[GHC] #11735: Optimize coercionKind

[GHC]

#11735: Optimize coercionKind
-------------------------------------+-------------------------------------
        Reporter:  goldfire          |                Owner:  (none)
            Type:  task              |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  7.10.3
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Compile-time      |  Unknown/Multiple
  performance bug                    |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:                    |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by simonpj):

 There are a number of different things going on.

 1.  comment:5 gets rid of a non-linearity in `coercionKind` for the
 `ForallCo` case.  This is a big win.  We definitely want it.

 2. The same non-linearity is present in `coercionKindRole` so I suggested
 getting rid of the duplication between the two.  That seems possible by
 caching the role in `NthCo`.  Doing so doesn't solve any known performance
 problems, but seems to be nicer code.

 3. We have a clear problem in `simplCast`, which comes from transforming
 {{{
 (f |> co) @t1 @t2 ... @tn
 ===>
 ((f @t1') |> co1) @t2 ... @tn
 ===> (etc)
 (f @t1' @t2' ... @tn') |> con
 }}}
   Doing these steps one at a time gives clearly non-linear behaviour.
 Hence comment:10.  I think comment:41 is in this territory.

 I think it'd be helpful to separate these issues, possibly into separate
 tickets.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/11735#comment:48>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler