[Haskell-cafe] Type infer

[Marco Túlio Gontijo e Silva]

Em Qui, 2007-01-25 às 16:58 +0000, Simon Peyton-Jones escreveu:
> | -----Original Message-----
> | From: haskell-cafe-bounces at haskell.org [mailto:haskell-cafe-bounces at haskell.org] On Behalf Of Marco
> | Túlio Gontijo e Silva
> | Sent: 25 January 2007 12:57
> | To: haskell-cafe
> | Subject: Re: [Haskell-cafe] Type infer
> |
> | Em Qua, 2007-01-24 às 20:36 -0500, Bryan Donlan escreveu:
> | > Marco Túlio Gontijo e Silva wrote:
> | > > Hello,
> | > >
> | > > I'm trying to define a partition__ function that is like
> | > > Data.Set.partition, but use State Monad:
> | > >
> | > >> import Data.Set
> | > >> import Control.Monad.State
> | > >
> | > >> partition__ f =
> | > >>     do
> | > >>     snapshot <- get
> | > >>     let
> | > >>         (firsts, rest) = Set.partition f snapshot
> | > >>     put rest
> | > >>     return firsts
> | > >
> | > > When I try to infer it's type in ghci I got:
> | > >
> | > > $ ghci
> | > >    ___         ___ _
> | > >   / _ \ /\  /\/ __(_)
> | > >  / /_\// /_/ / /  | |      GHC Interactive, version 6.6, for Haskell 98.
> | > > / /_\\/ __  / /___| |      http://www.haskell.org/ghc/
> | > > \____/\/ /_/\____/|_|      Type :? for help.
> | > >
> | > > Loading package base ... linking ... done.
> | > > Prelude> :load partition.hs
> | > > [1 of 1] Compiling Main             ( partition.hs, interpreted )
> | > > Ok, modules loaded: Main.
> | > > *Main> :type partition__
> | > > partition__ :: (MonadState (Set a) t, Ord a) => (a -> Bool) -> t (Set a)
> | > >
> | > > Ok, then I add
> | > >
> | > >> partition__ :: (MonadState (Set a) t, Ord a) => (a -> Bool) -> t (Set
> | > > a)
> | > >
> | > > to the file and then:
> | > >
> | > > *Main> :reload
> | > > [1 of 1] Compiling Main             ( partition.hs, interpreted )
> | > >
> | > > partition.hs:4:0:
> | > >     Non type-variable argument in the constraint: MonadState (Set a) t
> | > >     (Use -fglasgow-exts to permit this)
> | > >     In the type signature for `partition__':
> | > >       partition__ :: (MonadState (Set a) t, Ord a) =>
> | > >                      (a -> Bool) -> t (Set a)
> | > > Failed, modules loaded: none.
> | > >
> | > > Why do I need glasgow-exts to specify a type infered by GHCi without
> | > > -fglasgow-exts?
> | >
> | > I'd imagine the check that you're using -fglasgow-exts is performed when
> | > parsing type signatures from the parser. When you allow GHC to infer the
> | > type, it's pulling that from Control.Monad.State, which was compiled
> | > with -fglasgow-exts - it's simply not checking that all the types you
> | > might infer from there are legal without -fglasgow-exts.
> |
> | Makes sense, but isn't it a bug? It shouldn't be able infer types that
> | are not allowed without -fglasgow-exts, right?

> This isn't a type-soundness bug; but it could be considered a
> user-interface sort of bug.  After all, it's caused users to be
> puzzled.  It arises really because it was convenient for the
> implementation.

> Do go ahead and file it as a Trac bug if it tripped you up enough to be worth fixing.  A fix should not take long, but might be a bit fiddly.

But if I can't write this type signature without -fglasgow-exts, I
thought that it couldn't infer this type. For me it's strange that it's
ok to have a very generic function if I don't have a type signature, but
if I write it the function will not be so generic. Shouldn't it infer
only types that could be written?

Thanks.

-- 
malebria
Marco Túlio Gontijo e Silva
Correio (MSN): malebria at riseup.net
Jabber (GTalk): malebria at jabber.org
Ekiga: malebria at ekiga.net
IRC: malebria at irc.freenode.net
     malebria at irc.oftc.org
Skype: marcotmarcot
Telefone: 33346720
Celular: 98116720
Endereço:
    Rua Paula Cândido, 257/201
    Gutierrez 30430-260
    Belo Horizonte/MG Brasil