[MPICH] bug in evaluating MPIEXEC_MACHINES_PATH

William Gropp gropp at mcs.anl.gov
Thu Feb 15 07:35:05 CST 2007


Thanks for finding this.  remshell is an experimental process manager  
and hasn't been extensively tested.  This will be fixed for the next  
release, and I can send you a patch (with your next bug as well).
Bill

On Feb 15, 2007, at 3:19 AM, Martin Kleinschmidt wrote:

> when evaluating MPIEXEC_MACHINES_PATH, remshell throws away the last
> character of MPIEXEC_MACHINES_PATH, regardless if it is a slash or  
> not:
>
> $ setenv MPIEXEC_MACHINES_PATH `pwd`
> $ echo $MPIEXEC_MACHINES_PATH
> /data2/mk/518977.fats.theochem.uni-duesseldorf.de
> $ mpiexec -n 4 program
> Could not open machines file
> /data2/mk/518977.fats.theochem.uni-duesseldorf.d/machines
> Could not find machines for default architecture
> Unable to assign hosts to processes
>
> note the missing "e" at the end of the directory name!
> adding a slash for removal helps:
>
> $ setenv MPIEXEC_MACHINES_PATH /data2/mk/518977.fats.theochem.uni- 
> duesseldorf.de/
> $ echo $MPIEXEC_MACHINES_PATH
> /data2/mk/518977.fats.theochem.uni-duesseldorf.de/
> mpiexec -n 4 ~/prog/mrci/mrci.my_version/mrci
>
> (runs)
>
>
> The last character should only be thrown away if it is irrelevant.
>
>    ...martin
>
> -- 
> Wir muessen jetzt endlich den Arsch hoch kriegen und Eier zeigen.
> (Martin Pieckenhagen)
>




More information about the mpich-discuss mailing list