[GHC] #7258: Compiling DynFlags is jolly slow

[GHC]

#7258: Compiling DynFlags is jolly slow
-------------------------------------+-------------------------------------
        Reporter:  simonpj           |                Owner:  simonpj
            Type:  bug               |               Status:  new
        Priority:  normal            |            Milestone:
       Component:  Compiler          |              Version:  7.6.1
      Resolution:                    |             Keywords:  deriving-perf
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 tdammers):

 Here's some tentative evidence:

 [[Image(https://ghc.haskell.org/trac/ghc/attachment/ticket/7258/stgVarsSizeRatio.png)]]

 The 3 misbehaving examples (read, show, and to a lesser degree read-appl)
 show linear growth of the free-variables/other-stg ratio (which suggests
 roughly quadratic growth of the size of the overall STG size, assuming
 linear growth of the base STG size), while the well-behaved examples
 maintain a more-or-less constant ratio.

 The full profiling job (up to 400 fields) is still running, but this small
 sample (50, 100 and 150 fields) looks pretty convincing to me already.

 A quick glance at the generated STG suggests that there is indeed some
 deep nesting of closures going on.

-- 
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/7258#comment:85>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler