[GHC] #11011: Add type-indexed type representations (`TypeRep a`)

[GHC]

#11011: Add type-indexed type representations (`TypeRep a`)
-------------------------------------+-------------------------------------
        Reporter:  bjmprice          |                Owner:
            Type:  feature request   |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:
      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 bgamari):

 > > These representations are handled as a bit of a special case. I need
 to
 > [sic] Ooh. It's like a mystery novel.

 :-) That was meant to say that I needed to take another careful look at
 the implementation surrounding the representations primitive type since it
 seems a bit fragile at the moment.

 > Other points:
 > - I vote only pattern synonyms.

 That sounds reasonable to me.

 > - Serialization is used in Cloud Haskell, no?

 Yes; it's also used in GHC (there is a `Binary` instance). I'm a bit
 unsure at the moment of what this should look like.

 > - I like the idea of including the paranoid checks as an option for
 paranoid users. Although perhaps not, given that I believe it's likelier
 for a cosmic ray to flip a bit than for the fingerprints to give a false
 positive.

 I'm not opposed to providing it, although it would be good to state
 clearly what our threat model is here. What guarantees are we trying to
 keep? This question seems to become rather hairy in the presence of
 serialization.

 > As for names: I think a proper migration story needs to be articulated.
 In a few years, I'd hate to have all these extra Ts lying around. Use a
 new module name for the new features and keep the old module with the old
 interface around? Maybe the new module can be Data.Reflection.

 I'm not sure the `T`s are necessarily bad given that the quantified
 `TypeRep` still serves a purpose. Especially in light of my proposal in
 comment:18 it seems like the `T` prefix could be a useful convention.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11011#comment:20>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler