Juergen Schoenwaelder <j.schoenwaelder@xxxxxxxxxxxxxxxxxxxx> wrote: > On Tue, Jul 31, 2018 at 02:27:36PM +0000, Christer Holmberg wrote: > > > > >> Related to that, it would also be good to have an interoperability > > >> statement, saying that implementations that implement the draft will > > >> still work with implementations that do not. > > > > > > This primarily concerns clients: They need to be able to fallback to > > > using <edit-config> instead of <edit-data> and <get> instead of > > > <get-data> if they communicate with a non NMDA NETCONF server. I am > > > not sure whether this is a "SHOULD be able to fallback" or a "MUST be > > > able to fallback". > > > > If you use MUST, you guarantee that fallback will always work (assuming implementations follow the spec). If you use SHOULD, I think you'll need some additional discussion on when it doesn't apply, what to do then, etc. > > > > So, my suggestion (from a reviewer perspective) would be MUST. > > > > I am not sure about this. It is very well possible that in a few years > client implementations may require NMDA and instead of trying a > fallback they stop if the peer does not support NMDA. The complexity > of clients varies widely, ranging from implementations that can hide > the complexities behind an API to simple scripts without much fallback > complexity. If we write MUST, it will be ignored in practice by a > certain fraction of clients. I agree. I don't even think we should write SHOULD. IMO we shouldn't make such requirements on the clients at all. /martin
