haskell.compiler.ghc963Binary: drop

[wolfgangwalther]

This bootstraps mostly from ghc984Binary, at least for the major platforms. The bootstrap paths for darwin become (potentially much, with #440976 in mind) shorter. Unfortunately, the bootstrap paths for i686-linux become a bit longer, so it's a trade-off.

TLDR: I think this would be a major step towards consistent bootstrap paths from either GHC 9.4.8 or 9.8.4 for almost everything.

Things done

First, I tried bootstrapping everything from GHC 9.8.4 bindist:

984Binary	9.8.4	9.10.1	9.10.2	9.10.3	9.12.1	9.12.2
aarch64-darwin	✔️	❌	✔️	✔️	❔	❔
aarch64-linux	✔️	❌	✔️	✔️	❌	❌
aarch64-linux-musl	✔️	❌	✔️	✔️	❔	❔
i686-linux	❌	❌	❌	❌	❔	❔
x86_64-darwin	✔️	❌	✔️	✔️	❔	❔
x86_64-linux	✔️	❌	✔️	✔️	❔	❔
x86_64-linux-musl	✔️	❌	✔️	✔️	❔	❔

I only tested GHC 9.12 quickly for aarch64-linux, but that didn't look like it would go well, so backed out that change and kept its current bootstrap path. Since GHC HEAD does bootstrap with 9.8.4, we might look into this again eventually - it could be possible to patch this, so that we can consistently bootstrap from 9.8.4.

I decided to drop GHC 9.10.1, because I had that commit prepared already. Of course we have other options here, such as trying to patch it and/or change its bootstrap path. Personally, I'm not convinced that putting effort into this minor version is needed, though.

I also adjusted bootstrap paths for i686-linux:

• 9.8.4: ✔️

• 9.10.2: ✔️

• 9.10.3: ✔️

• Fits CONTRIBUTING.md, pkgs/README.md, maintainers/README.md and other READMEs.

---

Add a 👍 reaction to pull requests you find important.

[wolfgangwalther]

Resolved merge conflict.

@sternenseemann are you opposing any of the 4 commits explicitly? If yes, which one and why?

[yuyuyureka]

As it is, this PR looks like it will rot in a deadlock situation because of the thread above, and I know how bad this can feel for a PR author with no strong opinion on the topic.

I personally tend to agree with emilazy that the native 32-bit bootstrap is not really worth keeping around.
But I also am not a regular contributor to the Haskell parts of nixpkgs (other than @sternenseemann, it seems), so I respect his opinion on this.

@sternenseemann can you confirm once more: If seen as a source of unwanted complexity, is the pkgsi686Linux bootstrap around still important enough to you to warrant keeping it around?

[emilazy]

LGTM with two things I think should be fixed but aren’t strictly blocking. Since this has a concrete benefit of fixing Musl I think we should land it in the absence of reasons not to.

[sternenseemann]

I think we can retain i686 without adding conditionals even, I'm currently looking into it.

[wolfgangwalther]

I dropped the commit to remove i686 bootstrap.

I think we can retain i686 without adding conditionals even, I'm currently looking into it.

That would be great. Since I already have the conditionals and the PR fixes aarch64-linux musl + shortens the bootstrap path for the default GHC on darwin, I'd like to merge with the conditionals in place for now. If it works out what you try, we can put this on top!

@emilazy addressed your review, can you have another look?

[emilazy]

LGTM.

[wolfgangwalther]

Let's ship it then and proceed to #440976.