Re: [netmod] Proposal to enhance the YANG tree output

[Lou Berger <[e-mail]>]

On September 20, 2017 2:33:47 AM Martin Bjorklund <[e-mail]> wrote:

> Lou Berger <[e-mail]> wrote:
>>
>>
>> On 9/19/2017 9:42 AM, Martin Bjorklund wrote:
>> > Lou Berger <[e-mail]> wrote:
>> >>
>> >> On 9/19/2017 7:29 AM, Martin Bjorklund wrote:
>> >>> Lou Berger <[e-mail]> wrote:
>> >>>> Martin,
>> >>>>
>> >>>> Speaking as a contributor:
>> >>>>
>> >>>>
>> >>>> On 9/15/2017 7:40 AM, Martin Bjorklund wrote:
>> >>>>> Robert Wilton <[e-mail]> wrote:
>> >>>>>> On 15/09/2017 11:21, Ladislav Lhotka wrote:
>> >>>>>>> Andy Bierman píše v Čt 14. 09. 2017 v 08:43 -0700:
>> >>>>>>>> Hi,
>> >>>>>>>>
>> >>>>>>>>
>> >>>>>>>> Actually I liked the early pyang output that was concise and easy to
>> >>>>>>>> remember.
>> >>>>>>>> The current format gets very cluttered and there are too many little
>> >>>>>>>> symbols
>> >>>>>>>> to remember them all.
>> >>>>>>> I agree.
>> >>>>> Me too.  The current draft adds three new magic symbols: "mp" "@" and
>> >>>>> "/".
>> >>>>>
>> >>>>> "mp" is for a mount point, and it can be generated directly from the
>> >>>>> YANG modules.
>> >>>>>
>> >>>>> Directly under a "mp", "/" and "@" are used to indicate that a node 
>> is mounted
>> >>>>> or available through a parent reference, respectively.
>> >>>>>
>> >>>>> I actually question the usability of "/" and "@".
>> >>>> I agree that / and @ are something new, and enabled by schema mount. 
>> >>>> There have been repeated comments in the RTG WG that there needs to be
>> >>>> some way for vendors to convey what they have implemented with Schema
>> >>>> mount
>> >>> If that's the requirement, using the tree diagram is probably not the
>> >>> best way.  The tree diagram is intended to provide an overview of a
>> >>> given (set of) YANG module(s).
>> >>>
>> >>> A perhaps better way to convey the information is to create a file
>> >>> with an instantiated /schema-mounts tree.
>> >> using what syntax?  JSON and XML really isn't that easy for the (human)
>> >> reader to parse.
>> > Either JSON or XML.
>> This is fine for code, less so for humans.  We include both in the NI
>> draft, the tree provides a quick overview, the JSON provides the details. 
>>
>> >
>> >>>> and this is one way to help convey (a) what is expected of server
>> >>>> implementors and (b) what client implementors should expect.
>> >>>>
>> >>>    Hence the
>> >>>> current draft text:
>> >>>>
>> >>>>   In describing the intended use of a module containing a mount point,
>> >>>>    it is helpful to show how the mount point would look with mounted
>> >>>>    modules.
>> >>>>
>> >>>> The whole point of trees is to facilitate understanding for those who
>> >>>> are less familiar with a model than the authors, and IMO that's the
>> >>>> paramount perspective in this discussion.
>> >>> Fully agree!  I would say that we have to make sure that the diagrams
>> >>> can be understood by people less familiar with the technology than the
>> >>> authors.  Mixing schema and instance data does not help here.
>> >> Can you propose an alternative?
>> > As I have written before, I think the "/" is not needed, so I would
>> > remove that.  I would also not list the nodes from "parent-references"
>> > in the same tree ouput.  It is not clear to me that this level of
>> > detail is needed in the tree, and - as noted before - it isn't even
>> > correct to list e.g. "interfaces" when the parent-reference in fact
>> > selects a single interface.
>> >
>> >> The routing WG participants seem to
>> >> find these useful, we can also ask there for broader input if you'd like.
>> > One approach is to add the union of eveything that some people find
>> > useful.  In the end we have to look for WG consensus.  Several people
>> > have said that they prefer a less cluttered format.
>> In the context of listing enums...
>>
>> >
>> >>>>> Since a parent
>> >>>>> reference can be very specific, e.g. one specific interface, it isn't
>> >>>>> really accurate to show:
>> >>>>>
>> >>>>>                   +--mp vrf-root
>> >>>>>                      +--rw rt:routing/
>> >>>>>                      |  ...
>> >>>>>                      +--ro if:interfaces@
>> >>>> This is just a conditional and we have precedent on how to handle
>> >>>> conditional representation.   
>> >>>>
>> >>>>> And the trailing "/" on rt:routing doesn't add any information we
>> >>>>> don't already know.  Since vrf-root is a mount point, it follows that
>> >>>>> its children are mounted.
>> >>>> The issue is a bit more complex when considering some real use cases,
>> >>>> particularity when parent references and augmentations are used.  For
>> >>>> example consider the following *without* the use / or @:
>> >>>>
>> >>>> module: ietf-network-instance
>> >>>>   +--rw network-instances
>> >>>>      +--rw network-instance* [name]
>> >>>>         | ...
>> >>>>         +--rw (root-type)
>> >>>>            +--:(vrf-root)
>> >>>>               +--mp vrf-root
>> >>>>                  +--ro rt:routing-state
>> >>>>                  |  +--ro router-id?                 yang:dotted-quad
>> >>>>                  |  +--ro control-plane-protocols
>> >>>>                  |     +--ro control-plane-protocol* [type name]
>> >>>>                  |        +--ro ospf:ospf
>> >>>>                  |           +--ro instance* [af]
>> >>>>                  +--rw rt:routing
>> >>>>                  |  +--rw router-id?                 yang:dotted-quad
>> >>>>                  |  +--rw control-plane-protocols
>> >>>>                  |     +--rw control-plane-protocol* [type name]
>> >>>>                  |     +--rw ospf:ospf
>> >>>>                  |        +--rw instance* [af]
>> >>>>                  |           +--rw areas
>> >>>>                  |              +--rw area* [area-id]
>> >>>>                  |                 +--rw interfaces
>> >>>>                  |                    +--rw interface* [name]
>> >>>>                  |                       +--rw name if:interface-ref
>> >>>>                  |                       +--rw cost?   uint16
>> >>>>                  +--ro if:interfaces
>> >>>>                  |  ...
>> >>>>                  +--ro if:interfaces-state
>> >>>>                  |  ...
>> >>>>
>> >>>>
>> >>>> It's certainly not too hard to spot the top level mounts, but it is
>> >>>> impossible to distinguish the parent references from the actual mounts. 
>> >>> My proposal would be to not even show the parent references in the
>> >>> diagram.  If we do that, the '/' is not needed.
>> >>>
>> >>>> Further more, some mounts are hard to spot.  For example, notice ospf. 
>> >>>> Did you notice that it's a mount?
>> >>> This is actually not correct.  ospf is *not* a mount; it is an augment.
>> >>>
>> >> it's a mounted module that augments another mounted module.
>> > But why would it have a "/" just b/c it is augmented, when its sibling
>> > 'control-plan-protocol' doesn't have the "/"?  What addition info does
>> > the "/" convey?
>> That it is a (sub)tree that is present due to a mounted module.
>
> So is 'control-plane-protocol', so then it should also have '/',
> right?

Why, it's defined in ietf-routing (under rt:routing-state and rt:routing)?

Lou

>
>
> /martin