Re: [netmod] WG LC draft-ietf-netmod-module-tags-02 - 10/2/18 - 10/16/18

[Andy Bierman <[e-mail]>]

On Tue, Oct 16, 2018 at 12:42 PM, Christian Hopps <[e-mail]> wrote:

>
> Andy Bierman <[e-mail]> writes:
>
> On Tue, Oct 16, 2018 at 6:08 AM, Juergen Schoenwaelder <
>> [e-mail]> wrote:
>>
>> On Tue, Oct 16, 2018 at 08:31:43AM -0400, Christian Hopps wrote:
>>> >
>>> > On Oct 2, 2018, at 4:30 PM, Juergen Schoenwaelder <
>>> [e-mail]> wrote:
>>>
>>> > > - Standard tags defined in description statements
>>> > >
>>> > >  I do not like this. YANG has extension statements and having to
>>> > >  parse stuff out of free text description statements seems to be a
>>> > >  movement backwards.
>>> >
>>> > This is used by the human implementer of the module (i.e., they need to
>>> write code to implement the module). As such it was not intended for
>>> machine parsing.
>>> >
>>>
>>> I am personally not convinced. The whole reason why we have YANG is
>>> automation and I believe people will go and write tools to extract
>>> tags and having to extract them out of free form text looks like a
>>> step backwards.
>>>
>>>
>>
>> It is more than a step backwards.
>> There is an unexplained procedure for declaring the  module-tag
>> conformance,
>> in addition to the module-tag mappings.
>>
>> All YANG designers are supposed to learn the exact text to write (not
>> free-form at all)
>> and this draft updates 6087bis with procedures for declaring the
>> module-tags
>> in the description-stmt.
>>
>> Also, tool developers are supposed to parse the description-stmt looking
>> for the
>> module-tag definitions. But instead, tool developers are going to say "Use
>> our
>> proprietary YANG extension because we are never going to parse description
>> statements"
>>
>
> I've added the extension statement.
>

good


>
> > > - System management
>>> > >
>>> > >  What is 'system management' and a 'system management protocol'?
>>> >
>>> > These were derived from the work the RtgYangDT originally did where we
>>> were organizing everything under a single device tree. This tree concept
>>> was (rightly) abandoned to be replaced with use of tags. Examples of
>>> protocols would be Syslog, TACAC+, SNMP, Netconf, ... I've added that to
>>> the description.
>>> >
>>>
>>> I am generally not a fan of definition by example. Is SSH a 'system
>>> management protocol'?
>>>
>>>
>>
>> An example is not a definition.
>> The IETF is supposed to know the difference.
>>
>
> Do you have some suggestions for text that could replace the examples?
>
> Some of these things seem painful obvious, like do we *really* need to
> define what we mean by a routing protocol?
>
>

This draft needs to define the module-tag encoding wrt/
   - valid characters (e.g., some subset of UTF-8)
   - min/max length (e.g., implementation MUST support at least 64 chars
and can support larger)

Section 3 (Tag Prefixes) seems to imply that all modules tags follow a
structure to specify the naming authority.

According to the YANG module, the data type is a plain string, which
includes lots of
problematic chars and zero-length strings.

Does the string "routing" match "ietf:routing" or "vendor:routing"? How
about "routing:bgp"?
Is the char ":" allowed in a tag?
Is "vendor::::::::" a valid tag?

IMO this draft does not need to define any specific module-tag content but
it does need to define
in precise terms how a protocol encodes a module-tag and how a module-tag
match is determined.


Thanks,
> Chris.
>

Andy