[GHC] #10845: Incorrect behavior when let binding implicit CallStack object

[GHC]

#10845: Incorrect behavior when let binding implicit CallStack object
-------------------------------------+-------------------------------------
        Reporter:  nitromaster101    |                Owner:  gridaphobe
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  7.11
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #10846            |  Differential Rev(s):
-------------------------------------+-------------------------------------

Comment (by gridaphobe):

 Ah, I wasn't suggesting to treat `getCallStack` specially, it was just the
 simplest way to get GHC to instantiate the ?x's type with `CallStack` in
 the RHS of the local let-binder, which is a key ingredient in this bug.

 Suppressing the special case for `CallStack` seems to get us part of the
 way, we defer solving the CallStack until we get a toplevel implication

 {{{
   Implic {
     TcLevel = 3
     Skolems =
     No-eqs = False
     Status = Unsolved
     Given = $dIP_anO :: ?loc::CallStack
     Wanted =
       WC {wc_impl =
             Implic {
               TcLevel = 5
               Skolems =
               No-eqs = False
               Status = Unsolved
               Given =
               Wanted =
                 WC {wc_simple = [W] $dIP_anT :: ?loc::CallStack
 (CNonCanonical)}
               Binds = EvBindsVar<anU>
               the inferred type of y_anr :: t_anQ[tau:5] }}
     Binds = EvBindsVar<anY>
     the type signature for:
       f :: (?loc::CallStack) => [(String, SrcLoc)] }
 }}}

 Unfortunately the wanted in this implication is itself an implication, so
 GHC raises the TcLevel again when it enters the nested implication and our
 special case remains suppressed.

 I find it a bit strange though that the nested implication has no givens.
 Why emit an implication with no givens when we could just emit a simple
 wanted?

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/10845#comment:9>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler