standalone deriving for typeable fails

[jpbernardy]

import Data.Typeable

data D a = C1 a | C2 (D a)
deriving instance Typeable a => Typeable (D a)
--           deriving Typeable

x = show $ typeOf (undefined::D Int)

main = print x

yields:

[1 of 1] Compiling Main             ( T.hs, T.o )
ghc-6.8.1: panic! (the 'impossible' happened)
  (GHC version 6.8.1 for i386-unknown-linux):
        No match in record selector Var.tcTyVarDetails

[simonpj]

Oh bother! The deriving for Typeable is unique. If you use the deriving clause on the data decl, you'll get this instance:

instance Typeable1 D where
  typeOf1 _ = mkTyConApp (mkTyCon "D") []

So that's what the user should write in a standalone deriving clause too. thus

deriving instance Typeable1 D 

(There's a uniform instance for type applications.)

Sadly, GHC 6.8.1 rejects the above because "Typeable1 is not a derivable class".

Ok so this is broken and there's no workaround. I will milestone it for 6.8.1. What I propose is to require you to write "deriving instance Typeable1 D". That's not as obvious as the declaration you started with, but it fits with the current impl, and with the (current) Typeable design; see Section 7.4 of  the second SYB paper.

[simonpj]

PS: a smaller program that generates the error, when compiled with -dcore-lint is:

{-# OPTIONS -dcore-lint -XDeriveDataTypeable -XStandaloneDeriving #-}
module Foo where
data D a = C1 a | C2 (D a)
deriving instance Typeable a => Typeable (D a)

[igloo]

Is it hard to do the same magic for "deriving instance Typeable ..." that is done for "data ... deriving Typeable"? It seems a bit unpleasant that the same thing doesn't work in both cases.

[simonpj]

Fixed. Thank you for reporting it.

Merge to stable:

Tue Nov 20 12:57:32 GMT 2007  simonpj@microsoft.com
  * FIX Trac #1825: standalone deriving Typeable

[igloo]

Merged