On Thu, 8 Jul 2010 13:12:03 -0700, Barry Smith <bsmith at mcs.anl.gov> wrote: > > I will use Not Collective, Logically Collective, Neighbor-wise Collective > and Collective
There are many places now where "Collective on X" appears in the documentation, but there are actually further constraints about communicators matching, etc. I'm not familiar with sowing internals, but would it be feasible to automatically generate statements about collectiveness from the guards in the source (e.g. PetscCheckSameComm(), maybe a new PetscCollective(), or even PetscLogicallyCollective() that expanded to do a barrier in a debugging mode to help diagnose misuse)? Jed
