[Haskell-cafe] What's different about quantified constraints since GHC 9.2?

[Richard Eisenberg]

If I remember correctly, we added more aggressive expansion of superclasses in quantified-constraint givens. But it's all a bit surprising that it works, given that the superclass given will have a type family in an instance head, which is supposed to not happen (because inference would be fragile, not because there's anything unsound).

Richard

> On Sep 20, 2022, at 5:07 PM, Li-yao Xia <lysxia at gmail.com> wrote:
> 
> Dear Café,
> 
> There's a trick [1] involving quantified constraints (and usually type families) where a function has a quantified constraint (forall a. CF a) and which is explicitly instantiated with a well-placed type annotation (_ :: CF x => ...). Since GHC 9.2, this trick works without that type annotation. How did GHC get smarter at instantiating this quantified constraint? Below is a minimized example which compiles on GHC 9.2.1 but not 9.0.1 (haven't tested 9.0.2), unless you uncomment the last line.
> 
> Cheers,
> Li-yao
> 
> [1]: an Iceland_jack trick https://gitlab.haskell.org/ghc/ghc/-/issues/14860#note_151394
> 
> 
> {-# LANGUAGE TypeFamilies, FlexibleInstances, FlexibleContexts, QuantifiedConstraints, ScopedTypeVariables #-}
> module A where
> 
> type family F a
> class C a
> class C (F a) => CF a
> 
> f :: C (F a) => a
> f = undefined
> 
> g :: (forall a. CF a) => a
> g = f
>   -- :: forall a. CF a => a  -- needed until GHC 9.0.1 but not GHC 9.2.1 (and later)
> 
> _______________________________________________
> Haskell-Cafe mailing list
> To (un)subscribe, modify options or view archives go to:
> http://mail.haskell.org/cgi-bin/mailman/listinfo/haskell-cafe
> Only members subscribed via the mailman list are allowed to post.