[GHC] #9370: unfolding info as seen when building a module depends on flags in a previously-compiled module

[GHC]

#9370: unfolding info as seen when building a module depends on flags in a
previously-compiled module
-------------------------------------+-------------------------------------
        Reporter:  carter            |                Owner:  richardfung
            Type:  bug               |               Status:  new
        Priority:  high              |            Milestone:  8.2.1
       Component:  Compiler          |              Version:  7.8.3
      Resolution:                    |             Keywords:  newcomer,
                                     |  Inlining
Operating System:  Unknown/Multiple  |         Architecture:
 Type of failure:  Compile-time      |  Unknown/Multiple
  performance bug                    |            Test Case:
      Blocked By:                    |             Blocking:
 Related Tickets:  #8635             |  Differential Rev(s):
       Wiki Page:                    |
-------------------------------------+-------------------------------------

Comment (by richardfung):

 Ah yes sorry I shall try to keep my comments more self contained.

 The ticket addresses the issue that the unfolding info seen depends on
 previously compiled modules with --make. As I understand it, this can
 either cause ghc to inline when it shouldn't (in -O0) or to be unable to
 inline when it should.

 For example, if we have modules A and B, both which import module C, and
 we assume A is built first, then whether B sees the interface pragmas of C
 is dependent on whether they were read in when compiling A. If A is built
 with -O0, then even if B is built with -O it can not inline things from C
 because it doesn't have the unfolding info. Alternatively, if A is built
 with -O and B is built with -O0, B will still see the unfolding info and
 use it.

 The suggested fix is to disable -fignore-interface-pragmas being implied
 by -O0 and to make sm_inline = False when -O0.

 I do believe this solves both of the scenarios described above, but I'm
 concerned that if A is built with -fignore-interface-pragmas, and B is
 built with -O, B will still not have the unfolding info from C when
 compiling. This would be identical to what's currently happening when A is
 built with -O0 and B with -O.

--
Ticket URL: <http://ghc.haskell.org/trac/ghc/ticket/9370#comment:38>
GHC <http://www.haskell.org/ghc/>
The Glasgow Haskell Compiler