On Wed, Feb 26, 2020 at 08:42:57AM -0500, Dennis Dalessandro wrote: > On 2/25/2020 2:43 PM, Leon Romanovsky wrote: > > > I (and highly likely many other admins out there who don't follow this > > > list) am in full support of Haim's request for sorted output. > > > > I don't see any API or utils in the rdma-core which promised to provide > > sorted output. > > That is true, but when sys admins count on things in a certain way and the > rug gets pulled out from under them that's a headache at best. I'm not arguing about that and really understand, just don't want people to develop cargo cult around internal implementation of some tool. > > > > > > > Not everything can/will be pinned down in a manpage, but if you insist > > > then sure someone can submit a amendment to the ibstat documentation > > > which demands alphanumeric sorting for the NICs/ports. > > > > The manual is not enough, what about other tools? Should we update them > > too? Or do we need to make ibstat special while rest of the tools keep > > the list unsorted? > > > > Maybe I'm part of the problem, but any solution will require unified > > approach for whole rdma-core, so users like you will get stable and > > consistent result from any rdma-core API/tool. > > So am I understanding this right, the crux of the problem is the code > switched from using scandir() via libibumad to readdir(). The later of which > does not keep the same sorted order? > > Maybe let's look at why the change was done in the first place rather than > arguing over which way it should be. Is there a good reason to go with > readdir() approach if so then maybe that out weights the inconvenience of > not getting a sorted list. Or maybe it doesn't. I sure that Jason will remember more details, but from my point of view, everything started when we allowed to extend ibstat to look on sysfs instead of implementing netlink discovery, which will change order again. Thanks > > -Denny
