core-libraries-committee
core-libraries-committee copied to clipboard
haskell
Replace `Foreign.Marshal.Error.void` with a re-export of `Data.Functor.void`
These two functions have identical behavior and names, but differing types. (This can cause annoying and pointless "ambiguous occurrence" errors when both are in scope.) Data.Functor.void is more general than Foreign.Marshal.Error.void, but the proposed change is nevertheless a breaking one since the more general type of Data.Functor.void can cause type inference to fail in some cases.
But Foreign.Marshal.Error.void has been deprecated since base-4.6.0.0, which shipped with ghc-7.6.1 and was released in September 2012, more than a decade ago. Potentially-affected users have had plenty of time to migrate.
Side note: Foreign.Marshal.Error.void is specified in the Haskell2010 report; see this chapter. Otherwise I might propose to just delete this function instead of generalizing it.
Given that the function is deprecated for a decade, I don’t see a reason not to delete it. We are way far away from Haskell Report to bother about it.
@tomjaguarpaw @mixphix @hasufell @angerman @parsonsmatt @velveteer any more opinions about Foreign.Marshal.Error.void?
We are way far away from Haskell Report to bother about it.
I would like a base that conforms to the spec or an update to the spec.
Do we know why the deprecation was introduced?
I don't see base in the 7.6 tree https://gitlab.haskell.org/ghc/ghc/-/tree/ghc-7.6/libraries
I don't have a strong opinion between the three options (no change, replace with general version, remove it).
Do we know why the deprecation was introduced?
I don't see base in the 7.6 tree https://gitlab.haskell.org/ghc/ghc/-/tree/ghc-7.6/libraries
It was deprecated by @simonmar in https://gitlab.haskell.org/ghc/ghc/-/commit/99490045a60eca41e79c2158d839a65cf454de2b, Apr 2012, but there is no explanation or discussion attached.
Given the deprecation message I guess the reason was that the more general version of void should be preferred!
{-# DEPRECATED void "use Control.Monad.void instead" #-}
I think its removal is safe by now. I think if Haskell should stick to a Report, it should be a modern one, but such a Report has not yet been written.
@parsonsmatt @angerman @velveteer any more opinions on the preferred course of actions here?
I'm in favor of removing it or replacing with a re-export of Data.Functor.void.
Another point to consider: A fair number of packages have import Foreign hiding ( void ) which will trigger a -Wdodgy-imports warning if void is completely removed from Foreign.Marshal.Error. That's not a big deal: there are several acceptable ways of dealing with this warning, including silencing it. But it does probably create busy-work for the relevant maintainers, which could be avoided by replacing Foreign.Marshal.Error.void with a re-export of Data.Functor.void.
But I won't sleep any less soundly if the committee still prefers to delete the function instead.
@clyring we don't consider warnings breaking changes: https://github.com/haskell/core-libraries-committee/issues/262
Werror users get what they asked for.
I think we all agree that Foreign.Marshal.Error is not a right place for void :: Functor f => f a -> f (), it's just that CLC members may have different appetite when it comes to breakage.
As noted in the proposal, replacing void :: IO a -> IO () with void :: Functor f => f a -> f () is also a breaking change, because it can potentially make type checking ambiguous. So it would require an impact assessment, same as just removing it. So I guess it's a question which impact assessment (and patches) @clyring is motivated enough to perform.
If we go for re-export, let's deprecate it, so that we do not close the door to the eventual removal of the function.
@clyring unless there is some progress by the end of August, I'll have to close the proposal as abandoned.
Closing as abandoned, feel free to reopen when there are resources to continue.