antc2 / ghc-proposals Goto Github PK

I withdrew the 'Instance Apartness Guards' proposal, because it was getting no traction.

But the difficulties it was trying to address haven't gone away. Trac #15632, particularly this proposed rule is highlighting the nastiest of those difficulties.

@spj I have no idea what to do about (overlap + FDs).

The first thing to do is stop thinking about these as two extensions, somewhat orthogonal; instead see that FunDeps rely on overlaps. Because by all the theory these two instance are supposed to be equivalent, and they overlap

class C a b  | a -> b

instance C Int Bool
instance (b ~ Bool) => C Int b

But GHC exhibits different behaviour for them, as Trac #15632 is demonstrating: the direct C Int Bool can be 'fooled'/overridden into delivering C Int Char, if that second instance is

instance (b ~ Char) => C Int b

Then my suggested rules are trying to catch/prevent that: for those two instances, although the head C Int Bool is strictly more specific, the argument positions C Int ... are not more specific (they're the same), so those instances fail on rule B) iii).

This github Issue is for more ruminations.