Tweaking certain functions that sometimes break the MPI build, depending on the compiler.
I think I see what Doo Nam was trying to do, now. On certain MPI compilers, MPIBPool_RSD::create_comm() is creating problems because the variable "returnval" is created and set by an MPI function call, but only used for an assert() statement. I think the compiler optimizations are completely bypassing the assert() in release mode, which means that the compiler thinks that the variable's value is set but never used -- a warning that trips us up due to the warnings-as-errors. I'm switching these assert() checks to runtime_assert_string_msg() checks. The hit in performance will be negligible, and this stops the compiler from tripping. Plus, it's an assert worth making at runtime.