[GHC] #12387: Template Haskell ignores class instance definitions with methods that don't belong to the class

[GHC]

#12387: Template Haskell ignores class instance definitions with methods that don't
belong to the class
-------------------------------------+-------------------------------------
        Reporter:  RyanGlScott       |                Owner:
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Template Haskell  |              Version:  8.0.1
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  GHC accepts       |  Unknown/Multiple
  invalid program                    |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by simonpj):

 I made a little progress with this, which I'll record.

 The bug is here in `RnEnv.lookupSubBndrOcc`.  Usually this function looks
 up an unqualified `RdrName` and uses the parent info in the `GlobalRdrEnv`
 to disambiguate; and to check that the specified name is indeed a method
 of the parent class.

 But here are in the `Exact` `RdrName` case:
 {{{
 lookupSubBndrOcc warn_if_deprec the_parent doc rdr_name
   | Just n <- isExact_maybe rdr_name   -- This happens in derived code
   = do { n <- lookupExactOcc n
        ; return (Right n) }
 }}}
 In the example `'compare` refers precisely to `GHC.Classes.compare`, so we
 get an `Exact` `RdrName`.  But that `compare` may not even be in scope,
 and may not be in the `GlobalRdrEnv`.  We are omitting a check for the
 correct parent; hence the lack of error message.

 What to do?  Either

 * If it's not in scope it must presumably be imported and hence in the
 `TypeEnv`.  So we could look for parent-hood that way.  A bit of a pain.

 * Or we could postpone the check altogether to the typechecker.  I like
 this plan better because it works uniformly for local and imported things.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/12387#comment:1>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler