Remove cudainfo and amd-uprof from crossSet #210
Reference in New Issue
Block a user
Delete Branch "abonerib/jungle:cross-filter"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
These two packages result in evaluation errors when building
crossSet.riscv64due to their dependencies meta.platforms:@@ -111,0 +112,4 @@genCross = platform:# filter out packages by meta.platformsfilterAttrs (_: meta.availableOn final.pkgsCross.${platform}.stdenv.hostPlatform)final.pkgsCross.${platform}.bsc.pkgsTopLevel;IIUC this is adding packages with
cross = false(or not set). Not sure if we want to keep both crossSet and crossList the same. In that case, using canCrossCompile may be handy.In any case, I would place genCross before crossSet so it is defined before referenced.
You are correct, I think
crossSetshould be all the packages that should theoretically be possible to cross-compile (opt-out) andcrossListshould the subset that we care about and are used in the CI so we don't introduce regressions (opt-in).From hydra, I can monitor
crossSetstatus and keep track of why some packages fail to cross compile and work on fixing them. I have a branch which fixes mpich, once that is fixed, only clangOmpss2, tagaspi, mcxx and paraver will remain broken.I would like to keep the simple idea that either cross-compilation for a package works (for the pinned commit) and is maintained with
cross=trueor is broken (even if its is not).I assume with hydra you will be testing newer commits, which will add substantial maintenance burden, and currently, many extra packages. Do we want to make the set of packages tested by hydra be larger than the ones where we have
cross=true? Wouldn't make more sense to use the same set for both?Yes, but, right now I only track select repos (and have excluded cross-compilation entirely for the rolling tests)
Currently, I use hydra to determine what should be put in
cross=trueand fix any low-hanging fruit.I think we can wait until after nixpkgs 25.11, and then decide what to do for each package.
That being said, hydra caches build failures, so there is little overhead in having a "big" crossSet since most of the time the derivations won't change.
Okay, then fine for me.
97e81b2f91to3beae1f0193beae1f019to7989779c8f