nixpkgs-review pr 184041 failed on the jacinda package, but i tested on master and it fails for the same reason. seems like this should be marked as broken on (at least) aarch64-darwin. would it be worth including that in this PR, or not since it's unrelated?

Result of nixpkgs-review pr 184041 run on aarch64-darwin 1

nixpkgs-review pr 184041 failed on the jacinda package, but i tested on master and it fails for the same reason. seems like this should be marked as broken on (at least) aarch64-darwin. would it be worth including that in this PR, or not since it's unrelated?

Feel free to add a conditional markBroken for regex-rure (which is actually failing) in this PR. Be sure to link the actual issue in a comment as well: haskell/c2hs#33 (comment)

@sternenseemann Any recommendation for a cleaner method of doing this?

{
  # ...
  regex-rure = doDistribute ((if isDarwin && isAarch64 then x: x else markUnbroken) super.regex-rure);
  # ...
}

Seems to work fine, just looks a little confusing imo

Any recommendation for a cleaner method of doing this?

There's an override section in configuration-darwin.nix that only gets applied on aarch64-darwin (towards the bottom of the file).

This patch didn't seem to work correctly, trying to build .#jacinda failed with the original error instead of reporting regex-rure broken. Is the GHC configuration overriding the darwin configuration?

Don't remember the ordering precisely (you can check pkgs/development/haskell-modules/default.nix) in any case, you don't have to have any override in the regular configuration, if your patch only touches configuration-darwin.nix, it should wrok.

I can't work out exactly what's going on, but it seems like ...

• regex-rure is generally marked as broken in hackage-packages.nix
• configuration-darwin.nix "re"marks it as broken, which is ignored
• configuration-9.2.x.nix marks it as unbroken, which is the final accepted setting

I can update the PR with this commit so someone can maybe take a look and see if I am missing something. Otherwise I could

• leave the PR as-is, which correctly marks the package broken on aarch64-darwin
• drop the broken-marking entirely, and leave that for another PR

edit: i realize i am somewhat assuming that the package will build properly on linux, but is that the case? Is gcc used for C2hs on linux? if not, maybe i could remove the override in configuration-9.2.x.nix to resolve this.

@sternenseemann sorry for the ping, i've pushed a version that i can get to locally report regex-rure as broken, i was unable to get something working via the configuration-9.2.x.nix file. if it's better, i can just drop the version specific hacking and leave the broken packages as-is for now. i think this should be ready for review now, would it be fair to ping anyone? otherwise i can hold tight until someone has the bandwidth.

regex-rure is generally marked as broken in hackage-packages.nix

Ah, right I remember now. The problem is that regex-rure only builds with GHC 9.2 and above, so you can only expect haskell.packages.ghc924.regex-rure to work and it is marked as broken by default (since that follows the state in the default package set). Leaving the override as is is okay (albeit a bit ugly), there doesn't seem to be a better solution due to the overlay chaining.

Can you rebase on master? I think the regex-rure hack can be dropped now. The patch will need to be applied to 9.2.4 and 9.2.5 now.

Porting that stuff to 9.4.* can be done in a separate step then.

This patch should be applied for 9.0 (pkgs.haskellPackages) as well, no? Because of #194367 (comment)

This patch should be applied for 9.0 (pkgs.haskellPackages) as well, no? Because of #194367 (comment)

haskellPackages is 9.2 now, 9.0 had and has a patch for this already which no longer applies for Cabal 3.8 shipped with GHC 9.2.

Looks like the patch that got added to all builds for 9.2.x (https://github.com/haskell/cabal/commit/6c796218c92f93c95e94d5ec2d077f6956f68e98.patch) updates some of the same files as this patch, so I have to hardcode it to address those changes. I would say this should be fine, except for the note that says

Can be removed if the Cabal library included with ghc backports the linked fix

But I guess that could just be addressed with a revert of the appropriate commit?

I'm also still seeing the regex-rure failure show up in local testing, for what it's worth.

alright, i think this should be ready for another pass of reviewing. i can kick off nixpkgs-review, but as per comments above, is there anything i can do to more thoroughly verify things are working as expected? i can try running ghcid or ormolu as I did in my initial post

This mass-subscribed a bunch of people, who won't get unsubscribed from retrieving notifications.

Please open a new PR.