[openib-general] [RFC] [PATCH 3/7] ibrdmaverbs hdr files

[Krishna Kumar2]

To clarify, I am attempting to not duplicate any function, though data
structures will get duplicated (with the knowledge that one set will be
removed at some stage). The idea is to define all rdma_* functions
and make the exported ibv_ functions call these rdma_ exported API's.
(eg :
ibv_cmd_create_qp(...)
{
        return rdma_cmd_create_qp();
}
, etc). The un-exported ibv_ routines can be modified without any such
action.

This also ensures that if a bug is found in creation of a qp, the fix will 
go
into the "base" function - rdma_cmd_create_qp() and the same fix doesn't
have to be applied to ibv_cmd_create_qp.

Hope this clarifies my intention.

thanks,

- KK

openib-general-bounces at openib.org wrote on 07/20/2006 02:53:13 PM:

> Hi,
> 
> Are you are OK with :
> 
> 1. Changing all non-exported routines in libibverbs to the new API.
>     All static functions can be changed immediately instead of going
>     the "deprecation" way.
> 
> 2. Adding new duplicate API's for those API's that are exported to
>     other libraries.
> 
> Is this the right thing to do and acceptable ?
> 
> Thanks,
> 
> - KK
> 
> 
> Roland Dreier <rdreier at cisco.com> wrote on 07/10/2006 10:27:03 PM:
> 
> >     James> Long term, I don't think it makes sense having two
> >     James> libraries with the exact same functionality, but different
> >     James> function names.  Replacing ibv_ with rdma_ would be ideal,
> >     James> but disruptive to current users. Could you provide a
> >     James> backwards compatibility header (one that mapped each ibv_
> >     James> function to its equivalent rdma_ function)?
> > 
> > Agreed.  I think that the way forward is to provide a libibverbs
> > release that has both rdma_ and ibv_ prefixed identifiers, and also
> > provides both <infiniband/verbs.h> and <rdma/verbs.h> headers, etc.
> > Then in the next major libibverbs release, the ibv_ versions of
> > identifiers can be deprecated, and finally removed in the next major
> > release (which could be renamed to librdmaverbs).
> > 
> > Following that route is something that I've been meaning to work on,
> > and I'd be happy to accept patches in that direction.  But a big bang
> > approach that breaks the world is not a good idea I don't think.
> > 
> >  - R.
> 
> 
> _______________________________________________
> openib-general mailing list
> openib-general at openib.org
> http://openib.org/mailman/listinfo/openib-general
> 
> To unsubscribe, please visit 
http://openib.org/mailman/listinfo/openib-general
>