Make it build with ghc 9.12

[erikd]

Description

Make it build with ghc 9.12
Also note that:

• New code should be properly tested (even if it does not add new features).
• The fix for a regression should include a test that reproduces said regression.

[jasagredo]

This PR looks good, but I wonder if we want to merge it with those allow-newer.

Also I'd rather wait until the 10.3 branch is rebased onto latest main, before merging this one

[nfrisby]

All the changes look mild (🙌) except that ref=for-stackackage in the haskell.nix declaration.

[nfrisby]

I'm not familiar with the broader consequences of this change. The haskell.nix repo says:

The for-stackage branch also contains all cabal expressions from hackage.org as nix expressions.

This branch is used when working with stack projects in haskell.nix.

Could you summarize the potential reasons to not use it? And what is requiring its use here? The Consensus build process (eg CI) generally doesn't involve stack.

[erikd]

That is some Nix magic foo. @angerman might be able to explain.

[erikd]

@angerman responded in Slack and can be posted here:

Newer haskell.nix doesn't need hackage.nix to a large degree anymore, as .cabal -> .nix expressions are derived in the fly. As such hackage.nix (in it's old form doesn't exist anymore); we still keep it around to support stack and stackage. We are stuck with an old haskell.nix version until we can upgrade past ghc 8.10. Hence this rather annoying mismatch between hacakge.nix and haskell.nix, leading to older haskell.nix versions now needing to use the for-stackage branch (which is still periodically updated with new .nix expressions for the .cabal files in the hackage index).

[nfrisby]

Maybe a comment in the nix file, such as # workaround haskell.nix and hackage.nix being out of sync until we drop support for GHC 8.10 or some such

[erikd]

This PR looks good, but I wonder if we want to merge it with those allow-newer.

The allow-newer is guarded by if impl (ghc >= 9.12).

Also I'd rather wait until the 10.3 branch is rebased onto latest main, before merging this one

That is fine.

[lehins]

All the changes look mild (🙌) except that ref=for-stackackage in the haskell.nix declaration.

stackackage 🤣