This is part of the excellent error handing in ParMetis (sarcasm intended),
you would have found it rather quickly in the debugger. (ParMetis code follows)
idx_t BSearch(idx_t n, idx_t *array, idx_t key)
{
idx_t a=0, b=n, c;
while (b-a > 8) {
c = (a+b)>>1;
if (array[c] > key)
b = c;
else
a = c;
}
for (c=a; c<b; c++) {
if (array[c] == key)
return c;
}
errexit("Key %"PRIDX" not found!\n", key);
return 0;
}
So either it is a bug in ParMetis or ParMetis is being passed a "ill-formed"
graph and not detecting the "ill-formed" graph cleanly.
What to do? Well we can't spend out lives debugging other peoples libraries
so ..... I am not sure what to do next? Validate that Parmetis is getting
reasonable input?
Barry
On Apr 9, 2012, at 10:19 PM, Mohammad Mirzadeh wrote:
> -on_error_attach_debugger does not start the debugger. I traced it upto the
> following call: MatPartitioningApply(). It runs fine up until that function
> call and then gives the error.
>
> Also, to make things even look more weird, I got the following once I was
> running it:
>
> Key 10176 not found!
> INTERNAL ERROR: Invalid error class (66) encountered while returning from
> PMPI_Waitall. Please file a bug report.
>
> Do you happen to know what that could mean? The funny thing is, it only
> happened once!
>
> On Mon, Apr 9, 2012 at 7:57 PM, Jed Brown <jedbrown at mcs.anl.gov> wrote:
> On Mon, Apr 9, 2012 at 21:56, Mohammad Mirzadeh <mirzadeh at gmail.com> wrote:
> :( that's all I see. Do you mean running with -on_error_attach_debugger?
>
> Sure, or directly in a debugger. Does execution make it to main? Something is
> either not catching errors or is improperly calling exit() or similar.
>
