>I've come around to agreeing with you. Since this is an internal >function we are talking about, >we're probably better off having a separate flag which indicates >whether it was invoked by >rmda_create_qp() or rdma_create_qp_expanded(). That way, we do not >limit the future behavior >of ibv_create_qp_expanded() (by having the librdmarc library depend on >knowledge of the internal >behavior of ibv_create_qp_expanded() ).
This is debatable, but rdma_create_qp_ex() with create_flags = 0 (or 1 -1 or whatever) should give the exact same functionality as calling rdma_create_qp(). The expanded call should subsume all functionality as the existing call. I believe the verbs interface should be similar. This is more about the behavior of the API, than the internal implementation. - Sean _______________________________________________ general mailing list [email protected] http://lists.openfabrics.org/cgi-bin/mailman/listinfo/general To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general
