search

haskell / win32

Haskell support for the Win32 API
http://hackage.haskell.org/package/Win32
Other
97 stars 62 forks source link

registry001 on 2.6.2.0 fails on GHC CI due to compile error #181

Closed wz1000 closed 3 years ago

wz1000 commented 3 years ago

The definition of x is ambiguous due to the extra exports.

Patch:

diff --git a/tests/registry001.hs b/tests/registry001.hs
index ade0eba..27a3cb9 100644
--- a/tests/registry001.hs
+++ b/tests/registry001.hs
@@ -1,4 +1,4 @@
-import System.Win32
+import System.Win32 hiding ( COORD(..) )
 import Control.Exception
 import Control.Monad

It would be nice if this could be backported to 2.6.2.1 so we don't ship a version of Win32 with GHC 8.10.6 which fails to pass the testsuite. It is not exactly critical since it should have little user facing impact, but it would be nice to have (very soon if at all possible since 8.10.6 is very close to ready).

This was noticed when I bumped Win32 for https://gitlab.haskell.org/ghc/ghc/-/issues/20202

I do wish we had a better story for GHC tests which are really a part of submodules. This has come up twice in the last few weeks now.

Mistuke commented 3 years ago

I won't be able to do it for that same package version but can release a new one with the fix.

I do wish we had a better story for GHC tests which are really a part of submodules. This has come up twice in the last few weeks now.

There is a ticket on gitlab about decoupling the ghc testsuite from GHC. That shouldn't be too hard as the testsuite just reads a few config files. But no one has done so yet.

I'll make the change and release tonight.

Thanks!

wz1000 commented 3 years ago

I won't be able to do it for that same package version but can release a new one with the fix.

Yes. The version with the problem is 2.6.2.0, I suggested a bump to the last component since this wouldn't affect the user facing parts of the library.

Thinking about it more, it might just be best to backport the fix for #157, which would necessitate a major version bump. I'm not sure how much code will break due to #157 so I leave it up to your judgement.

Thanks!

Mistuke commented 3 years ago

Major versions bumps can't be done for win32 without bumping the bounds in various boot libraries for ghc (cabal and process).

But in this case that would also conflict with the existing 2.7 already (win32 is at 2.12)

Mistuke commented 3 years ago

Merged via 8103be38865de160fc1ef9e69e84ad87d2556e77 and release trigger for v2.6.2.1 is running and should be on hackage in 30mins.