[GHC] #14630: name shadowing warnings by record pattern synonyms + RecordWildCards or NamedFieldPuns

[GHC]

#14630: name shadowing warnings by record pattern synonyms + RecordWildCards or
NamedFieldPuns
-------------------------------------+-------------------------------------
        Reporter:  mizunashi_mana    |                Owner:  (none)
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  8.2.2
      Resolution:                    |             Keywords:
                                     |  PatternSynonyms
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Incorrect         |  Unknown/Multiple
  error/warning at compile-time      |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #11228, #11970    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by adamgundry):

 In the past (prior to Phab:D761), the `RecFieldEnv` stored the set of
 record field names defined in the current module, and `is_shadowed_gre`
 used this set for local identifiers and `isRecordSelector` for imported
 identifiers. We could go back to doing something similar.

 However, in order to implement `DuplicateRecordFields`, it's helpful for a
 GRE to simply know whether or not it is a record field (and the field
 label, if it is), because the renamer ends up special-casing the treatment
 of multiple in-scope GREs when they are all fields. I think we should keep
 doing this, and moreover make it correct for record pattern synonym
 fields, as a step towards #11228. That is, I think we should do something
 like Ryan's patch. The price we end up paying is a slightly more
 complicated definition of `AvailInfo`, but I don't think that's too
 terrible, is it?

 > I would imagine two distinct data types, one which tracks parenthood and
 and one which tracks whether a GRE is a record field.

 Yes, I think this would be worth doing instead of a new constructor in
 `Parent`. In the presence of record pattern synonyms, these two properties
 are independent, and they could simply be two separate fields of
 `GlobalRdrElt`.

 Unfortunately I doubt we can really make things fundamentally simpler
 without doing away with `DuplicateRecordFields` altogether; the
 simplifications I'm considering will move work from the typechecker to the
 renamer, but the issues here are all about the renamer.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14630#comment:13>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler