Prepare for GHC 9.2.* upgrade #3304
Conversation
ChrisPenner
force-pushed
the
ghc-9.2.3-cleanup
branch
from
7102b19
to
e38ae1e
Compare
ChrisPenner
force-pushed
the
ghc-9.2.3-cleanup
branch
from
e38ae1e
to
b0345a3
Compare
| -- Call False n as | ||
| -- | False -- known unsaturated call | ||
| -- = | ||
| -- Ins (Name (Env n 0) as) $ Yield (BArg1 0) |
There was a problem hiding this comment.
@dolio Were these "always failing" case matches here for a reason? They're triggering 'redundant match' in GHC 9.2.3
There was a problem hiding this comment.
They were just there to remind me to some day implement that feature. I don't think it's a great loss to just remove them, or comment them.
ChrisPenner
force-pushed
the
ghc-9.2.3-cleanup
branch
from
afbc34d
to
4652149
Compare
|
Looks like HLS isn't fully supported on 9.2.3 yet, I think we should wait until we have case-splitting at least since that's handy; I tried upgrading to 9.0.2, but that's a no-go due to a QuantifiedConstraints bug (I linked the issue in the PR description). I'll probably just remove the actual GHC upgrade and we can merge all the busy-work, that way we're ready to upgrade whenever HLS catches up 👍🏼 |
These are all the changes which are incompatible between 8.10.7 and GHC 9.2.3
| @@ -261,14 +261,7 @@ pretty0 | |||
| | otherwise -> | |||
| paren (p >= 11 || isBlock x && p >= 3) $ | |||
| fmt S.DelayForceChar (l "'") | |||
| <> ( case x of | |||
| Lets' _ _ -> id | |||
There was a problem hiding this comment.
This case is impossible because of the earlier pattern match on Lets' against x.
GHC 9.2.3 is smart enough to detect that and issues a warning here.
There was a problem hiding this comment.
I think we should keep that deleted comment
| @@ -997,7 +999,9 @@ wantRequest :: | |||
| Term v loc -> | |||
| Type v loc -> | |||
| (Type v loc, Wanted v loc) | |||
| wantRequest loc ~(Type.Effect'' es t) = (t, fmap (Just loc,) es) | |||
There was a problem hiding this comment.
GHC got confused and even though this pattern is irrefutable it wouldn't believe that the patterns were exhaustive, even if I added the COMPLETE pragma for Effect'', so I just used unEffect0 instead, which is all that the Effect'' pattern does.
| @@ -141,7 +141,7 @@ instance Show Stanza where | |||
| API apiRequests -> | |||
| "```api\n" | |||
| <> ( apiRequests | |||
| & fmap (\(GetRequest txt) -> Text.unpack txt) | |||
There was a problem hiding this comment.
This uni-pattern was actually a bug in the transcript parser before, 9.2.3 caught it 🎉
| @@ -32,4 +32,6 @@ reference_ = | |||
|
|
|||
| toShortHash :: ConstructorReference -> ShortHash | |||
| toShortHash (ConstructorReference r i) = | |||
| (Reference.toShortHash r) {ShortHash.cid = Just (tShow i)} | |||
There was a problem hiding this comment.
This was a partial record update before.
| @@ -0,0 +1,71 @@ | |||
| {-# LANGUAGE ExistentialQuantification #-} | |||
There was a problem hiding this comment.
This file was deleted on trunk; accidentally added back in a merge?
ChrisPenner
added
ready-to-merge
Overview
Instead of upgrading actual GHC, this is now a patch containing all the changes which will be necessary for the upgrade. It looks like HLS support will be coming soon, and we can complete the upgrade at that time by reverting this commit
Upgrades to GHC 9.2.3We'll want to upgrade Enlil at the same time: https://github.com/unisoncomputing/enlil/pull/74Built on #3117
Implementation notes
error, so now I could delete them 👌🏼 )error's<>to be defined instead ofmappend)Loose Ends
When we're ready to upgrade we can simply revert this commit to complete the process.
This change is