[GHC] #7594: GHCi becomes confused about IO type

[GHC]

#7594: GHCi becomes confused about IO type
-------------------------------+--------------------------------------------
  Reporter:  Khudyakov         |          Owner:                  
      Type:  bug               |         Status:  closed          
  Priority:  normal            |      Milestone:                  
 Component:  GHCi              |        Version:  7.6.1           
Resolution:  worksforme        |       Keywords:                  
        Os:  Unknown/Multiple  |   Architecture:  Unknown/Multiple
   Failure:  None/Unknown      |     Difficulty:  Unknown         
  Testcase:                    |      Blockedby:                  
  Blocking:                    |        Related:                  
-------------------------------+--------------------------------------------
Changes (by simonpj):

  * status:  new => closed
  * difficulty:  => Unknown
  * resolution:  => worksforme


Comment:

 Hmm.  Absolutely right, and this happens for 7.6.2 too. However with HEAD
 we get:
 {{{
 <interactive>:2:5:
     Couldn't match type `b' with `IO ()'
       `b' is untouchable
         inside the constraints ((:&:) Show Real a)
         bound by a type expected by the context:
                    (:&:) Show Real a => a -> b
         at <interactive>:2:1-11
       `b' is a rigid type variable bound by
           the inferred type of it :: b at <interactive>:2:1
 }}}
 This is actually right, although the error message is not so easy.  We
 know that
 {{{
 print :: Show a => a -> IO ()
 }}}
 We also know (by way of the argument `q`, that `c = (Show :&: Real) a`.
 So, to check that `print` is a valid argument to `app`, we must check that
 {{{
 From          (Show :&: Real) a
 prove that    (Show a, beta ~ IO ())
 }}}
 where `beta` is a unification variable (as-yet-unknown type) obtained by
 instantiating `app`.

 Well, you say, we can choose `beta` to be `IO ()` and we are done.  But
 not so: in general the "`From`" constraints might include GADT-style
 equalites, and GHC is careful NOT to unify under equality constraints.
 (See the [http://haskell.org/haskellwiki/Simonpj/Talk:OutsideIn Modular
 type inference with local assumptions] paper.)  That's why `beta` is
 "untouchable".

 Easily fixed by saying `app print q :: IO ()`, and that works fine.

 I am strongly dis-inclined to attempt to find out what's happening in 7.6!
 So I propose just to close this.  I might add a regression test though.

 Simon

-- 
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/7594#comment:1>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler