Tests failing with OCRv1.2 on x86-mpi

[DaoWen]

When running the CnC-OCR tests using OCRv1.2 on x86-mpi as the backend runtime, the tests almost always fail with a segfault. Rolling back to OCRv1.1 fixes the issue.

This actually seems to be a bug introduced into OCR somewhere between these two releases. The segfault happens during the CnC distributed context setup on x86-mpi. My initial impression is that there is some sort of race being introduced between the creation of a remote datablock (using affinity hints) and the creation of an EDT that depends on that datablock, since the segfault appears to always happen when resolving an EDT's dependence on a remotely-created datablock.

For now, the workaround is to use OCRv1.1 as the OCR backend. I'm going to distill the issue that's presenting in CnC-OCR down to a minimal OCR example and open a ticket in the OCR issue tracker. Hopefully this issue will be resolved in the next OCR release (assuming that this is really an OCR issue and not a bug in the CnC layer over OCR).