[GHC] #9315: Weird change in allocation numbers of T9203

[GHC]

#9315: Weird change in allocation numbers of T9203
-------------------------------------+-------------------------------------
              Reporter:  nomeata     |             Owner:  simonmar
                  Type:  bug         |            Status:  new
              Priority:  normal      |         Milestone:
             Component:  Runtime     |           Version:  7.9
  System                             |          Keywords:
            Resolution:              |  Operating System:  Unknown/Multiple
Differential Revisions:              |   Type of failure:  None/Unknown
          Architecture:              |         Test Case:
  Unknown/Multiple                   |          Blocking:
            Difficulty:  Unknown     |
            Blocked By:              |
       Related Tickets:              |
-------------------------------------+-------------------------------------

Comment (by simonmar):

 Replying to [comment:11 nomeata]:
 > >  The perf tests don't guarantee to pass unless you use the validate
 settings, I think that's unavoidable.
 >
 > True; but it seems that we would want to validate the settings used for
 release, aren’t we? Otherwise there might be a regression affecting our
 releases, and we won’t notice....

 There are *lots* of things that validate doesn't test for, and lots of
 regressions that it won't catch.  It's a compromise between coverage and
 speed, deliberately.  For instance, it doesn't test profiling at all - yet
 that's an important feature of our releases.  We need to make the full
 testsuite runs more visible to developers, and focus on fixing regressions
 that they find.

 I suspect that adding -O2 to libraries and compiler (we would have to do
 both) would make validate slower, but I don't know how much; feel free to
 measure it.  I do think validate is too slow already, though, and we
 should be finding ways to speed it up.

 The full testsuite run should use -O2 (I think I used to do that when I
 did the nightly builds), and the perf tests should have a separate set of
 figures for when libraries are built with -O2 where necessary.  Probably
 only this one test will need to have different results.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9315#comment:12>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler