Caitlin Bestler wrote:
The impact of this on the existing code would be quite wide,
but quite shallow. Many types would be renamed form 'ib_xyz'
to 'rdma_xyz', if being used in transport neutral code. Or
when staying infiniband specific fields that now qualified
as transport neutral would require "common." to be inserted
before the field name. Some constants are shifted from IB_XXX
to RDMA_XXX.

Embedding a struct rdma_object inside a struct ib_object essentially removes type checking from the API. For example, a user calling rdma_poll_cq() must know that they are using IB and reference the correct data structure, such as:

struct ib_wc wc;

rdma_poll_cq(cq, 1, &wc.common);

Similarly for posting RDMA requests through rdma_post_send().

- Sean
_______________________________________________
openib-general mailing list
[email protected]
http://openib.org/mailman/listinfo/openib-general

To unsubscribe, please visit http://openib.org/mailman/listinfo/openib-general

Reply via email to