[GHC] #16114: strange "instance .. => .. => .. where ..."

[GHC]

#16114: strange "instance .. => .. => .. where ..."
-------------------------------------+-------------------------------------
        Reporter:  j.waldmann        |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.6.3
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
                                     |  Unknown/Multiple
 Type of failure:  None/Unknown      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by simonpj):

 > Another option would be to re-investigate checkValidInstance

 I don't think so.  What I missed before is that ''the renamer needs to
 know'' which class is involved so it has the right "parent" for lookups.
 You can see this in the call to `getLHsInstDeclClass_maybe` in
 `RnNames.getLocalNonValBinders`.  And it is also the cause of this ticket,
 because the instance got past the renamer only by remaming `(==)` not to
 the `(==)` from the `Eq` class, but to a new unrelated top-level binder.
 Yikes.

 I think putting up with marginally less-cool errors may be a price we have
 to pay.

 Rather than "mal-formed" we could say
 {{{
 Class instances must be of the form
   context => C ty1 .. tyn
 where C is a class
 }}}
 or something more positive than "mal-formed".

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/16114#comment:5>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler