Backport flake.nix changes to 1.8 #3060

jaschutte · 2025-10-24T10:52:21Z

Cherry picked the commits from the main branch regarding the flakes.nix to the 1.8 branch.

Pins down the GHC version down to the minor version. As discussed in a PR in clash-cores (clash-lang/clash-cores#44 (comment)). Each supported GHC version is now postfixed with a minor version number. However the overlays are not. The last digit of the version number gets stripped when Nix imports an overlay. This way we can have one set of GHC version numbers pin down the right GHC version *and* import the right overlay. I `nix build .#clash-ghc` with all three versions succesfully.

The overlay files themselves have changed a lot in master, from being renamed to having bugs in them be solved.

In specific; a50b496

jaschutte · 2025-10-31T14:31:02Z

@DigitalBrains1 would it be possible to re-run the failing task? I think it's a fluke. Aside from that I think this provides a solid base for the mergify PRs to not fail.

martijnbastiaan · 2025-10-31T16:11:39Z

I've pressed the button, but I think for GHA you should have been able to do it yourself too. If that's not the case we'll have to look into how those permissions work..

kleinreact · 2025-10-31T16:23:01Z

 clash-ghc             > C:\Users\RUNNER~1\AppData\Local\Temp\ghc5C38.c: DeleteFile "\\\\?\\C:\\Users\\RUNNER~1\\AppData\\Local\\Temp\\ghc5C38.c": permission denied (Access is denied.)

Seems to be a reproducible one 🤔.

DigitalBrains1 · 2025-10-31T18:50:43Z

Very weird. Maybe there's a problem with left over files on a runner and we managed to hit the same runner. I don't know. The whole point of Stack is that it's the same config every time, and with the previous PR it passed, and this PR doesn't change one iota of the Stack setup. Let's try again. Third time's the charm?

[edit]
There we go. Third time's the charm.
[/edit]

DigitalBrains1

Ah, thanks! I have just a few points, I think we're almost there!

nix/overlay-ghc96.nix

DigitalBrains1 · 2025-11-01T14:00:08Z

nix/overlay-ghc96.nix

  # is basically abandonware it catches fire with brick 1.0+.
  brick = doJailbreak prev.brick_0_70_1;

+  vty = prev.callHackage "vty" "5.39" { };


Why is this

vty = prev.callHackage "vty" "5.39" { };

in ghc96 but

vty = doJailbreak (prev.callHackage "vty" "5.39" { });

in ghc98?

Ghc98 has a different set of packages (since the compiler is newer) and in this set of packages for ghc98 certain packages (deepseq) has the wrong version, however it still compiles fine. That is why the jailbreak is required in ghc98 and not in ghc62 (where deepseq has the right version).

DigitalBrains1 · 2025-11-01T14:00:57Z

nix/overlay-ghc96.nix


+  vty = prev.callHackage "vty" "5.39" { };
+
+  # Marked as broken in nixpkgs, since it needs on a newer hashable than the


Suggested change

# Marked as broken in nixpkgs, since it needs on a newer hashable than the

# Marked as broken in nixpkgs, since it needs a newer hashable than the

What does it mean when you say

Needs a newer x than the .cabal file currently uploaded to hackage.

That the package needs a newer version of x than is currently uploaded to Hackage? So it needs an unreleased version? (Usually, we view the latest that is on Hackage as the latest release of a Haskell package.)

While the .cabal file specifies the version number for a specific version of a package, it's the uploaded package itself that is at that version. So you can just shorten it to

Needs a newer x than the latest version on Hackage.

But it feels like you mean something else entirely, as it's rather surprising that a released version of a package strictly requires an unreleased version of another package, so I'm probably misinterpreting.

I didn't write these comments haha, I cherry picked them from the main branch. However I think the comments refer to the version of Hackage pinned by the flake. It could be that actual Hackage has a newer version however the pinned set of packages is older, so we jailbreak it.

Oh, sorry, I should have checked where it came from before raising it here.

Well, you know what, if they just come from master let's use them as they are.

nix/overlay-ghc98.nix

nix/overlay-ghc910.nix

nix/overlay.nix

jaschutte · 2025-11-03T20:30:26Z

I am noticing that the build for ghc964 and ghc982 are failing due to clash-cores. This is the same error which mandated the use of the -package ... flags in clash-cores, but they seemingly stopped working..?

The exact error:

 Running phase: installPhase
> Installing library in /nix/store/nk5x59gj85dkpfc4pafz7k9n2287pl5x-clash-cores-1.8.3/lib/ghc-9.6.4/lib/x86_64-linux-ghc-9.6.4/clash-cores-1.8.3-51cL0hdQstoGyd2JlzQBpU
> Error: Setup: '/nix/store/s461n4kgh0khrfpb6jqc8mf1hlqg20gn-ghc-9.6.4/bin/ghc'
> exited with an error:
> <command line>: cannot satisfy -package ghc-typelits-extra

This is weird since listing the buildInputs, ghc-typelits-extra-0.4.7 is listed (and so are the other ghc typelist extensions).

EDIT: I will apply the same patches I did at clash-cores
clash-lang/clash-cores@d9d1679#diff-206b9ce276ab5971a2489d75eb1b12999d4bf3843b7988cbe8d687cfde61dea0

And that seemed to have worked!

DigitalBrains1

And that seemed to have worked!

Nice!

Shall we squash-commit this? There's two problems with keeping the individual commits. First off, they don't individually pass CI, and I prefer every commit to pass CI. Secondly, the first commit message is cherry-picked from 762f7f0 and it says "I nix build .#clash-ghc with all three versions succesfully." which is confusing since this is not actually possible here :-).

jaschutte · 2025-11-04T13:10:56Z

I'm in favor of squash merging!

DigitalBrains1 · 2025-11-04T13:28:50Z

Okay, go ahead!

DigitalBrains1 · 2025-11-04T15:22:35Z

We decided that we want a Changelog entry, James, let's decide on a phrasing. I don't particularly care whether the Changelog entry is actually a part of a merged PR or not; writing the Changelog for a release is still a manual process anyway, so I can just add it.

First thought:

Changed:

Nix flake updated to make it more suitable for use in downstream projects

jaschutte force-pushed the flake-changes-1.8 branch from 7708534 to 9ae791c Compare October 24, 2025 10:56

Port overlays from master

a4ff702

The overlay files themselves have changed a lot in master, from being renamed to having bugs in them be solved.

jaschutte force-pushed the flake-changes-1.8 branch from e65abfd to a4ff702 Compare October 31, 2025 13:25

Bring commit up to date with main

b4b054e

In specific; a50b496

DigitalBrains1 requested changes Nov 1, 2025

View reviewed changes

Fix developer shells for ghc964 and ghc982

1cea021

DigitalBrains1 approved these changes Nov 4, 2025

View reviewed changes

jaschutte merged commit 52a5472 into 1.8 Nov 5, 2025
16 checks passed

jaschutte deleted the flake-changes-1.8 branch November 5, 2025 11:07

jaschutte mentioned this pull request Nov 5, 2025

Pin Nix GHC minor version (copy #2984) #3059

Closed

2 tasks


		vty = prev.callHackage "vty" "5.39" { };

		# Marked as broken in nixpkgs, since it needs on a newer hashable than the

Backport flake.nix changes to 1.8 #3060

Backport flake.nix changes to 1.8 #3060

Uh oh!

Conversation

jaschutte commented Oct 24, 2025

Uh oh!

jaschutte commented Oct 31, 2025

Uh oh!

martijnbastiaan commented Oct 31, 2025

Uh oh!

kleinreact commented Oct 31, 2025

Uh oh!

DigitalBrains1 commented Oct 31, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

DigitalBrains1 left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

DigitalBrains1 Nov 1, 2025

Choose a reason for hiding this comment

Uh oh!

jaschutte Nov 3, 2025

Choose a reason for hiding this comment

Uh oh!

DigitalBrains1 Nov 4, 2025

Choose a reason for hiding this comment

Uh oh!

DigitalBrains1 Nov 1, 2025

Choose a reason for hiding this comment

Uh oh!

DigitalBrains1 Nov 1, 2025

Choose a reason for hiding this comment

Uh oh!

jaschutte Nov 3, 2025

Choose a reason for hiding this comment

Uh oh!

DigitalBrains1 Nov 4, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

jaschutte commented Nov 3, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

DigitalBrains1 left a comment

Choose a reason for hiding this comment

Uh oh!

jaschutte commented Nov 4, 2025

Uh oh!

DigitalBrains1 commented Nov 4, 2025

Uh oh!

DigitalBrains1 commented Nov 4, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

5 participants

DigitalBrains1 commented Oct 31, 2025 •

edited

Loading

jaschutte commented Nov 3, 2025 •

edited

Loading