On Mon, Feb 18, 2013 at 3:00 PM, Jed Brown <jedbrown@mcs.anl.gov> wrote:

Even in Barry's original code, I don't understand why we can expect this check to be producing a collective result. Surely there can be some ranks that see everything evenly divide and others that do not.

It seems to me that we should either (a) explicitly disallow matrices with a block size to be built when the blocks are not filled or (b) when the user requests blocks, pad blocks so that we can always use an ISBlock.

It shouldn't be hard to build a test that will expose this problem. Just declare a block size (2), then assemble a matrix that does not fill out the block size. In the off-diagonal part of rank 0, we should have one full block, while in the off-diagonal part of rank 1, we should have anything that isn't a perfect block.