Update flake #3112

rowanG077 · 2025-12-23T22:09:41Z

This updates the nix flake to the newest nixos-unstable. Fully updates to the most recent GHC minor version as well as bump the typelit plugins.

clash-lib/clash-lib.cabal

.github/workflows/nix.yml

rowanG077 · 2025-12-24T09:23:31Z

I also ran into: #2823

in nixpkgs it's also patched out on >=9.6: https://github.com/NixOS/nixpkgs/blob/4cb741a91fd65971689b6272fd1fdc67355cd3e5/pkgs/development/compilers/ghc/ghc-9.6-llvm-use-new-pass-manager.patch

So I applied @christiaanb suggestion to remove it entirely.

jaschutte · 2025-12-24T14:19:41Z

Looks good overall! clash-ffi seems to not build on ghc984 with Nix though. I tried building it locally and it fails there too so it's not a CI bug.

I thought it maybe had to do with the removal of inherit (hfinal) clash-prelude in f6d9393bf297but adding that back Nix complains about having unsolicited arguments passed. Another item of note is that building clash-ffi for other versions doesn't give any errors, but it produces empty folders.

DigitalBrains1

I really thought we deliberately didn't use 9.10.3. Also please see #3112 (comment). Without adding it to CI, I really think we can't put it in our flake. And I suspect CI will fail.

[edit]
I'll run GitLab CI on it.
[/edit]

DigitalBrains1 · 2025-12-24T14:43:06Z

clash-ffi seems to not build on ghc984 with Nix though.

Perhaps it's the ffi-interface-tests test which doesn't work with plain Cabal either?

clash-compiler/.gitlab-ci.yml

Lines 43 to 46 in 2423678

    
           - GHC_VERSION: 9.8.4 
        
             # TODO: remove this when https://gitlab.haskell.org/ghc/ghc/-/merge_requests/12264#note_602406 
        
             # is fixed 
        
             SKIP_CLASH_FFI_EXAMPLE: "yes"

clash-compiler/.ci/gitlab/test.yml

Lines 157 to 160 in 2423678

    
           # TODO: remove this when https://gitlab.haskell.org/ghc/ghc/-/merge_requests/12264#note_602406 
        
           # is fixed 
        
           rules: 
        
             - if: $SKIP_CLASH_FFI_EXAMPLE != "yes"

DigitalBrains1 · 2025-12-24T15:05:58Z

Heh, silly me. If I would like to test 9.10.3 on GitLab CI, I'd first need to create a Docker image for that.

I'll try a local run.

@martijnbastiaan It wasn't an accident that we did not use 9.10.3, right? Something was up with it? I've forgotten what, but if we'd supported 9.10.3, we'd have done so to stay on Stackage.

[edit]
Hmmmm clash-testsuite locally runs to completion (no Symbiyosys or Modelsim tried). And the test suites for clash-prelude and clash-lib also succeed (but those are also part of GitHub Nix CI, so we knew that already). And yet, I think there was a problem.
[/edit]

[edit 2]
Ah maybe for Stackage we're just not on there due to #3007 ?
[/edit 2]

[edit 3]
ModelSim also passes on clash-testsuite
[/edit 3]

martijnbastiaan · 2025-12-25T08:49:09Z

I have no idea @DigitalBrains1, sorry. It wouldn't surprise me if it was deliberate though, every other GHC release seems to be broken in some way.

DigitalBrains1 · 2025-12-25T10:00:43Z

Maybe I'm just mixing stuff up with GHC 9.12, though.

GHC 9.12 is broken for us, and that is why we are not in Stackage nightly. But I'm not sure what the situation is for 9.10.

I think we can switch to 9.10.3, provided we add it to GitLab CI, because we want to know when we break something.

@christiaanb Do you agree? Or do you remember why we're still on 9.10.2?

martijnbastiaan · 2025-12-27T15:00:58Z

IMO just do it. If it didn't work, we should have left a note..

DigitalBrains1 · 2025-12-28T11:30:38Z

Right, I wasn't very explicit, but I also meant, let's upgrade to 9.10.3.

DigitalBrains1 · 2025-12-28T13:55:14Z

By the way, I'm going to do a Hackage revision of clash-prelude and clash-lib v1.8.4 to also allow doctest-parallel v0.4. I've asked our GitLab CI nicely to test this change for me, and if it is happy about it (I really expect it will be), I'll do that.

That way, we can get into Stack LTS 24 without any fussing about with disabling tests because the version is out-of-bounds.

Which brings me to this question: we're going to backport this PR to 1.8, right? Our own downstream projects use the flake from the stable Clash release, so it'd be good to get this change into an upcoming stable release. We definitely don't want the next release to shrink its bounds on doctest-parallel :-). If this PR can and should be backported, could you add the backport 1.8 label?

DigitalBrains1 · 2025-12-29T11:33:52Z

@rowanG077 , I've created a branch peter/nix-ghc-minor-update where I did all the things needed to bump to 9.10.3 in CI. I created and published new Docker images, which is the major part of it. I also noted that the script wasn't updated to drop old versions of GHC we no longer support in master, so I yanked them as well.

I suggest you include the commit in this PR.

Of course, the [skip ci] commit at the HEAD is not part of it; it's just to prevent GitLab CI from running costly CI for no purpose. It's especially costly because the changes to CI affect caching.

BTW, the backport to 1.8 of this PR will need some more work for the changes to CI. I'll do that as well (it's not much work if you've done it before); just don't expect a backport to work right away.

rowanG077 · 2025-12-29T11:37:40Z

@DigitalBrains1 Awesome! Will take a look at this later this week and finalize this!

DigitalBrains1

My requested changes have been taken care of, approved.

However, I think you will still need to do something about 9.8.4 on Nix. I suspect ffi-interface-tests is your culprit #3112 (comment).

rowanG077 · 2025-12-30T10:23:58Z

The reason the clash-ffi build succeeds on GHC 9.10.3 and not on GHC 9.8.4 seems to be that cabal errors on empty package definition with the cabal that is used with GHC 9.8.4 but no longer with GHC 9.10.3.

What I have done now is marked clash-ffi broken on the GHC version that it doesn't function on in nix as well and removed the builds from the matrix in CI.

@jaschutte To do this I had to change the order of the overlays. With ghcOverlay first there was no way to mark clash-ffi broken only on some specific GHC version. I don't think this has further impact for us but I'd like you to confirm that!

I think if @jaschutte is oke with this and CI passes it can be merged.

rowanG077 · 2025-12-30T12:11:06Z

Scratch that, It seems that clash-ffi has no problem running on 9.8.4 or 9.10.3 (fingers crossed for CI), so I have:

Removed the guards on buildable in the clash-ffi cabal file.
Transformed the test executable to a proper test-suite so nix picks it up and runs it. I couldn't dig up a reason it originally was a bare executable, do you know @kleinreact?
Removed the CI logic that skips clash-ffi on the apparently not broken versions.

There seems to be no reason this was not a test-suite to begin with. Making it a test-suite makes it automatically build and ran for nix builds of clash-ffi.

kleinreact · 2025-12-30T12:28:27Z

Transformed the test executable to a proper test-suite so nix picks it up and runs it. I couldn't dig up a reason it originally was a bare executable, do you know @kleinreact?

I don't remember the exact issue, but it was some problem with adding C sources as part of the test-suite. Happy to hear if the issue got fixed such that we can convert it to a proper test-suite now.

This reverts commit 1e6967a.

This reverts commit 973e91a.

It is just for `clash-ffi`, but we are about to disable that

Co-authored-by: Rowan Goemans <[email protected]>

christiaanb · 2026-01-02T16:28:07Z

clash-ffi/clash-ffi.cabal

  -- TODO: remove this once https://gitlab.haskell.org/ghc/ghc/-/merge_requests/12264#note_602406
  -- is fixed
-  if impl(ghc < 9.8.3) || impl(ghc == 9.10.1)
+  if impl(ghc < 9.8.3) || impl(ghc == 9.10.1) || impl(ghc > 9.10.2)


Why not:

if impl(ghc < 9.8.3) || impl(ghc >= 9.10.1)

Because GHC 9.10.2 has the issue.

Of course we could do

if impl(ghc < 9.8.3) || impl(ghc > 9.10.2)

and just ignore the fact that 9.10.1 would actually work.

Ah, perhaps we should reverse the logic and do:

-- GHC 9.8.4 and 9.10.2 have issues, see https://gitlab.haskell.org/ghc/ghc/-/merge_requests/12264#note_602406 if impl(ghc == 9.8.4) || impl(ghc == 9.10.2) buildable: False else buildable: True

You're thinking nobody will be trying 9.8.3 anymore? In that case, it would indeed match everything in practice.

What happened to 9.8.3? :-) Do you know why it was pulled from GHCup? I couldn't find anything at a really quick glance, and my dinner is cooling.

Ah, sorry, I missed that 9.8.3 also has the bug... so yeah, then it would be:

-- GHC 9.8.3, 9.8.4 and 9.10.2 have issues, see https://gitlab.haskell.org/ghc/ghc/-/merge_requests/12264#note_602406 if impl(ghc == 9.8.3) || impl(ghc == 9.8.4) || impl(ghc == 9.10.2) buildable: False else buildable: True

Perhaps we can just leave the logic as is, but update the comment to specify which GHCs are affected

Fine by me! What do we do with

-- TODO: remove this once https://gitlab.haskell.org/ghc/ghc/-/merge_requests/12264#note_602406 -- is fixed

I'm not sure what it suggests we do, but I think it's saying: once GHC 9.8 and GHC 9.10 both have received a minor version update that fixes the bug, remove this stanza. I wonder whether that would even be a good idea.

Also, after reading the discussion about 9.8.3 I linked to in my previous message, I doubt there will ever be another 9.8 release.

You wrote the comment; what did you intend?

Ah, I meant that when a fix was released, we should properly fix the bounds. The original version didn’t allow building on e.g. GHC 9.12

Okay I've updated the comments from

-- TODO: remove this once https://gitlab.haskell.org/ghc/ghc/-/merge_requests/12264#note_602406 -- is fixed

to

-- GHC 9.8.3, 9.8.4 and 9.10.2 have issues, see https://gitlab.haskell.org/ghc/ghc/-/merge_requests/12264#note_602406

I note that in .ci/build.sh the phrasing was a bit different:

clash-compiler/.ci/build.sh

Lines 42 to 44 in d0f65c4

# TODO: remove this and put it back into tests when

# https://gitlab.haskell.org/ghc/ghc/-/merge_requests/12264#note_602406

# is fixed

I think it'll be quite a while before we can actually put it back into tests; I suppose that will be once we drop 9.8 support from Clash. In the interest of clarity and maintainability, I've rewritten it a bit so it doesn't duplicate shell code and is easier to grok.

Are you satisfied with how it is now, @christiaanb ?

christiaanb · 2026-01-02T16:30:10Z

clash-ffi/clash-ffi.cabal

  -- TODO: remove this once https://gitlab.haskell.org/ghc/ghc/-/merge_requests/12264#note_602406
  -- is fixed
-  if impl(ghc < 9.8.3) || impl(ghc == 9.10.1)
+  if impl(ghc < 9.8.3) || impl(ghc == 9.10.1) || impl(ghc > 9.10.2)


Why not

if impl(ghc < 9.8.3) || impl(ghc >= 9.10.1)

jaschutte · 2026-01-07T11:32:05Z

@jaschutte To do this I had to change the order of the overlays. With ghcOverlay first there was no way to mark clash-ffi broken only on some specific GHC version. I don't think this has further impact for us but I'd like you to confirm that!

Nope! That should be good. I don't know what depends on clash-ffi and if marking that broken causes anything bad, but the overlay change should be fine!

Just to be sure I nix build clash-protocols with this version of clash-compiler, which seems to evaluate fine at least. It fails compilation due to wrong dependency versions but that's unrelated to this PR. Tldr; this is fine by me :)

rowanG077 requested a review from jaschutte December 23, 2025 22:09

rowanG077 commented Dec 23, 2025

View reviewed changes

clash-lib/clash-lib.cabal Show resolved Hide resolved

rowanG077 force-pushed the nix-ghc-minor-update branch 2 times, most recently from 1ba5d8c to 76353ee Compare December 23, 2025 22:19

kleinreact reviewed Dec 24, 2025

View reviewed changes

.github/workflows/nix.yml Outdated Show resolved Hide resolved

rowanG077 force-pushed the nix-ghc-minor-update branch 3 times, most recently from cf6bc92 to 1b9d874 Compare December 24, 2025 10:56

DigitalBrains1 requested changes Dec 24, 2025

View reviewed changes

rowanG077 added 8 commits December 29, 2025 15:02

Update flake

d0b50fa

Update nix haskell overlays

27703db

doctest <0.4 -> <0.5

2acf9da

GA: Update GHC matrix for nix jobs

4ff7798

flake: Use ghc 9.10.3

7351ac2

clash-ghc: Remove Opt_LlvmTBAA from unwanted

0539f07

Add changelog

db71253

Flake: Fix renamed packages

aadf718

rowanG077 force-pushed the nix-ghc-minor-update branch from 1b9d874 to da40b2e Compare December 29, 2025 14:03

DigitalBrains1 approved these changes Dec 29, 2025

View reviewed changes

rowanG077 and others added 2 commits December 30, 2025 10:12

flake: Fix lib.optional usage

e918199

Bump CI versions

c4f95e3

rowanG077 force-pushed the nix-ghc-minor-update branch from da40b2e to a103740 Compare December 30, 2025 10:20

rowanG077 force-pushed the nix-ghc-minor-update branch from 08e1ee0 to bf88deb Compare December 30, 2025 10:26

clash-ffi: Remove seemingly wrong buildable guards

1e6967a

rowanG077 force-pushed the nix-ghc-minor-update branch from bf88deb to feb0ff3 Compare December 30, 2025 12:05

rowanG077 added 3 commits December 30, 2025 13:17

clash-ffi: Make ffi_interface-tests a proper test-suite

147d59e

There seems to be no reason this was not a test-suite to begin with. Making it a test-suite makes it automatically build and ran for nix builds of clash-ffi.

nix: Make clash-ffi work including tests

e227a2d

ci: Remove conditional test run for clash-ffi

973e91a

rowanG077 force-pushed the nix-ghc-minor-update branch from 7774a32 to 973e91a Compare December 30, 2025 12:18

DigitalBrains1 and others added 6 commits December 30, 2025 14:35

Revert "clash-ffi: Remove seemingly wrong buildable guards"

7be94b3

This reverts commit 1e6967a.

Revert "ci: Remove conditional test run for clash-ffi"

11f645a

This reverts commit 973e91a.

GitLab CI: Enable clash-ffi on GHC 9.10.3

f6404fe

Unneeded clause in overlay-ghc98.nix

9292eb4

It is just for `clash-ffi`, but we are about to disable that

Make clash-ffi buildable on GHC 9.10

f99af1b

Nix: Mark clash-ffi broken on the GHC version it does not work on

2330bb4

Co-authored-by: Rowan Goemans <[email protected]>

DigitalBrains1 force-pushed the nix-ghc-minor-update branch from e3a7e78 to 2330bb4 Compare December 30, 2025 14:56

christiaanb reviewed Jan 2, 2026

View reviewed changes

jaschutte approved these changes Jan 7, 2026

View reviewed changes

Finalize handling of GHC issue in clash-ffi

5dba3d2

	# TODO: remove this and put it back into tests when
	# https://gitlab.haskell.org/ghc/ghc/-/merge_requests/12264#note_602406
	# is fixed

Update flake #3112

Are you sure you want to change the base?

Update flake #3112

Conversation

rowanG077 commented Dec 23, 2025

Uh oh!

Uh oh!

Uh oh!

rowanG077 commented Dec 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jaschutte commented Dec 24, 2025

Uh oh!

DigitalBrains1 left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

DigitalBrains1 commented Dec 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

DigitalBrains1 commented Dec 24, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

martijnbastiaan commented Dec 25, 2025

Uh oh!

DigitalBrains1 commented Dec 25, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

martijnbastiaan commented Dec 27, 2025

Uh oh!

DigitalBrains1 commented Dec 28, 2025

Uh oh!

DigitalBrains1 commented Dec 28, 2025

Uh oh!

DigitalBrains1 commented Dec 29, 2025

Uh oh!

rowanG077 commented Dec 29, 2025

Uh oh!

DigitalBrains1 left a comment

Choose a reason for hiding this comment

Uh oh!

rowanG077 commented Dec 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rowanG077 commented Dec 30, 2025 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

kleinreact commented Dec 30, 2025

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jaschutte commented Jan 7, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

rowanG077 commented Dec 24, 2025 •

edited

Loading

DigitalBrains1 left a comment •

edited

Loading

DigitalBrains1 commented Dec 24, 2025 •

edited

Loading

DigitalBrains1 commented Dec 24, 2025 •

edited

Loading

DigitalBrains1 commented Dec 25, 2025 •

edited

Loading

rowanG077 commented Dec 30, 2025 •

edited

Loading

rowanG077 commented Dec 30, 2025 •

edited

Loading