>Won't it be confusing if the rdma_create_qp_ex() call with create_flags >= 0 >succeeds, while the same call with create_flags != 0 fails (in the case >where >userlevel is running against an older kernel which does not have the >ib_uverbs_create_qp_ex() interface)?
As soon as we introduce new flags, we will have a situation where create_flags=X works, but create_flags=Y won't. (Either because of the different kernels, or devices.) This isn't really any different. >I prefer to keep the ex interface separate -- there is no point in >using the >qp_ex call if there is no intent to use the expanded feature. I can envision an application that allows user controlled parameters where calling qp_ex would always work, exactly so the user doesn't have to do an if statement to see which call to invoke. As soon as an ex type function is introduced, you could argue that the existing call should be deprecated. - 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
