strong type derivation in ghc 6.12 compared to 6.12

[Seyed Hosein Attarzadeh Niaki]

I have problems with upgrading some already working code for GHC 6.12.
The following code:

-----------------------------------------------------------
{-# LANGUAGE DeriveDataTypeable,
             GeneralizedNewtypeDeriving, TemplateHaskell  #-}

 import Data.Data
 import Language.Haskell.TH.Syntax

 class Nat a
 class (Data a, Lift a) => ProcType a


 instance (Lift a, Nat s) => Lift (FSVec s a) where
  lift (FSVec xs) = undefined


 newtype Nat s => FSVec s a = FSVec {unFSVec :: [a]}
  deriving (Eq, Typeable, Data)

 -- The following line does not cause a problem in GHC 6.10
 -- instance (Typeable s, Nat s, ProcType a) => ProcType (FSVec s a)
-----------------------------------------------------------

works fine in ghc 6.10.4 and I get (extra stuff removed):
//
*Main> :i FSVec
newtype (Nat s) => ...
instance (Data a, Typeable s, Nat s) => Data (FSVec s a)
  -- Defined at ghctest.hs:16:26-29
...
//

but in 6.12.2 what I get is:
//
*Main> :i FSVec
newtype (Nat s) => ...
instance (Data s, Data a, Nat s) => Data (FSVec s a)
...
//

So, although the last line in the above code compiles correctly in ghc 
6.10.4, in 6.12.2 GHC complains with:
//
ghctest.hs:19:10:
    Could not deduce (Data s)
      from the context (Typeable s, Nat s, ProcType a)
      arising from the superclasses of an instance declaration
                   at ghctest.hs:19:10-64
    Possible fix:
      add (Data s) to the context of the instance declaration
    In the instance declaration for `ProcType (FSVec s a)'
Failed, modules loaded: none.
//

Why are they different? Why 6.12 derives a stronger constraint? Is it a 
bug or it was not working correctly before?

-- 
Hosein Attarzadeh