llvmPackages_14.openmp: unbreak

[dtzWill]

Description of changes

Fixup old patches causing OpenMP to be marked broken,
one of which can be removed in favor of pointing the build
at the required tools directly.

Mostly get tests working, but few failures I haven't tracked down
caused me to ultimately disable them for now.

Help testing or pointing to something that should work on Linux with
this is appreciated :).

Things done

• Built on platform(s)
  • x86_64-linux
  • aarch64-linux
  • x86_64-darwin
  • aarch64-darwin
• For non-Linux: Is sandbox = true set in nix.conf? (See Nix manual)
• Tested, as applicable:
  • NixOS test(s) (look inside nixos/tests)
  • and/or package tests
  • or, for functions and "core" functionality, tests in lib/tests or pkgs/test
  • made sure NixOS tests are linked to the relevant packages
• Tested compilation of all packages that depend on this change using nix-shell -p nixpkgs-review --run "nixpkgs-review rev HEAD". Note: all changes have to be committed, also see nixpkgs-review usage
• Tested basic functionality of all binary files (usually in ./result/bin/)
• 22.05 Release Notes (or backporting 21.11 Release notes)
  • (Package updates) Added a release notes entry if the change is major or breaking
  • (Module updates) Added a release notes entry if the change is significant
  • (Module addition) Added a release notes entry if adding a new NixOS module
  • (Release notes changes) Ran nixos/doc/manual/md-to-db.sh to update generated release notes
• Fits CONTRIBUTING.md.

[dtzWill]

Looks like others need fixing too, see #162423 . Especially libunwind and libcxxabi. libunwind is trickier re:patch.
@Ericson2314 IIRC you upstreamed some patches to LLVM re:GNUInstallDirs, what's the status there? And don't suppose you have patches for these already or would be willing to take a look? Thanks!

[dtzWill]

(force-push'ing after rebasing onto master, and -rc updates before that)

[Ericson2314]

@dtzWill #154465 updated the patches somewhat, a tiny bit more was perhaps upstreamed since. I am not sure at what point the branch-offs occured upstream.

[Ericson2314]

github.com/Ericson2314/llvm-project has the commits I made by patches from, but I would rebase them so it is unclear whether we have enough to stuff for prior releases from there :(.

[Ericson2314]

@dtzWill Did you just remove parts of the patch that already were applied, or what?

[dtzWill]

@dtzWill Did you just remove parts of the patch that already were applied, or what?

That's what it ended up being, yes, I think. Not how I went about it but yes, applied and seemed safe to drop bits that were already applied (since seemed you were upstreaming things). The rest still seem useful, and seem to be needed (and doing what they seem intended to).

Bummer about the patches rebase'd and maybe lost from around the branch-off. These things happen, but ugh sorry. How do you want to handle this, then? (don't suppose you have a super high git reflog limit and they're still sticking around in there? hopeful face)

What we have here (openmp) doesn't seem super broken (bits all end up in expected places, compared to openmp13 -- includes are almost entirely the same, etc.), but I haven't checked the produced cmake bits all work (esp across multiple outputs). Haven't sorted what a good test might be, since our in-tree users are Darwin or a clangStdenv (so can't just point nixpkgs-review at it or something :)).

The other bits (the other subprojects) I/we might need (or really appreciate at the least) some help with. Can you take a look? Or is there a way to check things work as expected?

---

BTW I've done a hopefully-not-too-rough attempt at packaging up MLIR and flang over here: #163878 . I noticed your github has a gnu installdirs for MLIR (5ad96992ef909e6b99d8c693a285f06c9fc645fe), is any of that needed? To fix libraries being installed in the wrong location, I addressed it same as in another patch (but I don't see that in your commit). Well anyway, this can be discussed over in that PR if you have time/are interested 😄.

[siraben]

Result of nixpkgs-review pr 163868 run on aarch64-darwin 1

1 package built:

• llvmPackages_14.openmp

[Ericson2314]

@dtzWill is the issue that the GNU install dirs LLVM patches only contain too much stuff, or do they sometimes not contain enough stuff? If just tthe former I might have reflog, but honestly I would just start with what's in Nixpkgs as more of a known-good state.

[Ericson2314]

Also note I did save the original backports in split-prefix-*. I should probably put 13's in there for safe keeping, and 14's once it is fixed up.