[ofa-general] Bugs in opensm/libvendor

[Sasha Khapyorsky]

On 09:43 Thu 18 Dec     , Hal Rosenstock wrote:
> >
> > Are you saying we need to return potentially more than 127 paths even if
> > num_paths = 127 was specified explicitly? I don't think that this is a good idea.
> 
> I thought that was your proposal in order not to change the API for this.

My proposal is to make OSMV_QUERY_PATH_REC_BY_* stuff (in
opensm/libvendor/osm_vendor_*_sa) to be IBA complaint. And to use
OSMV_QUERY_USER_DEFINED when we would like to create custom queries
(for example without num_paths defined).

> I also thought you "liked" the extensions provided by OpenSM.

Yes, I'm. So I think it would be nice for OpenSM to handle queries where
num_paths is not specified. But in case when num_paths is requested
OpenSM should not ignore this and return records accordingly.

> > Assuming you care about breaking backward compatibility
> 
> Don't you ?
> 
> > where this osm_vendor_sa API (OSMV_QUERY_PATH_REC_BY_*) is used today?
> 
> saquery and osmtest.

I can care about saquery - I always thought that OSMV_QUERY_USER_DEFINED
is better and more useful than OSMV_QUERY_PATH_REC_BY_*.

I don't expect any hurt for osmtest (however didn't check this yet).

Sasha