Re: [GHC] #10340: Type inference regression with Any

29 Apr 2015

      #10340: Type inference regression with Any
-------------------------------------+-------------------------------------
        Reporter:  akio              |                   Owner:
            Type:  bug               |                  Status:  new
        Priority:  normal            |               Milestone:
       Component:  Compiler (Type    |                 Version:  7.10.1
  checker)                           |                Keywords:
      Resolution:                    |            Architecture:
Operating System:  Unknown/Multiple  |  Unknown/Multiple
 Type of failure:  GHC rejects       |               Test Case:
  valid program                      |                Blocking:
      Blocked By:                    |  Differential Revisions:
 Related Tickets:                    |
-------------------------------------+-------------------------------------

Comment (by goldfire):

 I just took a look, and the problem here is fairly simple. I'm not sure
 what the right solution is, though.

 Here are the constraints that get to the solver:
 {{{
         [W] $dMonadState_a1UB :: MonadState
                                    s_a1Uy[tau:1] m_a1Uz[tau:1]
 (CNonCanonical)
         [W] cobox_a1UD :: m_a1Uz[tau:1] ~ State Any (CNonCanonical)
         [W] cobox_a1UE :: s_a1Uy[tau:1] ~ Any (CNonCanonical)
 }}}

 In short order, we get a fuv for `Any`

 {{{
 [W] cobox_a1UT :: Any ~ s_a1US[fuv:0] (CFunEqCan)
 }}}

 and a unification for `m_a1Uz`.

 {{{
 setWantedTyBind m_a1Uz[tau:1] := State s_a1US[fuv:0]
 }}}

 The solver then attacks `cobox_a1UE`, which canonicalizes to

 {{{
 [W] cobox_a1UW :: s_a1US[fuv:0] ~ s_a1Uy[tau:1] (CTyEqCan)
 }}}

 Because an fuv is on the left, no unification takes place. See `Note
 [Orient equalities with flatten-meta-vars on the left]` in !TcFlatten.

 When we then canonicalize the `MonadState` constraint, we get

 {{{
 [W] $dMonadState_a1UB :: MonadState s_a1Uy[tau:1] (State s_a1US[fuv:0])
 (CDictCan)
 }}}

 which can't get solved and is reported to the user.

 The "obvious" solution to me is to re-orient the `CTyEqCan` above
 (`cobox_a1UW`) so that unification gets the job done for us. But that
 `Note [Orient equalities with flatten-meta-vars on the left]` in
 !TcFlatten tells me not to. Another solution is to allow a Wanted to
 rewrite a Wanted here, but we've previously been convinced that this is a
 bad idea.

 Incidentally, I don't think this problem is at all restricted to `Any`. I
 haven't tested any other code, but I imagine this will bite whenever we
 want an instance with repeated pattern variables to match a type with
 irreducible type families.

--
Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/10340#comment:1
GHC http://www.haskell.org/ghc/
The Glasgow Haskell Compiler