GHC complains about missing instance in conjunction with GADTs

I wanted to do something like that:

data Pair :: (* -> *) -> * where
    Pair :: a b -> b -> Pair a

data Sel :: * -> * where
    A :: Sel Bool
    B :: Sel Integer

showSnd :: Pair Sel -> String
showSnd (Pair A bool)    = show bool
showSnd (Pair B integer) = show integer

However, GHC 6.4.1 with -fglasgow-exts complains in the second last and last line that there is no instance for (Show b). But it is clear that bool :: Bool and integer :: Integer and that therefore bool and integer are showable. In addition, Chris Kuklewicz said the following:

I cut and pasted your code into ghci-6.4.2 with -fglasgow-exts and there was no

error. And I can (print (showSnd (Pair A True)) and (print (showSnd (Pair B 8))) but when I try to compile it with ghc-6.4.2 it complains about the Show instance. I find this deeply wierd, and it seems like a bug.

Trac metadata

Trac field	Value
Version	6.4.1
Type	Bug
TypeOfFailure	OtherFailure
Priority	normal
Resolution	Unresolved
Component	Compiler
Test case
Differential revisions
BlockedBy
Related
Blocking
CC
Operating system
Architecture

changed weight to 5

added Tbug Trac import labels

Seems all right with GHC 6.4.2 to me. I dn't nkow what Chris thinks doesn't work.

Mind you, type classes and GADTs don't work togeher properly, still, so it's good luck that this one is ok.

Simon

simonpj wrote:

Seems all right with GHC 6.4.2 to me. I dn't nkow what Chris thinks doesn't work.

What Chris means is that the example code is accepted by GHCi 6.4.2 (which is the right behavior, in my opinion) but it's not accepted by the ordinary GHC compiler of the same version.

The odd thing is that it works with 6.4.2 (batch, not GHCi). Maybe Chris can try again?

My comment about luck is simply that GHC is badly broken when it comes to a mixture of type classes and GADTs. It's something high on my priority list, but for you can't rely on it to behave well, or even predictably.

assigned to @simonpj

mentioned in commit 5a6c7d62

This case seems fine in 6.6, but I assume the bug is being kept open for other, similar issues?

Trac metadata

Trac field	Value
Test case	- → gadt20

closed

I think this is actually ok in 6.6, and in any case I'm fixing the type-class interaction, and we have a test case to remind us if it goes bad on us.

So I'm closing this bug.

Simon

Trac metadata

Trac field	Value
Resolution	Unresolved → ResolvedFixed

changed milestone to %6.8.1

added Pnormal label