[GHC] #4462: -dcore-lint error in simplifier phase 0 when profiling

[GHC]

#4462: -dcore-lint error in simplifier phase 0 when profiling
-------------------------------------------------+--------------------------
    Reporter:  wkahl                             |        Owner:                       
        Type:  bug                               |       Status:  new                  
    Priority:  normal                            |    Milestone:                       
   Component:  Compiler                          |      Version:  7.0.1 RC1            
    Keywords:  profiling, simplifier, core-lint  |     Testcase:  Agda (darcs, current)
   Blockedby:                                    |   Difficulty:                       
          Os:  Linux                             |     Blocking:                       
Architecture:  x86_64 (amd64)                    |      Failure:  Compile-time crash   
-------------------------------------------------+--------------------------

Comment(by wkahl):

 Actually,
 {{{
 dist/build/Agda/Syntax/Parser/Parser.hs
 }}}
 is generated by Happy, and Cabal invokes Happy, without asking, and
 without opportunity to override, with {{{-agc}}}.
 I have manually regenerated
 {{{
 dist/build/Agda/Syntax/Parser/Parser.hs
 }}}
 with {{{-ag}}}, and see, as expected, the same segfaults (and the same
 core-lint error).

 I have the following times to segfault:

 Common RTS flags:
 {{{
 +RTS     -C0 -i0 -K64M -M12G -H12G -S
 }}}

 {{{
 Binary      Additional flags   Time       Alloc Sum
 agda                           17min     870,907,904,328
 agda_debug                     27min     870,907,904,288
 agda_prof  +RTS      -xc       49min   1,138,794,235,424
 agda_debug +RTS -DS           656min     870,907,945,760
 }}}

 {{{-xc}}} unfortunately gives no cost-center stack for the segfault. But
 there are 20 exceptions in these runs, generating 403kBytes of additional
 output.

 The allocation sums are calculated by adding up the first column of the
 {{{+RTS -S}}} output, and are all constant across several runs. (The
 {{{-DS}}} runs were with the old parser; the two fast runs both allocated
 870,907,945,760 bytes with the old parser.)



 Agda links with Haskeline, but that is not used in my kind of invocation.
 The only other potentially suspicious part seems to be

 {{{
   build-depends:
                     syb == 0.1.*
 }}}

 for which I have {{{syb-0.1.0.3}}}. Do you think that syb is worth
 investigating for this?

 Do you have any other ideas what I could do to locate this segfault?


 Wolfram

-- 
Ticket URL: <http://hackage.haskell.org/trac/ghc/ticket/4462#comment:6>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler