Fix build and tests with GHC-9.4.3 and 9.2.5 #3364

mouse07410 · 2022-11-24T14:08:28Z

Fixes build with GHC-9.4.3.
Includes workaround for apparently-broken implicit-hie 0.1.3.

.github/workflows/test.yml

pepeiborra · 2022-11-24T19:40:53Z

haskell-language-server.cabal

+  if impl(ghc >= 9.4)
+    build-depends: ghc-exactprint >= 1.6.1


why is this needed?

Ah, a good question. Without this seemingly unnecessary constraint, Cabal somehow manages to pull ghc-exactprint-1.6.0 - presumably because there's a dependency of dependency that's asking for it. And CI tests kept failing.

With this addition, all the relevant tests passed (that is - at least, until I started applying changes to address suggestions ;).

I really think this needs to go in the right place. It shouldn't be that hard to figure out. At the very least it needs a clear comment explaining this.

I really think this needs to go in the right place

I attempted to find the right place - starting with attempts to put it in plugins/hls-refactor-plugin and ghcide. Failed miserably in both cases - tests kept failing. Please feel free to pitch a hand and locate that right place - I don't know nearly enough to dig deeper. I tried to search/grep for exactprint, and did not come up with anything helpful (besides the above, which did not help either).

If you do know that right place, please share. Or, perhaps, create a new PR - I'm already up to my ears in the bog of CI that I don't understand.

The current workaround was prompted by observation that

$ ghcup compile hls -g master --ghc 9.4.3 -- --index-state=@<current_date> --constraint="ghc-exactprint >= 1.6.1"

works.

How would you want me to explain it? Just copy-paste this post into the .cabal file?

It is not necessary, afaict, this worked for me out of the box:

> cabal update > cabal build -w ghc-9.4.3 exe:haskell-language-server

Interesting - perhaps, index-state changed since...? Should I update index-state in this PR too?

Also, have you tried just cabal build -w ghc-9.4.3 (to build whatever's there)?

Since you updated the index-state in cabal.project, you locally need to make sure your global index-state is newer than the one in cabal.project (which you added). So just running cabal update in the project or outside of it should do the trick.

Yeah, works just the same. It'd be weird if cabal didn't choose the most recent version of 'ghc-exactprint'.

Co-authored-by: Pepe Iborra <pepeiborra@me.com>

…ge-server

…l 9.4.3

georgefst · 2022-11-25T09:25:38Z

plugins/hls-fourmolu-plugin/hls-fourmolu-plugin.cabal

+    build-depends: fourmolu ^>= 0.8.2
+  if impl(ghc >= 9.4)
+    build-depends: fourmolu ^>= 0.9
+


I'm a bit confused by exactly why this is done this way (same as #3364 (comment), basically). But if it passes CI I'm fine with it.

I'll revisit all these bounds next time I come along to bump the version.

I'm a bit confused by exactly why this is done this way

Mainly, because I don't know any better. It seemed to be a crude but correct way to get tests pass.

I'll be happy if somebody with more knowledge fixes this the right way.

mouse07410 · 2022-11-25T13:05:33Z

I don't know what to do with this test failure (GHC-9.4.3, ubuntu-latest):

doc
      local define:                                                                   OK (1.50s)
      local empty doc:                                                                OK (1.52s)
      local single line doc without '\n':                                             FAIL (expected: Completion doc doesn't support ghc9) (1.51s)
        test/exe/Main.hs:2044:
        expected: ["*Defined at line 3, column 1 in this module*\n* * *\ndocdoc\n"]
         but got: ["*Defined at line 3, column 1 in this module*\n"] (expected failure)
      local multi line doc with '\n':                                                 FAIL (expected: Completion doc doesn't support ghc9) (1.51s)
        test/exe/Main.hs:[204](https://github.com/haskell/haskell-language-server/actions/runs/3545320619/jobs/5953373235#step:8:205)4:
        expected: ["*Defined at line 4, column 1 in this module*\n* * *\n abcabc\n"]
         but got: ["*Defined at line 4, column 1 in this module*\n"] (expected failure)
      local multi line doc without '\n':                                              FAIL (expected: Completion doc doesn't support ghc9) (1.51s)

fendor · 2022-11-25T13:17:46Z

The testcases you mention are expected failures and require no action.
However,

   addDependentFile
    file-changed
      test:                                                                           FAIL
        Exception: Timed out waiting to receive a message from the server.
        Last message received:
        {
            "jsonrpc": "2.0",
            "method": "window/logMessage",
            "params": {
                "message": "Live bytes: 29.27MB Heap size: 331.35MB",
                "type": 3
            }
        }
        Use -p '/file-changed.test/' to rerun this test only.

might be a genuine bug.

mouse07410 · 2022-11-25T14:09:37Z

Very likely so. But I don't see how this PR could be related, or how to fix it. Or even whether the fix should belong to this PR.

fendor · 2022-11-25T14:40:43Z

Well, if the PR that adds support for GHC 9.4.3, and the test-suite for GHC 9.4.3 breaks, then it is related because GHC changed something, somehow and now some assumption we previously had does not hold anymore. In other words, HLS is not compatible with GHC 9.4.3 out of the box.

A fix may very well belong into this PR, and the fix is likely to be rather minor, but potentially very hard to track down.
It is in theory possible to disable this specific test if you have a good reason, but I think fixing it is the right course of action.

mouse07410 · 2022-11-25T15:05:47Z

. . . HLS is not compatible with GHC 9.4.3 out of the box.

I call it "limited support". ;-)

It is in theory possible to disable this specific test if you have a good reason, but I think fixing it is the right course of action.

My only/main reason to ignore this test is that I don't have the slightest clue how to go about fixing, let alone actually fixing it.

The change to make HLS build and pass the majority of the tests seems fairly small and simple. The choice appears to be between providing some support for GHC-9.4.3 now, or waiting for an unknown length of time until somebody with enough knowledge figures out what's wrong and makes all the tests pass.

Also, what should we do with GHC-9.2.5? Stackage released LTS for it, but there's no official HLS support yet.

mouse07410 · 2022-11-25T16:53:52Z

Is there any way to make tests continue with other GHC versions rather than cancel, if one fails?

fendor · 2022-11-26T14:23:44Z

Is there any way to make tests continue with other GHC versions rather than cancel, if one fails?

No, this is by design to spare some CI resources.

mouse07410 · 2022-11-26T23:53:00Z

Is there any way to make tests continue with other GHC versions rather than cancel, if one fails?

No, this is by design to spare some CI resources.

I understand that. But here it looks like we need to know which GHC versions do and which do not fail the tests, with the given changes.

@fendor

Remove forcing `ghc-exactprint >= 1.6.1`, checking if @fendor 's suggestion is correct

pepeiborra · 2022-11-27T21:48:44Z

@mouse07410 you could mark the test as an expected failure in ghc 9.4.5. This will allow CI to continue past the error.

mouse07410 · 2022-11-28T03:27:30Z

@mouse07410 you could mark the test as an expected failure in ghc 9.4.5. This will allow CI to continue past the error.

Please excuse my ignorance. How to do that?

fendor · 2022-11-28T08:04:58Z

Use the knownBrokenFor function in front of the test-case, e.g. knownBrokenFor [GHC GHC94] "Not supported on GHC 9.4.3" $ testSession .... Another example at https://github.com/haskell/haskell-language-server/blob/master/ghcide/test/exe/Main.hs#L1989

mouse07410 · 2022-11-28T18:12:47Z

Use the knownBrokenFor function in front of the test-case, e.g. knownBrokenFor [GHC GHC94] "Not supported on GHC 9.4.3" $ testSession .... Another example at https://github.com/haskell/haskell-language-server/blob/master/ghcide/test/exe/Main.hs#L1989

I see, thanks. Frankly, I think it is beyond my competency to start messing with the code. Perhaps, one of the maintainers would care to tweak these?

wavewave · 2022-12-12T23:56:20Z

The currently broken master CI makes us hard to work on. (We are currently trying to land some changes for GHC 9.4 such as #3317 but could not make progress related to CI failures)
Is there any blocker for this to be merged and make CI succeed again? Thanks.
cc: @pepeiborra @michaelpj

pepeiborra · 2022-12-13T17:26:18Z

The currently broken master CI makes us hard to work on. (We are currently trying to land some changes for GHC 9.4 such as #3317 but could not make progress related to CI failures)
Is there any blocker for this to be merged and make CI succeed again? Thanks.
cc: @pepeiborra @michaelpj

@wavewave what is broken in master CI and what parts of this PR would help unblock you?

9999years · 2022-12-13T19:52:06Z

@wavewave what is broken in master CI and what parts of this PR would help unblock you?

C'mon, just look at the most recent master commit. It tells you exactly which checks failed:

924b932

You can see that most of the checks in #3317 are failing. We need this PR to fix those checks. I believe this PR will be necessary but perhaps not sufficient to fix those checks.

pepeiborra · 2022-12-13T20:00:50Z

C'mon, just look at the most recent master commit. It tells you exactly which checks failed:

924b932

Can you be more explicit? I don't see any checks failing other than the Nix ones.

You can see that most of the checks in #3317 are failing. We need this PR to fix those checks. I believe this PR will be necessary but perhaps not sufficient to fix those checks.

I see that there is a build error in a dependency:

Building library for apply-refact-0.10.0.0..
[508](https://github.com/haskell/haskell-language-server/actions/runs/3681170875/jobs/6227583497#step:4:509)
[1 of 5] Compiling Refact.Compat    ( src/Refact/Compat.hs, dist/build/Refact/Compat.o, dist/build/Refact/Compat.dyn_o )
[509](https://github.com/haskell/haskell-language-server/actions/runs/3681170875/jobs/6227583497#step:4:510)

[510](https://github.com/haskell/haskell-language-server/actions/runs/3681170875/jobs/6227583497#step:4:511)
src/Refact/Compat.hs:183:15: error:
[511](https://github.com/haskell/haskell-language-server/actions/runs/3681170875/jobs/6227583497#step:4:512)
Error:     Not in scope: type constructor or class ‘ErrorMessages’
[512](https://github.com/haskell/haskell-language-server/actions/runs/3681170875/jobs/6227583497#step:4:513)
    |
[513](https://github.com/haskell/haskell-language-server/actions/runs/3681170875/jobs/6227583497#step:4:514)
183 | type Errors = ErrorMessages
[514](https://github.com/haskell/haskell-language-server/actions/runs/3681170875/jobs/6227583497#step:4:515)
    |               ^^^^^^^^^^^^^
[515](https://github.com/haskell/haskell-language-server/actions/runs/3681170875/jobs/6227583497#step:4:516)
Error: Process completed with exit code 1.

Can you clarify how this PR fixes that?

wavewave · 2022-12-13T22:43:04Z

@pepeiborra
I probably do not follow the current status about CI correctly. My question was at more high level, and the CI badge on master commits is shown red cross mark, and the PR title says "it fixes build", so was asking whether we can merge and fix the CI. Thanks for trying to clarify the actual issue further.
The above issue you linked is another upstream issue for apply-refact, which should be solved by updating hackage with GHC-9.4 compatible version (the master of apply-refact is already updated) (note that this couldn't be solved locally here by source-repository-package since the newer version of apply-refact is breaking GHC 9.2 build.

pepeiborra · 2022-12-14T10:09:40Z

Superseded by #3396

michaelpj · 2022-12-14T11:46:20Z

It is certainly a problem that our CI checks are flaky, yeah. But I don't believe that master or anything else is actually broken.

Fix build and tests with GHC-9.4.3.

efba374

mouse07410 requested review from Anton-Latukha, Ailrun and pepeiborra as code owners November 24, 2022 14:08

pepeiborra reviewed Nov 24, 2022

View reviewed changes

mouse07410 and others added 6 commits November 24, 2022 15:57

Update .github/workflows/test.yml

322c03b

Co-authored-by: Pepe Iborra <pepeiborra@me.com>

Address @pepeiborra suggestions

5e51777

Address @pepeiborra suggestions

b3d0bd2

Merge branch 'master' of https://github.com/mouse07410/haskell-langua…

3e6d8ca

…ge-server

Disable GHC-9.4.3 on WIndows for now - test complains it can't instal…

12efbd6

…l 9.4.3

Attempt to work around fourmolu 0.9.0.0 build failure in Windows test

4b5bdb8

mouse07410 requested a review from georgefst as a code owner November 25, 2022 04:26

georgefst reviewed Nov 25, 2022

View reviewed changes

georgefst approved these changes Nov 25, 2022

View reviewed changes

Add build with GHC-9.2.5

f1e2481

mouse07410 changed the title ~~Fix build and tests with GHC-9.4.3.~~ Fix build and tests with GHC-9.4.3 and 9.2.5 Nov 25, 2022

mouse07410 added 2 commits November 25, 2022 13:45

Merge remote-tracking branch 'upstream/master'

eca61d1

Temporarily disable GHC-9.2.5 for Windows

4e489cb

mouse07410 added 2 commits November 26, 2022 22:56

Merge remote-tracking branch 'upstream/master'

362b8d2

Update haskell-language-server.cabal

23a6279

Remove forcing `ghc-exactprint >= 1.6.1`, checking if @fendor 's suggestion is correct

Merge remote-tracking branch 'upstream/master'

c359743

fendor mentioned this pull request Dec 2, 2022

Add CI flows for GHC 9.2.5 #3376

Merged

Merge remote-tracking branch 'upstream/master'

c6f1a0a

Merge branch 'master' into master

8a491f5

pepeiborra mentioned this pull request Dec 13, 2022

Add CI flows for 9.4.3 #3396

Merged

Merge branch 'master' into master

b474547

pepeiborra closed this Dec 14, 2022

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Fix build and tests with GHC-9.4.3 and 9.2.5 #3364

Fix build and tests with GHC-9.4.3 and 9.2.5 #3364

mouse07410 commented Nov 24, 2022 •

edited by gitpod-io bot

Loading

pepeiborra Nov 24, 2022

mouse07410 Nov 24, 2022 •

edited

Loading

michaelpj Nov 25, 2022

mouse07410 Nov 25, 2022 •

edited

Loading

fendor Nov 27, 2022

mouse07410 Nov 27, 2022

fendor Nov 27, 2022

georgefst Nov 25, 2022

mouse07410 Nov 25, 2022

mouse07410 commented Nov 25, 2022

fendor commented Nov 25, 2022

mouse07410 commented Nov 25, 2022

fendor commented Nov 25, 2022

mouse07410 commented Nov 25, 2022 •

edited

Loading

mouse07410 commented Nov 25, 2022

fendor commented Nov 26, 2022

mouse07410 commented Nov 26, 2022

pepeiborra commented Nov 27, 2022

mouse07410 commented Nov 28, 2022

fendor commented Nov 28, 2022

mouse07410 commented Nov 28, 2022

wavewave commented Dec 12, 2022

pepeiborra commented Dec 13, 2022

9999years commented Dec 13, 2022 •

edited

Loading

pepeiborra commented Dec 13, 2022

wavewave commented Dec 13, 2022

pepeiborra commented Dec 14, 2022

michaelpj commented Dec 14, 2022

Fix build and tests with GHC-9.4.3 and 9.2.5 #3364

Fix build and tests with GHC-9.4.3 and 9.2.5 #3364

Conversation

mouse07410 commented Nov 24, 2022 • edited by gitpod-io bot Loading

pepeiborra Nov 24, 2022

Choose a reason for hiding this comment

mouse07410 Nov 24, 2022 • edited Loading

Choose a reason for hiding this comment

michaelpj Nov 25, 2022

Choose a reason for hiding this comment

mouse07410 Nov 25, 2022 • edited Loading

Choose a reason for hiding this comment

fendor Nov 27, 2022

Choose a reason for hiding this comment

mouse07410 Nov 27, 2022

Choose a reason for hiding this comment

fendor Nov 27, 2022

Choose a reason for hiding this comment

georgefst Nov 25, 2022

Choose a reason for hiding this comment

mouse07410 Nov 25, 2022

Choose a reason for hiding this comment

mouse07410 commented Nov 25, 2022

fendor commented Nov 25, 2022

mouse07410 commented Nov 25, 2022

fendor commented Nov 25, 2022

mouse07410 commented Nov 25, 2022 • edited Loading

mouse07410 commented Nov 25, 2022

fendor commented Nov 26, 2022

mouse07410 commented Nov 26, 2022

pepeiborra commented Nov 27, 2022

mouse07410 commented Nov 28, 2022

fendor commented Nov 28, 2022

mouse07410 commented Nov 28, 2022

wavewave commented Dec 12, 2022

pepeiborra commented Dec 13, 2022

9999years commented Dec 13, 2022 • edited Loading

pepeiborra commented Dec 13, 2022

wavewave commented Dec 13, 2022

pepeiborra commented Dec 14, 2022

michaelpj commented Dec 14, 2022

mouse07410 commented Nov 24, 2022 •

edited by gitpod-io bot

Loading

mouse07410 Nov 24, 2022 •

edited

Loading

mouse07410 Nov 25, 2022 •

edited

Loading

mouse07410 commented Nov 25, 2022 •

edited

Loading

9999years commented Dec 13, 2022 •

edited

Loading