the in-tree regressiontest download can get out of sync with code
If the code gets updated, either through pulling new changes or especially by checking out a new branch, the wrong regressiontest tarball can end up being used to verify mismatching code.
If implementing invalidation is easy enough it might be worth doing it despite the plans for the regressiontests to be merged into the repo.
#1 Updated by Szilárd Páll 3 months ago
I thought some URL inspection trickery could be used to see if the on-server redirect points to a URL with a new filename than the one that's been download before. However, the source of the issue is that the tarball does not have a version in the filename, so one can't know what the version is without re-downloading. If we could have a simple redirect on the server side, this would be rather easy.
#2 Updated by Szilárd Páll 3 months ago
BTW, actually the one thing we should fix is that we make
make clean consistent. We should either have it remove the regressiontest downloads too (like the fftw tarball) or add a new clean target, e.g. a
make clean-all that removes the aforementioned downloads.