Update refactor/splice/hlint/fourmolu/retrie/gadt plugin for GHC 9.4 #3317
Conversation
9999years
force-pushed
the
rebeccat/ghc94
branch
from
3123648 to
b2d735b
Compare
|
Let me know if you need any help with ls-refactor-plugin! |
9999years
force-pushed
the
rebeccat/ghc94
branch
from
4ade99c to
39188b4
Compare
9999years
force-pushed
the
rebeccat/ghc94
branch
2 times, most recently
from
2a2df37 to
3459f48
Compare
flake.nix
Outdated
| # The GHC 9.4 pull request builds upon these unpublished changes, | ||
| # which is why we include them. | ||
| # | ||
| # Also, there is no PR for these changes. The fourmolu maintainers |
There was a problem hiding this comment.
this isn't true, for example the first commit is this pr: fourmolu/fourmolu#229
There was a problem hiding this comment.
OK, but this is just for testing. We'll have to release a new Fourmolu before this PR can build outside of Nix.
| -- the error messages are significantly different from the ones without that, and | ||
| -- thus the code action triggers (such as in hls-refactor-plugin) are not fired properly. | ||
| -- So we disable the error demotion. | ||
| demoteIfDefer = id |
There was a problem hiding this comment.
I actually couldn't figure out yet what would be the best action here. What I observed is that due to the error message difference between the case with and without defer, several hls-refactor-plugin suggestions are not fired (which uses matchRegex on error messages). With defer turned on, on GHC 9.4, there is no way to obtain source location (which relied on a fragile mechanism of parsing error message), so if we do not want to have this brute-force disabling, we need to make those suggestions and tests disabled.
There was a problem hiding this comment.
Disabling deferred type errors will have an impact on many other IDE features, including diagnostics and code navigation. For this reason, I think the only alternatives here are:
- update the regexes to account for the changes in 9.4 (incrementally, in other PRs if necessary)
- migrate the suggestion providers to use typed diagnostics
There was a problem hiding this comment.
Thanks for the feedback!
I will revert this disabling and mark the failed tests due to this with comments. The fix for the failed tests will not be small as previously the necessary error source location info was obtained from the error message which is gone with the deferred type error flag, so let us address that in a separate PR. After all, typed diagnostics is the best direction, so we had better move on in that direction.
There was a problem hiding this comment.
reverted the change and mark the affected test as broken.
9999years
force-pushed
the
rebeccat/ghc94
branch
from
a20ee04 to
1b027bf
Compare
|
retrie 1.2.1 is now available in Hackage |
- Build against GHC 9.4 API - Increase upper bound on `hlint`
9999years
requested review from
michaelpj,
wavewave,
pepeiborra,
brandonchinn178,
fendor and
konn
and removed request for
brandonchinn178,
michaelpj,
wavewave,
pepeiborra and
fendor
|
Sorry I don't know how the GitHub reviewers mechanic really works I guess |
wz1000
changed the title
|
I've added 9.4 support for the retrie and gadt plugins to this PR. |
wz1000
force-pushed
the
rebeccat/ghc94
branch
from
3787c9e to
410d0d0
Compare
wz1000
changed the title
WIP tracking work on GHC 9.4 support for HLS plugins, particularly interested in
hls-refactor-plugin.We will need
fourmoluto cut a new release before CI can start building this PR successfully. fourmolu/fourmolu#242 (comment)