[GHC] #13253: Exponential compilation time with RWST & ReaderT stack with `-02`

[GHC]

#13253: Exponential compilation time with RWST & ReaderT stack with `-02`
-------------------------------------+-------------------------------------
        Reporter:  phadej            |                Owner:  bgamari, osa1
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:  8.8.1
       Component:  Compiler          |              Version:  8.0.1
      Resolution:                    |             Keywords:
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Compile-time      |  Unknown/Multiple
  performance bug                    |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #15630            |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by tdammers):

 I went wild on the debug logging when compiling bad.hs, with GHC HEAD and
 the patch from comment:24 in place. One thing that sticks out is that at
 some point, total Core size goes from about 7,700 to 685,000:

 {{{
 Result size of Simplifier iteration=4
   = {terms: 7,759, types: 9,295, coercions: 847, joins: 9/84}
 Result size of Simplifier
   = {terms: 7,759, types: 9,295, coercions: 847, joins: 9/84}
 !!! Simplifier [Main]: finished in 1123.65 milliseconds, allocated
 1251.533 megabytes
 *** SpecConstr [Main]:
 Result size of SpecConstr
   = {terms: 685,052,
      types: 613,224,
      coercions: 22,309,
      joins: 1,571/11,551}
 }}}

 Subsequent simplifier runs bring this down to 70,000 eventually, and then
 it jumps up to 96,000 again at the CorePrep stage, but it never goes back
 down to anywhere near those 7,000.

 If, however we enable the `{-#NOINLINE#-}` pragma on `mhelper`, then that
 same SpecConstr step maintains Core size exactly (suggesting that it
 doesn't do anything at all).

 `mhelper` is ` ::  MonadIO m => ...`; if we inline it, then `m` can be
 specialized to `Handler`, which is a synonym for `ReaderT () IO`. I don't
 know why this particular specialization blows up right there and then, but
 that's what I've figured so far.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/13253#comment:37>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler