No, the type instance must match the class heading.

I can use

instance Foo [_a] where
  type Bar [_a] = Int
  whatever = ... where
    bar :: _a -> Int
    bar = ...

but that is a needlessly messy thing to request of every package everywhere.

The arguments being pattern matched in a class associated type aren't really just bindings, they reference the surrounding context and so shouldn't be treated the same as the basic type family case.

It isn't just the class associated type being mangled, it is every type variable that comes from the instance head in the entire body of every instance that happens to have a class associated type in it.

Note that in the example above I added another ScopedTypeVariables reference to the same parameter, but it _also_ must be mangled despite having absolutely nothing to do with the class associated type.

The existing convention has worked since 6.10 or so, when all of this stuff was invented in the first place, and the new state of affairs is clearly worse.

-Edward

On Sun, Jan 17, 2016 at 3:16 AM, Andrew Farmer <xichekolas@gmail.com> wrote:

Can't you just:

instance Foo [a] where
  type Bar [_a] = Int

(At least I think I did that somewhere...)

On Jan 16, 2016 9:24 PM, "Edward Kmett" <ekmett@gmail.com> wrote:
As a data point I now get thousands of occurrences of this warning across my packages.

It is quite annoying.

class Foo a where
  type Bar a

instance Foo [a] where
  type Bar [a] = Int

is enough to trigger it.

And you can't turn it off by using _ as

instance Foo [_] where
  type Bar [_] = Int

isn't legal.

I've been avoiding it for now by using

  if impl(ghc >= 8)

    ghc-options: -fno-warn-unused-matches

but this is a pretty awful addition to this warning as it stands.

-Edward

On Mon, Jan 11, 2016 at 2:12 PM, Henning Thielemann <lemming@henning-thielemann.de> wrote:

On Mon, 11 Jan 2016, Richard Eisenberg wrote:

On Jan 9, 2016, at 6:44 PM, Henning Thielemann <lemming@henning-thielemann.de> wrote:

instance (Natural n) => Num.Integer (Un n) where
   type Repr (Un _n) = Unary


GHC-7.6.3 and GHC-7.4.2 complain:
   Type indexes must match class instance head
   Found `Un _n' but expected `Un n'
   In the type synonym instance declaration for `Num.Repr'
   In the instance declaration for `Num.Integer (Un n)'


GHC-7.8.4, GHC-7.10.3 and GHC-8.0 are happy with the difference.

I'm surprised this is accepted at all. Looks like hogwash to me. I think you should post a bug report.

 Ok, but then GHC must not warn about the unused argument of Repr.

_______________________________________________
Glasgow-haskell-users mailing list
Glasgow-haskell-users@haskell.org
http://mail.haskell.org/cgi-bin/mailman/listinfo/glasgow-haskell-users


_______________________________________________
Glasgow-haskell-users mailing list
Glasgow-haskell-users@haskell.org
http://mail.haskell.org/cgi-bin/mailman/listinfo/glasgow-haskell-users