(More from the annals of oh-so-easy in yacc.)

On Thu, 4 Oct 2018 at 12:45 AM, Anthony Clayden <anthony_clayden@clear.net.nz> wrote:
A typical consequence of combining FunDeps + Overlapping instances is that you have to make the result parameter more general than it needs be, then use a TypeCast constraint to improve it.

... TypeCast, which does use FunDeps to mutually improve/unify its two parameters. Nowadays in GHC you'd use the (~) constraint.

I got so fed up writing out TypeCast constraints; and I'm so used to writing infix (~) in GHC, I implemented (~) in Hugs. This is hard-coded syntax in constraints; I've not implemented type operators. (Mark you, it's also hard-coded in GHC, because (~) is a reserved symbol.)

This needed only changes to the yacc syntax. From that I insisted that "~" is a legitimate name for a class. There is a downside that equality constraints are printed with the '~' prefix -- because all class names are prefix.

To be clear: this is not as powerful or well-integrated as (~) in GHC. You still need at the term level to explicitly cast.

Then I got so fed up writing out explicit typeCast calls ..., I picked up the postfix operators idea
https://mail.haskell.org/pipermail/hugs-users/2018-September/000909.html
and invented a postfix operator (~::) to do the job.
Trailing double-colon says I'm doing something typeful; tilde connects it to the equality constraint.

Then here's the classic, compiled in Hugs, also exhibiting the FunDeps + Overlaps combo

class TypeEq t t' r  | t t' -> r  where
  typeEq :: t -> t' -> r

instance TypeEq t t TTrue  where
  typeEq _ _ = TTrue

instance (TFalse ~ f) => TypeEq t t' f  where
  typeEq _ _ = (TFalse ~::)      -- without explicit cast, complains type is not general enough


AntC