Wouter Deconinck

The fix seems fine (we are indeed using `fail-on-missing` by default through the spack package). It is still not entirely clear in advance which features require connectivity, and which don't...

Since you had to patch it in nix, you'll also appreciate the fragility of such an approach. In fact, it will already break with #15467... (`substituteInPlace CMakeLists.txt --replace 'if(NO_CONNECTION)' 'if(FALSE)'`...

Qt6 (i.e. `qt-base`) doesn't require `qmake` anymore. In spack, `qt-base` uses CMake and doesn't install `qmake`.

> `qt-base` does indeed install `qmake` Right, I see now. Do you want to modify `qt-base` as part of this PR or should I file a separate PR for that...

> > qt-base does not build > Maybe @wdconinc @sethrj can help with this one. Probably best to file a separate issue on that with more info (system etc). I...

@oliviermattelaer Distributions (not only spack) can only package *usable* versions that are provided under a stable link. Can mg5nlo provide such stable links for non-lts versions?

> Would you be interested in that? No, as it would be a url that would again change its hash unpredictably.

@oliviermattelaer What would help the most here, for spack, is if we could keep the last numbered version instead of removing them when lts is created. So, for example, https://launchpad.net/mg5amcnlo/2.0/2.9.x/+download/MG5_aMC_v2.9.19.tar.gz...

Just got bit by this again. If this is indeed due to a bad index on the buildcache, is there a way to make package removes from the buildcache maintain...

They can be used as is, but with a performance penalty due to emulation.