[GHC] #8177: Roles for type families

[GHC]

#8177: Roles for type families
-------------------------------------+------------------------------------
        Reporter:  simonpj           |            Owner:  goldfire
            Type:  bug               |           Status:  new
        Priority:  normal            |        Milestone:
       Component:  Compiler          |          Version:  7.6.3
      Resolution:                    |         Keywords:
Operating System:  Unknown/Multiple  |     Architecture:  Unknown/Multiple
 Type of failure:  None/Unknown      |       Difficulty:  Unknown
       Test Case:                    |       Blocked By:
        Blocking:                    |  Related Tickets:
-------------------------------------+------------------------------------
Changes (by goldfire):

 * owner:   => goldfire


Comment:

 Q1: Yes, that's possible. I didn't do it mostly because my stomach lurched
 when thinking about parsing role annotation concrete syntax for data
 family instances -- I was going to wait until someone shouted. I
 considered having data instances go through role inference without
 allowing for role annotations, but that meant it would be impossible to
 create a properly abstract data family instance.

 Q2: That's possible but currently unimplemented (see earlier posts to this
 bug report).

 So, if we could just figure out a decent concrete syntax, I could make it
 all work. The hardest part will be updating the parser and `HsSyn` stuff.
 The role inference/checking should be almost no work at all.

 As a strawman syntax, what about

 {{{
 type role instance T [nominal]
 }}}

 for the case above. Note the word `instance` to specify that we're talking
 about the roles on an instance, not on the family `T`. Then, role names
 would appear wherever type variables appear in the corresponding instance
 declaration. Underscores would be supported in these positions, as well
 (like normal role annotations). The corresponding data (or newtype)
 instance would have to be in the same module, as usual. The appearance of
 the word `type` (as opposed to `data`) in the role annotation syntax is
 regrettable, but it conforms to the universal use of the word `type` in
 all role annotations.

 Though it makes me want to cry, I suppose we should allow ''associated''
 role instance annotations, in class instance definitions alongside
 associated data/newtype instances.

 Does anyone out there actually want to use this feature? Eager users would
 help motivate all this! :)

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/8177#comment:5>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler