On Jan 18, 2012, at 3:49 PM, Jed Brown wrote:
> On Wed, Jan 18, 2012 at 12:49, Barry Smith <bsmith at mcs.anl.gov> wrote:
> Debugging fragment that should never been committed. I wanted more
> information from MPI about the problem it detected then was traditionally
> available. Maybe when the option -on_error_abor it used it should not set
> this MPI property?
>
> Minor technical issue:
>
> We currently set the error handler before checking options. Which should we
> do?
Set it to return on error as soon as possible
>
> 1. delay setting the error handler until we have checked for the option.
>
> 2. save the default error handler and restore it when we process
> -on_error_abort
Change it to MPI_ERRORS_ARE_FATAL when processing options if the user
requests -on_error_abort.
Barry
>
> Uncomment the line
>
> pushed
