GitLab

While working on #13512, I noticed a buglet in the way that GHC implicitly quantifies type variables in associated type family instances. If you compile the following program with -ddump-rn-ast on HEAD:

{-# LANGUAGE TypeFamilies #-}
module Bug where

class C a where
  type T a b

instance C [a] where
  type T [a] b = Either a b

This is what the T instance's AST will look like:

       (ClsInstD
        ...
        (ClsInstDecl
         ...
         (L
          ...
          (HsSig
           (NoExtField)
           (HsOuterImplicit
            [{Name: a_agi}])
           ...
         [(L
           ...
           (TyFamInstDecl
            ...
            (FamEqn
             ...
             (L
              ...
              {Name: Bug.T})
             (HsOuterImplicit
              [{Name: a_agi}
              ,{Name: b_agj}])
             ...)))]))))

The suspicious part is the second HsOuterImplicit for the T TyFamInstDecl. It claims that it implicitly quantifies both a and b, but this isn't true—the a should already be in scope due to the first HsOuterImplicit for the instance head. I would expect the second HsOuterImplicit to only implicitly quantify b.

I'm not aware of any user-observable behavior that results from this bug, but it does cause issues when implementing a fix for #13512. This bug can be fixed independently of #13512, so I decided to open a separate issue for it.