[GHC] #14880: GHC panic: updateRole

[GHC]

#14880: GHC panic: updateRole
-------------------------------------+-------------------------------------
        Reporter:  RyanGlScott       |                Owner:  goldfire
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler (Type    |              Version:  8.2.2
  checker)                           |
      Resolution:                    |             Keywords:  TypeInType
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Compile-time      |  Unknown/Multiple
  crash or panic                     |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #15076            |  Differential Rev(s):  Phab:D4769
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by tdammers):

 Replying to [comment:59 simonpj]:
 > 4. In `unitFV` I see
 > {{{
 > unitFV :: Id -> FV
 > unitFV var fv_cand in_scope acc@(have, haveSet)
 >   | var `elemVarSet` in_scope = acc
 >   | var `elemVarSet` haveSet = acc
 >   | fv_cand var = (var:have, extendVarSet haveSet var)
 >   | otherwise = acc
 > }}}
 > Notice that ''before inserting we test for membership''.  Could that
 possibly be a win?  If so, we can use it in the `VarSet` version.

 I believe that the "test before insert" thing is needed because we're
 accumulating a set (`haveSet`) and a list (`have`) in parallel, instead of
 just the set, so in order to keep the two in sync, we cannot simply insert
 unconditionally, as that would produce duplicates in the list.

 I don't think it can possibly produce a performance advantage beyond
 skipping the list cons, *unless* we are in a special situation where:

 - `in_scope` is much smaller than `haveSet` (such that testing membership
 of `in_scope` is significantly faster than testing `haveSet` membership);
 AND:
 - `in_scope` is likely to match new entries (because when it doesn't
 match, the extra test just eats up clock cycles without gaining anything)

 One interesting thing might be to invert the `fv_cand` condition and
 change the ordering, so that the `fv_cand` test is done before the set
 membership tests, something like:

 {{{#!haskell
 unitFV :: Id -> FV
 unitFV var fv_cand in_scope acc@(have, haveSet)
   | not (fv_cand var) = acc
   | var `elemVarSet` in_scope = acc
   | var `elemVarSet` haveSet = acc
   | otherwise = (var:have, extendVarSet haveSet var)
 }}}

 But that won't change the performance of `VarSet` the slightest.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14880#comment:63>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler