Re: [GHC] #14498: GHC internal error: "not in scope during TC but it passed the renamer"

30 Nov 2017

      #14498: GHC internal error: "not in scope during TC but it passed the renamer"
-------------------------------------+-------------------------------------
        Reporter:  Iceland_jack      |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler (Type    |              Version:  8.2.1
  checker)                           |             Keywords:
      Resolution:                    |  PatternSynonyms
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Compile-time      |  Unknown/Multiple
  crash or panic                     |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #14288            |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by RyanGlScott):

 To be clear here: I'm not proposing to change anything regarding the rules
 for ordinary functions (I'll save that for #14288, which will require a
 GHC proposal). I'm only proposing to change the rules for pattern
 synonyms, since (1) pattern synonyms are a more experimental feature, and
 (2) the current rules governing them are clearly broken, as this ticket
 demonstrates.

 Replying to [comment:8 simonpj]:
...
1.  Where are the implicitly-added foralls?  Example
{{{
pattern SS ::
     forall (t :: k').
     ()
  => forall (a :: kk -> k') (n :: kk).
     (t ~ a n)
  => blah
}}}
  This is short for
{{{
pattern SS ::
     forall {k'}. forall (t :: k').
     ()
  => forall {kk}. forall (a :: kk -> k') (n :: kk).
     (t ~ a n)
  => blah
}}}
Ideally, this would be the case. The typechecker thinks this way, but
 unfortunately, the renamer does not. It thinks it's this:

 {{{#!hs
 pattern SS ::
      forall {k' kk}. forall (t :: k').
      ()
   => forall (a :: kk -> k') (n :: kk).
      (t ~ a n)
   => blah
 }}}

 That is, the renamer believes everything implicitly quantified is put up
 front, as as a result, tries to bring `kk` into scope over the body of
 `SS`, leading to the internal error seen above.

 I propose to simply not bring `kk` into scope in the renamer. That's all.
 (I tried implementing this last week, but this turns out to be quite
 tricky, so I gave up.)

-- 
Ticket URL: http://ghc.haskell.org/trac/ghc/ticket/14498#comment:9
GHC http://www.haskell.org/ghc/
The Glasgow Haskell Compiler