Re: Fundeps and quantified constructors
--- Tom Pledger [EMAIL PROTECTED] wrote: anatoli writes: : | The same error message is given for | | data Foo a = (Eq b) = MkFoo b Since the type variable a is orphaned, how about reducing it to this? data Foo = forall b . Eq b = MkFoo b This is possible (the semantics is different of course). What I want to do is, given a multi-parameter context, orphan one type variable leaving the other intact, and that's not possible now. I can only orphan all or preserve all. so that MkFoo :: Eq b = b - Foo | I don't know whether this is intended behaviour; IMHO | it should be treated identically to | | data Foo a = MkFoo (Eq a = a) Isn't the Foo-ed a hidden by the Eq-ed a in this example too? i.e. data Foo = MkFoo (forall a . Eq a = a) It appears that both examples are wrong, or at least Hugs thinks so. The first MkFoo happily accepts any type, not just of class Eq: Main :t MkFoo head MkFoo head :: Foo ([a] - a) The second accepts nothing of interest: Main :t MkFoo 'a' ERROR: Inferred type is not general enough *** Expression: 'a' *** Expected type : Eq a = a *** Inferred type : Eq Char = Char Hm... The more I experiment with this, the less I understand. It seems that the only correct place to put the quantifier is before the data constructor, and the context may go either before the type constructor or after the quantifier. This is very unfortunate. How can I correctly write something to this effect: data Quick a = Eq a = Unsorted [a] | Ord a = Sorted (Tree a) ??? Regards -- anatoli __ Do You Yahoo!? Get personalized email addresses from Yahoo! Mail - only $35 a year! http://personal.mail.yahoo.com/ ___ Haskell mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell
Re: Fundeps and quantified constructors
Hi everybody: I think I've found what's the problem. Still no solution in sight :( The problem has nothing to do with fundeps. Consider an example: data Foo a = (Eq a) = MkFoo a This gives the same error message: type variable a is not locally bound. Apparently, 'a' in 'Eq a' hides 'a' in 'Foo a' instead of using it. The same error message is given for data Foo a = (Eq b) = MkFoo b I don't know whether this is intended behaviour; IMHO it should be treated identically to data Foo a = MkFoo (Eq a = a) Hugs accepts either data Eq a = Foo a = MkFoo a or data Eq a = MkFoo (Eq a = a) with no problem, but neither syntax works with Collection: the first one would look like data forall c. Collection c e = AnyColl e = MkColl c which is a syntax error, and the second one data AnyColl e = forall c . MkCall (Collection c e = c) or alternatively data AnyColl e = MkCall (forall c . Collection c e = c) gives me 'Ambiguous type signature in type component' error. { Apparently Hugs is still buggy in the fundep/multiparameter classes area; I have an example where Hugs assigns garbage types to things when fundeps are involved. This is probably a hugs-bugs issue. } Now to the question: --- Tom Pledger [EMAIL PROTECTED] wrote: Does it introduce any new implementation issues? When a constructor like MakeSomeCollection is applied, some implementations gather a dictionary of methods appropriate for use with the existentially quantified type, right? I believe that in the general case all implementation must do so, in one form or another. Is that more difficult when it involves a non-locally bound type variable? Even when that type variable is constrained by a fundep? I believe there should be no implementation problems. In order to gather the dictionary, an implementation must know the type of the argument to MakeSomeCollection. When this type (say T) is known, the compiler just finds 'instance Collection T whatever' declaration. There can be only one such instance (because of the fundep). All needed methods are specified there. Regards -- anatoli __ Do You Yahoo!? Yahoo! Auctions - Buy the things you want at great prices. http://auctions.yahoo.com/ ___ Haskell mailing list [EMAIL PROTECTED] http://www.haskell.org/mailman/listinfo/haskell