[mpich-discuss] intercomm merge then intercomm create bug/misunderstanding?

[Rajeev Thakur]

I think your logic is ok, although I would move the if (rank==0) outside the
else and use rank in new_global instead of comm_world (because the
controller is set up in terms of rank 0 in new_global). Then you don't need
a program_c. 

But it makes no difference in the output. There must be a bug in the
implementation.

Rajeev



> -----Original Message-----
> From: owner-mpich-discuss at mcs.anl.gov 
> [mailto:owner-mpich-discuss at mcs.anl.gov] On Behalf Of Brad Penoff
> Sent: Friday, June 27, 2008 4:38 PM
> To: mpich-discuss at mcs.anl.gov
> Subject: Re: [mpich-discuss] intercomm merge then intercomm 
> create bug/misunderstanding?
> 
> On Fri, Jun 27, 2008 at 8:39 AM, Robert Latham 
> <robl at mcs.anl.gov> wrote:
> > On Fri, Jun 27, 2008 at 01:19:43AM -0700, Brad Penoff wrote:
> >> However, when I do this, I keep getting this error that I'm
> >> having difficulty understanding:
> >>
> >> mpirun -np 3 ./spawn_merge_create_simple 2      #  the arg 2 is how
> >> many procs to spawn
> >
> > Hey Brad
> >
> > I get this same error as well with MPICH2 from SVN (the 
> gang here has
> > done a lot of work on the dynamic process path since the release...)
> >
> > If I only mpiexec one process and let mpi_comm_spawn build the rest,
> > things seem to go ok.    I can't add too much to help you 
> out: I only
> > break MPI-2 dynamic process routines, not work on them :>
> 
> hey Rob,
> 
> Thanks, ya I hadn't tried the latest MPICH2 from SVN.  For the record,
> on MPICH2 1.0.7, this same code I sent you guys a link to also works
> if I run only one process.  This says there's either a bug in MPICH2
> or my "program_a" logic for creating the intercommunicator is
> incorrect...
> 
> I noticed on the new MPICH2 webpage that the bug tracker is a work in
> progress... Do you know if there is a bug formally logged against
> this?
> 
> Thanks again,
> brad
> 
>