[GHC] #14880: GHC panic: updateRole

[GHC]

#14880: GHC panic: updateRole
-------------------------------------+-------------------------------------
        Reporter:  RyanGlScott       |                Owner:  goldfire
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:  8.8.1
       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):

 Test run results:

 - `wip/T14880-baseline`: pre-patch GHC HEAD.

   `("bytes allocated", "1144547520")`

 - `wip/T14880`: patch fully applied.

   `("bytes allocated", "1207856632")`

 - `wip/T14880-just-tvs`: only implement the "use VarSets, not FV" part of
 #14880

   `("bytes allocated", "1176403280")`

 - `wip/T14880-nondet-fv`: baseline (same as `wip/T14880-baseline`), but
 using
   the `NDFV` replacement instead of `FV` (getting rid of the list that
 `FV`
   drags around to maintain insertion order).

   `("bytes allocated", "1144547040")`

 - `wip/T14880-nondet-no-in-scope`: same as `wip/T14880-nondet-fv`, but
 instead
   of adding to the `in_scope` set, remove items from the accumulator.

   `("bytes allocated", "4064821552")`

 - `wip/T14880-nondet-fv-2`: same as `wip/T14880-nondet-fv`, but with the
 `NDFV`
   type rewritten to update sets in place rather than building a
 computation
   accumulator-style.

   `("bytes allocated", "1222408136")`

 So it seems that somehow my results got mixed up somewhere, because there
 is a whopping 400% increase in memory use when removing the in-scope
 feature. There must be something else going on there though, because
 neither the fully applied patch nor the `just-tvs` version deviate this
 severely, despite also not using this in-scope pre-filtering.

 Disregarding the `nondet-no-in-scope` figure, we thus get:

 - applying full patch: 1145M -> 1208M
 - only introducing VarSet: 1145M -> 1176M (worse than baseline, better
 than full patch)
 - making FV nondeterministic: 1145M -> 1145M (no effect)
 - rewriting (ND)FV to not use accumulator style: 1145M -> 1222M (worse
 than full patch)

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/14880#comment:93>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler