I noticed this when putting up a PR to wolfi that changed lots of packages.
It appeared to me that the solver did not consider runtime dependencies, but c-i was doing test immediately after build.
So for a package P that depended on A for build and A and C for runtime, there was no guarantee that C would be built before P.
It is possible that this is acceptable, and the build/test should simply be changed to do "build all the things and then test all the things". I don't have a good argument against that, but the current process in github actions was 'make package/' and then 'make test/'.
I noticed this when putting up a PR to wolfi that changed lots of packages.
It appeared to me that the solver did not consider runtime dependencies, but c-i was doing test immediately after build. So for a package P that depended on A for build and A and C for runtime, there was no guarantee that C would be built before P.
It is possible that this is acceptable, and the build/test should simply be changed to do "build all the things and then test all the things". I don't have a good argument against that, but the current process in github actions was 'make package/' and then 'make test/'.