Re: [netmod] YANG Versioning: discussion around 7950 bis or errata (from Key Issue #1)

["Jan Lindblad (jlindbla)" <[e-mail]>]

Jürgen,

Fair enough. We'll discuss more later then. Thanks for providing quick feedback and an outline of what you see as favorable.

Best Regards,
/jan

On 3 Oct 2023, at 19:32, Jürgen Schönwälder <[e-mail]> wrote:

Jan,

I need to see the text. Using the existing versioning draft as a base
for this minimal solution scares me since this document went over
board. I need to see the text that details in which situations an NBC
change is allowed. I need to see the definition of the extension to
mark NBC changes. Only then I can tell whether we made progress.

If the WG does draft-schoenw-netmod-yang-relaxed-versioning-00 without
the import statement change (section 2), which was the reason to bump
the YANG language version number, then I am OK.

/js

On Tue, Oct 03, 2023 at 05:06:42PM +0000, Jan Lindblad (jlindbla) wrote:
Jürgen,

Thanks for the clarifications. It seems to me we're in complete agreement here.

Here are the use cases I can think of:
a) Tools that don't care about modules being backwards-compatible are fine. They will see good-old 7950 and 6020 YANG modules which are supposed to be backwards compatible, but sometimes are not. Occasionally they will see an unknown extension statement that they ignore.
b) Tools which flag YANG modules that are not backwards compatible and are unaware of the extension statement will flag all incompatibilities as errors. Not ideal, but an uncommon case that is pretty easy to handle in practice.
c) Tools which flag YANG modules that are not backwards compatible and are *aware* of the extension statement will flag incompatibilities without proper markup as errors, and allow the rest. This is great.
d) 7950 and 6020 module authors are (theoretically) bound by the strict sec 11 and sec 10 backwards compatibility rules, like today.
e) Module authors that need to introduce backwards incompatible changes may do so provided some appropriate extension statement defined in an upcoming document is applied.

To me, this makes sense. If it does to you too, Jürgen, that would certainly make me happy.

Best Regards,
/jan




On 2 Oct 2023, at 18:06, Jürgen Schönwälder <[e-mail]> wrote:

RFC 7950 is clear that extensions must be designed such that they can
be ignored by YANG parsers and tools.

If you use 'mandatory, are you talking about 'mandatory' in an RFC
8407 sense (and not in an YANG language sense)? The difference here is
between 'mandatory to use by module authors' versus 'mandatory to be
understood by tools'.

/js

On Mon, Oct 02, 2023 at 03:52:00PM +0000, Jason Sterne (Nokia) wrote:
Hi guys,

I think we'll need to be concrete about exactly which parts/extensions in Module Versioning we're talking about. And it will likely be a slightly different debate/discussion for each one.

I think the top level rev:non-backwards-compatible extension (and it being mandatory) is important to bundle in with the NBC rule change to SHOULD NOT.

The rev:recommended-min is useful IMO but may not be critical to include & bundle into the first versioning RFC. I still think it is useful for the YANG ecosystem to have this though.

In Key Issue #2 we've raised the question about the rev:revision-label-scheme already.

We should probably discuss each of these different & separate ideas/concepts in individual threads though..

Jason

-----Original Message-----
From: Jürgen Schönwälder <[e-mail]>
Sent: Monday, October 2, 2023 11:45 AM
To: Jan Lindblad (jlindbla) <[e-mail]>
Cc: Jason Sterne (Nokia) <[e-mail]>; Kent Watsen
<[e-mail]>; [e-mail]
Subject: Re: [netmod] YANG Versioning: discussion around 7950 bis or errata
(from Key Issue #1)


CAUTION: This is an external email. Please be very careful when clicking links or
opening attachments. See the URL nok.it/ext for additional information.



Jan,

I am certainly not against documenting NBC changes. This can be done
using extension statements. Whether such extensions are defined in the
same document or not at the end is a procedural question.

That said, any extensions that go beyond something that can be safely
ignored (e.g., extensions that for example influence how imports are
resolved) do for me require a new YANG language version. It would help
if people could acknowledge that we have agreement on this. Otherwise,
I fear that we may repeat the same discussion we had again several
months later.

/js

On Mon, Oct 02, 2023 at 02:34:31PM +0000, Jan Lindblad (jlindbla) wrote:
Jürgen, WG,

I agree that a document that updates 7950 would be the preferred solution
here, rather than a bis or errata.

I'm quite attracted, however, by the idea to bundle the softening of 7950 with
the requirement to document any incompatibilities introduced. This way, we get
something useful back as we provide the needed flexibility. This is something I
would have an easy time to explain to YANG practitioners, and it seems pragmatic
to me.

I agree completely that YANG extensions cannot change YANG at all for clients
that are not in on them. In the key issue #1 debate, however, I believe most
people agreed that we should allow non-backwards compatible changes to some
degree. To also require that any such non-backwards compatible changes are
documented using an extension statement is not to muddy the waters in my
opinion. Quite the contrary, actually. People's understanding of what's going on
will likely be improved by this requirement, for clients and server implementors
alike.

We can certainly discuss the pros and cons of requiring users to document their
non-backwards compatible changes once we have the key issue #1 behind us.

Best Regards,
/jan


On 29 Sep 2023, at 07:45, Jürgen Schönwälder
<[e-mail]> wrote:

Jason,

the must/should change is technically a change of the language. We can
do a short RFC to do that so that we get unstuck and oour AD allows us
again to publish YANG modules where bug fixes or alignment with other
modeled technologies is desirable.

Adding decorations that can be ignored is something one can do with
YANG extensions.  However, once such extensions change the behaviour
of YANG language constructs, we get into muddy waters.

I prefer to clearly separate changes of the language from additional
decorations that can be ignored and do not influence the behaviour of
YANG implementations (i.e., they can be ignored).

/js

On Thu, Sep 28, 2023 at 08:57:42PM +0000, Jason Sterne (Nokia) wrote:
Hi all,

IMO - We've already started moving out of the "stuck" situation. We no longer
have to debate whether a new YANG 1.2 is needed for allowing an NBC change.
That will be the end of a big distraction and circular discussions for the WG.

I'm not so convinced we want to rush and do a separate RFC just for that one
part of Module Versioning (and one part of the original versioning requirements).
It is a key/critical part, but we should continue discussing what other parts we'd
want to also tackle as part of the "first" versioning RFC.

I'm very doubtful we should relax MUST to SHOULD NOT without also at least
making the rev:non-backwards-compatible marker mandatory (as per Module
Versioning). The marking is a key part of making this all better for consumers of
modules and clients (one of the main problems is the current silent NBC changes
happening).

We should also clarify that marking an element as "status obsolete" is NBC.
That has major impact on clients who are trying to continue using an old version
of the module.

(and there are likely at least a few other pieces from Module Versioning that
should be in a "first" RFC)

Jason

-----Original Message-----
From: netmod <[e-mail]> On Behalf Of Jürgen Schönwälder
Sent: Thursday, September 28, 2023 9:12 AM
To: Reshad Rahman <[e-mail]>
Cc: Kent Watsen <[e-mail]>; [e-mail]
Subject: Re: [netmod] YANG Versioning: discussion around 7950 bis or errata
(from Key Issue #1)


CAUTION: This is an external email. Please be very careful when clicking links or
opening attachments. See the URL nok.it/ext for additional information.



The truth is that we did bug fixes in the past. We now have maneuvered
us into a situation where work is put on hold because we do not even
do bug fixes anymore (and yes, I know, the line between bug fixes,
alignment with moving targets and other changes is vague and needs to
be decided on a case by case basis). The fastest way to get unstuck is
to write this one page content RFC that changes MUST to SHOULD and
then we at least get out of the being stuck situation.

/js

On Thu, Sep 28, 2023 at 01:00:23PM +0000, Reshad Rahman wrote:
As a client (consumer of models), I do not want only the MUST -> SHOULD
change, IMO that would be worse than the current situation.
Regards,Reshad.
 On Wednesday, September 27, 2023, 09:16:10 PM EDT, Kent Watsen
<[e-mail]> wrote:

This was my thought as well, that it would be best to have the smallest-possible
draft update 6020/7950.  That way, when someone follows the “Updated” links,
they’re not overloaded with material that could’ve been left out.
Jason was saying that just doing MUST/SHOULD by alone isn’t great, that at
least the "rev:non-backwards-compatible” extension statement should be
included and, by extension I suppose, the rules for editing the revision history.
Presumably revision labels could be left out.  IDK what minimal is possible.
K. // contributor



On Sep 27, 2023, at 7:06 PM, Rodney Cummings
<[e-mail]> wrote:


It is easy to write a short RFC updating RFC 7950, changing one sentence from
MUST to SHOULD.


I agree. I found that I cannot enter a response to the poll, because I disagree
with both Option 1 and Option 2.

My concern is that there are many people out there who are implementing
YANG, but who do not follow discussions on this mailing list. I'm concerned that
there is a serious risk that those people will interpret the change from MUST to
SHOULD as "backward compatibility is irrelevant for YANG". We all know that
the
concern is about bug fixes and so on, but without explaining that in a short and
focused manner (i.e., the short RFC described above), that will be lost in the
noise
of the larger draft-ietf-netmod-yang-module-versioning change.

draft-ietf-netmod-yang-module-versioning is a great draft, but I think it should
move forward as an independent RFC, distinct from the MUST/SHOULD change.

Rodney Cummings

-----Original Message-----
From: netmod <[e-mail]> On Behalf Of Jürgen Schönwälder
Sent: Tuesday, September 26, 2023 5:24 PM
To: Jason Sterne (Nokia) <[e-mail]>
Cc: [e-mail]
Subject: Re: [netmod] YANG Versioning: discussion around 7950 bis or errata
(from Key Issue #1)

It is easy to write a short RFC updating RFC 7950, changing one sentence from
MUST to SHOULD. This is inline with the goal to not change the language, i.e.,
to
keep the version numbers.

/js

On Tue, Sep 26, 2023 at 03:00:19PM +0000, Jason Sterne (Nokia) wrote:

Hello NETMOD WG,

We've had a poll going for a few weeks to determine if we require YANG 1.2 for
allowing ("SHOULD NOT") NBC changes (see "Poll on YANG Versioning NBC
Approach").

As part of that, some discussion has happened on the list around
potentially doing an errata for RFC7950/6020 or a bis of 7950/6020 (if
rough consensus is reached for option 1 of the poll)

7-8 of us discussed this in the YANG Versioning weekly call group today.

First of all: this question of mechanics (errata vs bis vs Module Versioning draft)
is orthogonal to the poll. Let's first and separately resolve the poll and confirm if
we need YANG 1.2 or not (that's the fundamental question the poll is resolving -
everything else is a subsequent issue to be discussed). We'll let the chairs
confirm
when/if rough consensus on the poll has been reached.

But *if* the answer to the poll is option 1, then the weekly call group was
unanimous that we should not do an errata for RFC7950/6020 and we should
not
do a 7950/6020 bis. We should just continue with the Module Versioning draft
which will update 7950 and 6020.

The primary reason is that we shouldn't just change MUST NOT to SHOULD NOT
without also tying it together with the mandatory top level rev:non-backwards-
compatible extension when an NBC change is done. Changing the NBC rule to
SHOULD NOT needs to be in the same RFC as the mandatory rev:non-
backwards-
compatible tag.

Other reasons:

*   an errata probably isn't correct since this isn't fixing an intent that was
present back when 7950 was written (it was clearly the intent at the time to
block NBC changes)
*   a bis would be odd without actually introducing other changes to YANG and
changing the version (this discussion is all based on "if the answer to the poll is
option 1")

Jason (he/him)




_______________________________________________
netmod mailing list
[e-mail]
https://www.i/


etf.org%2Fmailman%2Flistinfo%2Fnetmod&data=05%7C01%7C%7C22464d2aa09
441


f1b1bd08dbbedf65ad%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C6
38313


638956186415%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIj
oiV2luM


zIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C3000%7C%7C%7C&sdata=DgsZVlBTQt
qJjR
tVXs%2Bze%2BrOanijgDEuCn93gbN9Jyw%3D&reserved=0



--
Jürgen Schönwälder              Constructor University Bremen gGmbH
Phone: [phone]         Campus Ring 1 | 28759 Bremen | Germany
Fax:   [phone]         <https://constructor.university/>

_______________________________________________
netmod mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/netmod

_______________________________________________
netmod mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/netmod
_______________________________________________
netmod mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/netmod


--
Jürgen Schönwälder              Constructor University Bremen gGmbH
Phone: [phone]         Campus Ring 1 | 28759 Bremen | Germany
Fax:   [phone]         <https://constructor.university/>

_______________________________________________
netmod mailing list
[e-mail]
https://www.ietf.org/mailman/listinfo/netmod

--
Jürgen Schönwälder              Constructor University Bremen gGmbH
Phone: [phone]         Campus Ring 1 | 28759 Bremen | Germany
Fax:   [phone]         <https://constructor.university/>

_______________________________________________
netmod mailing list
[e-mail]<mailto:[e-mail]>
https://www.ietf.org/mailman/listinfo/netmod


--
Jürgen Schönwälder              Constructor University Bremen gGmbH
Phone: [phone]         Campus Ring 1 | 28759 Bremen | Germany
Fax:   [phone]         <https://constructor.university/>

--
Jürgen Schönwälder              Constructor University Bremen gGmbH
Phone: [phone]         Campus Ring 1 | 28759 Bremen | Germany
Fax:   [phone]         <https://constructor.university/>


--
Jürgen Schönwälder              Constructor University Bremen gGmbH
Phone: [phone]         Campus Ring 1 | 28759 Bremen | Germany
Fax:   [phone]         <https://constructor.university/>

_______________________________________________
netmod mailing list
[e-mail]<mailto:[e-mail]>
https://www.ietf.org/mailman/listinfo/netmod