Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

Subject: Re: [OMPI devel] [OMPI svn] svn:open-mpi r23767
From: Ralph Castain (rhc_at_[hidden])
Date: 2010-09-18 10:54:00


On Sep 18, 2010, at 8:46 AM, Ralf Wildenhues wrote:

> Hi Jeff,
>
> * Jeff Squyres wrote on Sat, Sep 18, 2010 at 01:39:44PM CEST:
>> On Sep 18, 2010, at 10:03 AM, Jeff Squyres wrote:
>>
>>> So it's still on my list to go back and figure out what changed
>>> (i.e., it didn't complain about valgrind's version before this -- so
>>> something must have changed. We don't use --without-valgrind when
>>> configuring, so I'm not sure why it somehow became a fatal error).
>>> :-)
>>
>> I figured out what changed -- it was something we did in ompi_mca.m4.
>> So I sent Brian a patch (the m4 is a little dicey -- I don't want to
>> make a change without checking with him). However, he's out on
>> vacation, so it may be a little bit before I get an answer.
>
> You /could/ send it to the list as well, you know, maybe the hive has an
> opinion about it, too. ;-)
>

Good point - nothing secret about it!

Index: config/ompi_mca.m4
===================================================================
--- config/ompi_mca.m4 (revision 23770)
+++ config/ompi_mca.m4 (working copy)
@@ -564,10 +564,12 @@

    # try to configure the component. pay no attention to
    # --enable-dist, since we'll always have makefiles.
- m4_ifdef([MCA_$1_$2_$3_CONFIG],
+ AS_IF([test "$should_build" = "1"],
+ [m4_ifdef([MCA_$1_$2_$3_CONFIG],
             [MCA_$1_$2_$3_CONFIG([should_build=$should_build],
                                  [should_build=0])],
             [m4_fatal([MCA_$1_$2_$3_CONFIG macro not found])])
+ ])

    AS_IF([test "$should_build" = "1"],
          [MCA_PROCESS_COMPONENT($1, $2, $3, $4, $5, $6, $7, $compile_mode)],
@@ -619,7 +621,7 @@
            if test "$should_build" = "1" ; then
                OMPI_CONFIG_SUBDIR([$1/mca/$2/$component],
                                   [$ompi_subdir_args],
- [should_build=1], [should_build=2])
+ [should_build=1], [should_build=0])
            fi

            if test "$should_build" = "1" ; then

The problem is that the opal/memchecker framework disables itself if you don't --enable-memchecker by setting DISABLE_memchecker. The check for $DISABLE_memchecker is still intact and sets $should_build properly (to 0), but without the AS_IF above, we always run the valgrind configure.m4 CONFIG macro.

Is there a reason you removed this AS_IF in the ompi-agen branch? And/or is it ok to put it back now that we're back on the trunk?

Additionally, I think the should_build=2 in the 2nd hunk is a typo and should be should_build=0. Can you confirm?

> Cheers,
> Ralf (at least it could help weed out bugs Brian would otherwise find ...)
> _______________________________________________
> devel mailing list
> devel_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/devel