Hi!
Ghc seems to have some problems in inferring the most general types of
RULES pragmas. The bug was not present in 5.02.1 but shows up in a cvs
version checked out yesterday.
Here's the problem:
Consider the following example from the user's guide:
{-# RULES
"map/map" forall f g xs. map f (map g xs) = map (f.g) xs
#-}
When I compile with -ddump-rules I get the following rule:
"map/map" __forall {@ b
@ b
@ a
f :: (b -> b)
g :: (a -> b)
xs :: [a]}
GHC.Base.map @ b @ b f (GHC.Base.map @ a @ b g xs)
= GHC.Base.map @ a @ b (GHC.Base.. @ b @ b @ a f g) xs ;
(And it also turns out this rule is listed among the imported rules, not
the local ones, which is a bit confusing.)
The rule says that f must have type b -> b for some b. This is clearly Bad
as the matching algorithm also matches types and the rule will not be
fired in all the cases it should.
When I add a type signature for f then g get type b -> b. In order to get
the desired behaviour I have to add type signatures to both f and g.
I hope this is easy to fix.
Cheers,
/Josef
PS. Would it be possible to add a -fwarn-pragma which warns when a pragma
is ignored? It is quite frustrating when some rules are silently ignored
and you keep wonder why the rules don't fire. I guess I've learnt to use
-ddump-rule the hard way...
_______________________________________________
Glasgow-haskell-bugs mailing list
[EMAIL PROTECTED]
http://www.haskell.org/mailman/listinfo/glasgow-haskell-bugs
