it is actually rebuilt). Assuming your file system supports reflinks (btrfs or xfs), performance should be comparable.

I'll open an issue with more details.

As a workaround in freedesktop-sdk we changed from ninja install to cmake --install https://gitlab.com/freedesktop-sdk/freedesktop-sdk/-/blob/master/include/_private/cmake-conf.yml?ref_type=heads#L8

[aleixpol]

Can confirm that BUILDBOX_STAGER=copy-or-link fixes the issue.

It's already doing cmake --install, the problem as far as I understand it is that we are still doing the cmake build run too where it says + sh -c -e $'cmake --build _builddir -- ${JOBS}\n'.

[abderrahim]

It's already doing cmake --install, the problem as far as I understand it is that we are still doing the cmake build run too where it says + sh -c -e $'cmake --build _builddir -- ${JOBS}\n'.

Your log contains env DESTDIR="/buildstream-install" cmake --build _builddir --target install, which is not the same as what I linked.

This one is still going through ninja, and triggers the buildbox-fuse bug I was referring to.

[juergbi]

https://gitlab.com/BuildGrid/buildbox/buildbox/-/merge_requests/1065 will fix this issue.

[juergbi]

The upstream MR is now merged. The issue should be resolved when using BuildBox 1.3.42+.

Do we want to close this or do we still want to tweak cmake commands and keep this issue open?

[aleixpol]

Do we want to close this or do we still want to tweak cmake commands and keep this issue open?

Trusting it's indeed fixed, I'll close this issue.

I'll still leave #93 open as I think it still has value.