Re: [netmod] mount-point in anydata

[Martin Bjorklund <[e-mail]>]

Lou Berger <[e-mail]> wrote:
> Martin,
> 
> On 1/17/2017 7:29 AM, Martin Bjorklund wrote:
> > Hi,
> >
> > Currently, the schema mount draft says that the "mount-point"
> > extension only can be defined in an "anydata" node.  However, this
> > doesn't really work, since RFC 7950 says:
> >
> >    An anydata node is treated as an opaque chunk of data.  This data
> >    can be modified in its entirety only.
> >
> > But the idea with schema mount is to build a composite model that can
> > be manipulated just like a normal model, or a model that is
> > augmented.  If we mount models in an "anydata" node, clients would
> > have to replace the enitire mounted subtree in order to change e.g. a
> > single leaf.
> >
> > For this reason, I propose that we go back to the previous model where
> > "mount-point" would be allowed in "container" and "list".  Note that a
> > client that doesn't know anything about these mounts would see some
> > nodes in some unknown namespace; just like in the case that there is
> > an augment that the client doesn't know about.
> 
> would it be better to have the  schema mount draft updates 7950 to allow
> for sub-tree changes under anydata?  It "feels" cleaner and I'd expect
> to have no real impact on servers as this would only impact servers
> supporting the draft.  I'm unsure about client impact though...

Actually, this would have bigger impact on servers, than on clients; a
client could still treat the entire thing as an opaque blob - but
that's not what we want.  I prefer a solution that doesn't change the
core YANG semantics.  The text in 7950 is pretty clear, e.g.,:

   Any "operation" attributes present on subelements of an anydata node
   are ignored by the NETCONF server.



/martin