[GHC] #9805: Use TrieMaps to speed up type class instance lookup

[GHC]

#9805: Use TrieMaps to speed up type class instance lookup
-------------------------------------+-------------------------------------
        Reporter:  ezyang            |                   Owner:  ezyang
            Type:  task              |                  Status:  new
        Priority:  normal            |               Milestone:
       Component:  Compiler (Type    |                 Version:  7.9
  checker)                           |                Keywords:
      Resolution:                    |            Architecture:
Operating System:  Unknown/Multiple  |  Unknown/Multiple
 Type of failure:  None/Unknown      |               Test Case:
      Blocked By:                    |                Blocking:
 Related Tickets:                    |  Differential Revisions:
-------------------------------------+-------------------------------------

Comment (by simonpj):

 I've taken a look at your straw man.  I really don't understand it.

 What is `TyBox`??

 Standing back, I'm sure that the template variables should be deBruijn-
 ised.  And a `VarMap` should look like
 {{{
 data VarMap a = VM { vm_bvar   :: BoundVarMap a  -- Bound variable
                    , vm_fvar   :: VarEnv a       -- Free variable
                    , vm_tvar   :: TmplVarMap a } -- Template variable

 type TmplVar = Int  -- Like BoundVar but a different namespace
 type TmplVarMap = IntMap
 }}}
 That is, in a key, a variable could be bound by a forall (`vm_bvar`), or
 free (`bm_fvar`), or a template variable (`vm_tvar`).
 So a key `(T (forall a. a -> b -> c))`, where b is a template variable and
 c is free will effectively be treated as
 {{{
   T (forall. bv1 -> tv1 -> c)
 }}}
 where I used `bv1` for `BoundVar` 1 and `tv1` for `TmplVar` 1.

 When doing a `matchT` we must maintain a
  * `CmEnv`, for deBruijn of the foralls
  * `TmplVarMap (CmEnv,Type)` for the substitution so far, mapping template
 variables to types.  Well, actually to a kind of lexical closure of
 `(CmEnv, Type)` (c.f. #9960).

 When matching, if the key is say (tv1,tv1), then the second time we
 encounter
 `tv1` it'll already be bound in the substitution, so we must check for
 equality with the equality function mentioned in #9960.

 I don't think an occurs-check can happen when ''matching''.

 Does that make sense?  Happy to talk more, or Skype.

 Simon

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9805#comment:5>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler