Two recompilation avoidance related bug fixes #3452
Conversation
|
Do we definitely want to include the mitigation for issue 2? would it be better to just fix in GHC and let it be in the next minor 9.4 release? |
It is unlikely to be fixed in a minor GHC release because the simplest way to perform the fix involves an API change. |
|
That's a shame :( |
| filePathMap = unsafePerformIO $ newIORef HashMap.empty | ||
| {-# NOINLINE filePathMap #-} | ||
|
|
||
| shareFilePath :: FilePath -> FilePath |
There was a problem hiding this comment.
This and shareUsages are actually called in IO, so we could just have them be in IO and have slightly less unsafety.
There was a problem hiding this comment.
I moved shareFilePath to a different module, and I think it makes sense for it to be pure so that it can be used in other places in the future.
|
I've tried this MR and I can confirm that HLS is way faster and seems to not rebuild multiple files at each changes. However, some files are failing with: Because And any change in any file which appears in my workspace error log with such error does not seem to be taken into account anymore in files which are depending on them. |
wz1000
force-pushed
the
wip/usage-fixes
branch
from
d7efbc3
to
ff7d8ed
Compare
Thanks, I've fixed this. |
Thank you. I've tested and it is fast like it had never been before AND it does not fail like described above. I'm pushing the fix in production tomorrow, I'll tell you if developers are happy or grumpy ;) |
wz1000
force-pushed
the
wip/usage-fixes
branch
from
ff7d8ed
to
c918470
Compare
1. Recompilation avoidance regresses in GHC 9.4 due to interactions between GHC and HLS's implementations. Avoid this by filtering out the information that causes the conflict See https://gitlab.haskell.org/ghc/ghc/-/issues/22744. 2. The recompilation avoidance info GHC stores in interfaces can blow up to be extremely large when deserialised from disk. See https://gitlab.haskell.org/ghc/ghc/-/issues/22744 Deduplicate these filepaths.
wz1000
force-pushed
the
wip/usage-fixes
branch
from
1c6601d
to
ae6ea9b
Compare
1. Recompilation avoidance regresses in GHC 9.4 due to interactions between GHC and HLS's implementations. Avoid this by filtering out the information that causes the conflict See https://gitlab.haskell.org/ghc/ghc/-/issues/22744. 2. The recompilation avoidance info GHC stores in interfaces can blow up to be extremely large when deserialised from disk. See https://gitlab.haskell.org/ghc/ghc/-/issues/22744 Deduplicate these filepaths. (cherry picked from commit 00f4e61)
1. Recompilation avoidance regresses in GHC 9.4 due to interactions between GHC and HLS's implementations. Avoid this by filtering out the information that causes the conflict See https://gitlab.haskell.org/ghc/ghc/-/issues/22744. 2. The recompilation avoidance info GHC stores in interfaces can blow up to be extremely large when deserialised from disk. See https://gitlab.haskell.org/ghc/ghc/-/issues/22744 Deduplicate these filepaths.
Recompilation avoidance regresses in GHC 9.4 due to interactions between GHC and HLS's implementations.
Avoid this by filtering out the information that causes the conflict
See Recompilation avoidance in GHC 9.4 conflicts with recompilation avoidance in HLS #3450.
The recompilation avoidance info GHC stores in interfaces can blow up to be
extremely large when deserialised from disk. See https://gitlab.haskell.org/ghc/ghc/-/issues/22744
Deduplicate these filepaths.