Open MPI logo

Open MPI Development Mailing List Archives

  |   Home   |   Support   |   FAQ   |   all Development mailing list

From: Timothy B. Prins (tprins_at_[hidden])
Date: 2005-07-22 17:42:28


Problem 2 should be fixed in the trunk. We now detect if the proper
version of bproc is available, and we will abort configure if the proper
version is not available and --with-bproc is passed.

To disable bproc entirely, simply pass --without-bproc or --with-bproc=no
to configure (this functionality was previously broken).

The current bproc modules support LANL bproc >=3.2.0. They do not
currently run on any version of Scyld or older LANL bproc. If anyone is
curious, we detect the bproc version by looking for sys/bproc_common.h,
which is only present in LANL bproc v 3.2.0 and greater. By happy
coincidence, v3.2.0 is the exact version where all the features we use are
first available.

Sorry, but this change should require an autogen/configure....

Tim

> There appear to be 2 bproc problems in the tree right now:
>
> 1. I mailed Tim Prins / Greg Watson about one of them already (trying
> to compile bproc on systems that don't have bproc -- perhaps related to
> the .ompi_unignore from last night?).
>
> 2. The second was noticed by Joel Krauska from Cisco (he'll probably be
> on this list shortly). Here's a mail he sent to me last night:
>
>> I haven't done the due diligence to attempt to disable the bproc stuff
>> -- it just appears to not agree with my system and it's in the stock
>> "make".. I will later read the documentation, but it looks like
>>
>> ompi/orte/mca/pls/bproc_seed/pls_bproc_seed.c:446
>> rc = bproc_vrfork(num_nodes, node_list, daemon_pids);
>>
>> conflicts with my scyld system's
>> /usr/include/sys/bproc.h
>> int bproc_vrfork (int *nodes, int nnodes);
>
> Right now, I think we're just checking for bproc.h to determine if the
> system has bproc -- we're not doing anything to figure out *which*
> bproc you have (LANL vs. Scyld).
>
> Does anyone have any Scyld machines lying around? The prototype for
> vrfork() is one indicator, but not the easiest to test fork -- are they
> any other telltale #define's or such that we can use for testing in
> configure?
>
> --
> {+} Jeff Squyres
> {+} The Open MPI Project
> {+} http://www.open-mpi.org/
>
> _______________________________________________
> devel mailing list
> devel_at_[hidden]
> http://www.open-mpi.org/mailman/listinfo.cgi/devel
>