Re: [core] CBOR Encoding of Data Modeled with YANG

[Andy Bierman <[e-mail]>]

Hi,

It is not clear to me how manual numbering of objects can be practical.
Automatic numbering really means "write a program that does the manual
numbering".  YANG allows new nodes to be added anywhere.
Only the relative order of nodes is maintained across revisions.
Some nodes, like "augment" are not ordered at all and they
can change order from 1 revision to the next.  If a grouping in an
external module is extended, then all "uses" of the grouping will be
extended.
If "import without revision" is used, then the specific external grouping
is not even knowable and it is impossible to number the expanded "uses"
correctly.
If any module ever has more than 1024 objects, they will be ignored,
since the numbering has a hard limit on objects per module.

These seem like significant issues that cannot be ignored.


Andy


On Tue, Dec 8, 2015 at 5:10 AM, Alexander Pelov <[e-mail]> wrote:

> Dear Juergen,
>
> Thanks for the great remarks! Indeed, we have had lots of lengthy
> discussions on the way structured data node IDs work, and there are several
> ways to handle the situations you describe.
>
> We've reached a consensus to first detail the most demanded and least
> controversial part - the YANG-CBOR mapping. Having structured data node IDs
> allows for all existing designs to operate with this.
>
> I've provided you with the baseline algorithm, which covers one aspect of
> the data node IDs. Structured data node IDs also allow for private module
> ID space, and for non-managed modules (e.g. as the example you provide).
> They also allow for having COMI-style hashes if you need them. The draft
> which will describe the full operation will come shortly after the
> YANG-CBOR draft.
>
> I think that it would be nice if you can contribute, as your remarks can
> help clear the readability of the document and point on examples that must
> be provided to cover the full spectrum of YANG use-cases.
>
> Best,
> Alexander
>
>
> Le 08/12/2015 12:53, Juergen Schoenwaelder a écrit :
>
>> On Tue, Dec 08, 2015 at 12:11:00PM +0100, Alexander Pelov wrote:
>>
>>> Dear Juergen,
>>>
>>> See inline.
>>>
>>> Le 8 déc. 2015 à 11:55, Juergen Schoenwaelder <
>>>> [e-mail]> a écrit :
>>>>
>>>> On Tue, Dec 08, 2015 at 11:16:58AM +0100, Alexander Pelov wrote:
>>>>
>>>>> Dear Juergen,
>>>>>
>>>>> The algorithm is straightforward - assign in increasing order a data
>>>>> node ID to each element on the schema tree. If you have a revision of a
>>>>> module, you do the diff between the old and the new tree, and add the
>>>>> new data node IDs at the end of the numbering.
>>>>>
>>>> So I need access to the complete revision history in order to determine
>>>> data node IDs, correct?
>>>>
>>> It is not a must. You can manually indicate the node IDs of the newly
>>> added elements. If you want a completely automated allocation of IDs, then
>>> yes. The way I think this will work is - the publisher of a module will use
>>> the automatic tool to generate the statements, which will then be indicated
>>> in the YANG definition.
>>>
>>> I want independently developed systems to interoperate. In general, I
>> can't assume modules have data node IDs (since a YANG module like lets
>> say ietf-interfaces simply does not define them).
>>
>> What about imports without a revision (a nasty
>>>> source of ambiguity)?
>>>>
>>>> There are two points in your question. The first is straightforward -
>>> Data node IDs make sense within the scope of a module. Changing a module
>>> does not affect the different modules which import it, with the possible
>>> exception of groupings, where the classical data node ID algorithm
>>> assignment kicks in (or manual ID assignment if the module author prefers
>>> so). In any case, I would refer you to RFC 6020, section 5.1.1. where the
>>> question of acceptability of imported modules is treated.
>>>
>> I am talking about imports _into_ a module that may be ambigous from
>> the viewpoint of the importing module. Section 5.1.1 is about import
>> by revision, my question was about what happens if there is an import
>> without revision. We had lengthy discussions in NETMOD what this means
>> for conformance and the outcome roughly is that this requires a
>> runtime lookup to find out which module revisions have been used by a
>> given implementation, that is, your algorithm would not only need the
>> complete revision history but also runtime information in order to
>> calculate unique data node ids.
>>
>> There are alternate designs possible in the solution space. I like to
>> understand the trade-offs between them.
>>
>> /js
>>
>>
> _______________________________________________
> core mailing list
> [e-mail]
> https://www.ietf.org/mailman/listinfo/core
>