Re: [netmod] [Lsr] I-D Action: draft-ietf-lsr-ospfv3-extended-lsa-yang-10.txt

[tom petch <[e-mail]>]

From: Lsr <[e-mail]> on behalf of Rob Wilton (rwilton) <rwilton=[e-mail]>
Sent: 07 April 2022 10:25

I basically agree with Acee, and I think that we should do (b):

        b) Change the types as suggested and accept that doing so breaks
        modules where zone indexes are meaningful.

<tp>

I am concerned that such behaviour will damage the standing of the IETF at large.

We clearly laid down rules as to what updates were regarded as compatible so that authors of software could be confident that their work was robust and future-proof.  We did it with SNMP, inter alia, and we have carried that forward with YANG.  To tear up that understanding , creating who knows how much disruption, can only harm the standing of IETF.

Much has been said about how implementations have assumed that the address types do not include a zone but no evidence has been put forward for that assertion.

I have always assumed that software uses libraries and that the libraries have been written with an understanding of the specifications such that if a zone is received over the wire in conformance with the specification but where the display, field or such like does not allow for a zone, then, tolerant of what to accept, the zone is silently discarded and the address is used without the zone.  But, like the assertion that keeping the zone will cause who knows what damage, I have not done the research to substantiate that assumption.

Tom Petch 

I appreciate that this is an NBC change, but I believe that this is the most intuitive definition and is the best choice longer term.  I also note that the base ipv4-address/ipv6-address types in OpenConfig (where they use the OC copy/version of inet-types and not ietf-inet-types) don't allow a zone to be specified and assumes the default zone.  They have separate types in cases where a zone is allowed to be specified, i.e., aligned to what (b) proposes.

For modules that are using/wanting zones (if any), then they can migrate to the new explicit zone type.   draft-ietf-netmod-yang-module-versioning, if it keeps its import "revision-or-derived" extension, would also allow such modules to indicate the dependency on the updated revision/definition of ietf-inet-types.yang.

Of course, the description associated with the updated ietf-inet-types.yang revision should clearly highly the non-backwards-compatible change to the types.

Rob


-----Original Message-----
From: iesg <[e-mail]> On Behalf Of Jürgen Schönwälder
Sent: 07 April 2022 08:35
To: Acee Lindem (acee) <[e-mail]>
Cc: [e-mail]; The IESG <[e-mail]>; [e-mail]
Subject: Re: [netmod] [Lsr] I-D Action: draft-ietf-lsr-ospfv3-extended-lsa-yang-10.txt

Here is roughly what happened:

- RFC 6020 (published ~12 years ago) introduced the ip-address
  type. It included an optional zone index part since zone indexes
  are necessary in certain situations (e.g., configuring services
  listening on link-local addresses or clients connecting to services
  listening on link-local addresses).

- RFC 6991 (published ~9 years ago) added the ip-address-no-zone types
  since people felt that it is useful to also an ip address type
  without the optional zone part for situations where a zone is not
  applicable. The name 'ip-address-no-zone' was picked since the name
  ip-address was already taken.

I understand that the names resulting from this evolution of the YANG
module confuse people not looking up the type definitions. Let me note
that using a type allowing for an optional zone for a leaf that never
needs a zone is not a fatal error (its like using an int where a short
is sufficient) while using a type not allowing for a zone for a leaf
that may need zones is a fatal error (using a short where an int is
required) requiring an update of the definition of the leaf to fix.

What are our options?

a) Do nothing and accept that types are called as they are.
b) Change the types as suggested and accept that doing so breaks
   modules where zone indexes are meaningful.
c) Deprecate the types and create a new module defining new types
   so that modules can opt-in to use better names.
d) Deprecate the -no-zone types and move back to have a single
   type for IP addresses.

Any other options?

How are we going to pick between them?

/js

On Wed, Apr 06, 2022 at 09:02:23PM +0000, Acee Lindem (acee) wrote:
> Jürgen and netmod WG,  +IESG,
>
> It is not just the IETF models that are using the inet:ip-address for the standard IPv4/IPv6 addresses without zones. Every vendor’s native models and the OpenConfig models use the base types and expect the standard IP address notation. If we don’t fix this, it is something that people can point to as another example of the IETF being out of touch with reality.
>
> I thought about more, and it might make the backward compatibility easier if we just leave the existing ip-address-no-zone, ipv4-address-no-zone, and ipv6-address-no-zone types and add *-zone types for the remote possibility that someone actually wants to include the zone.  In the existing RFC 6991 BIS document, we could merely remove the zone from the ip-address, ipv4-address, and ipv6-address types and classify this as we would any other bug fix. While including the zone was the original intent of the base types, this is what those of us who work on software products would classify as a requirements bug.
>
> Thanks,
> Acee
>
> From: Andy Bierman <[e-mail]>
> Date: Tuesday, April 5, 2022 at 3:21 PM
> To: Juergen Schoenwaelder <[e-mail]>, Andy Bierman <[e-mail]>, Acee Lindem <[e-mail]>, "[e-mail]" <[e-mail]>, "[e-mail]" <[e-mail]>
> Subject: Re: [netmod] [Lsr] I-D Action: draft-ietf-lsr-ospfv3-extended-lsa-yang-10.txt
>
>
>
> On Tue, Apr 5, 2022 at 12:02 PM Jürgen Schönwälder <[e-mail]<mailto:[e-mail]>> wrote:
> On Tue, Apr 05, 2022 at 10:03:25AM -0700, Andy Bierman wrote:
> > >
> > > The best outcome would be to fix ip-address to not include the zone,
> > > introduce ip-address-zone, and deprecate ip-address-no-zone. My take all
> > > the is that all the existing usages do not require zone and this would be a
> > > fix as opposed to a change.
> > >
> > >
> > I don't think this will harm our implementations.
> > The type is still string. The pattern will change but that is handled by a
> > library.
> > Whatever pattern is used will get handled the same way.
>
> Either a zone is allowed to be present or it is not, this does make a
> difference, its not a cosmetic change.
>
>
> True. The code will probably accept the pattern then fail trying to use the string.
> If the client sends the form with a zone.
>
>
>
>
> > The same problem exists for 'date' and 'date-no-zone' types,
> > but they are not used very much.
>
> Perhaps we should call types a, b, c, and so on - this may force
> people to read the descriptions. ;-)
>
> For some reason, the smarter the person, the less likely they are to
> read any of the documentation before using some software.
> I call it the "it should work the way I would design it" phenomenon :-)
>
> You have to admit that Acee's suggestion is more intuitive than the current
> definitions.
>
> Clearly an NBC change.
> IMO it is more useful to put some YANG extension magic in these specific typedefs
> than just bumping a major revision number. This is a great use-case for the version DT.
>
> There probably is no solution path where nobody has to change any YANG or any code
> and everything still works.
>
>
>
> /js
>
> Andy
>
> --
> Jürgen Schönwälder              Jacobs University Bremen gGmbH
> Phone: [phone]         Campus Ring 1 | 28759 Bremen | Germany
> Fax:   [phone]         <https://www.jacobs-university.de/>

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

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