[GHC] #11970: Simplify Parent for patten synonyms

[GHC]

#11970: Simplify Parent for patten synonyms
-------------------------------------+-------------------------------------
        Reporter:  simonpj           |                Owner:  mpickering
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  7.10.3
      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 mpickering):

 I have looked a bit at this. The problem from last time is that when
 renaming exports we need to reject data constructors if they are exported
 with a type constructor which is not their parent.

 This is how the code was structured before my changes.

 1. `tcg_rdr_env` is converted into a `kids_env` which is a map from
 `NameEnv [GlobalRdrElt]`. The intention of this map that we can lookup a
 type constructor and find which data constructors are allowed to be
 bundled with it.
 2. `lookupChildren` uses this map to check that all bundled constructors
 are with the correct type constructor.

 The problem which I ran into was that as we only have `GlobalRdrElt`s, at
 this stage I couldn't see a way of telling whether a certain
 `GlobalRdrElt` arose from a data constructor or a pattern synonym. It
 seemed to me that the sole purpose of the `Parent` datatype was to
 indicate when it should be allowed to export an identifier with a certain
 type constructor (its parent). Because of this, I made the necessary
 changes to `Parent` to indicate whether a `GRE` arose from a pattern
 synonyn.

 Do you see a better way of structuring this? Unless it is possible to
 distinguish between a pattern synonym and a data constructor at this point
 then it isn't safe to ignore an entry in the export list as it might be a
 data constructor which we are not allowed to export with that type
 constructor.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11970#comment:3>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler