Re: [Sidrops] Requiring Two Implementations

Hi all,

Apologies for the top-post: my non-terrible mail client is in hospital :)

I agree, on a balance, that running code should be a pre-requisite for standards track documents.
Implementors should be making a conscious choice whether to implement early or await a stable specification, and to accept that this may require more iterations than if they wait for a spec to stabilize.

As long as we are robust as a group in insisting that pleas of "oh, but this will break my early implementation" not de-rail improvements to the spec, I think we can avoid the negative side effects that Geoff warns of.

Cheers,

Ben

________________________________
From: Sidrops <sidrops-bounces@ietf.org> on behalf of Job Snijders <job=40fastly.com@dmarc.ietf.org>
Sent: Monday, July 24, 2023 7:24:16 pm
To: Christopher Morrow <christopher.morrow@gmail.com>
Cc: SIDR Operations WG <sidrops@ietf.org>; SIDROps Chairs <sidrops-chairs@ietf.org>; sidrops-ads@ietf.org <sidrops-ads@ietf.org>
Subject: Re: [Sidrops] Requiring Two Implementations - before exit of WGLC

Dear Working Group,

On Mon, Jul 24, 2023 at 05:53:04PM -0700, Christopher Morrow wrote:
> Today SIDROPS has no requirement, except tacitly, to to have working
> and interoperable implementations of the proposed changes.
>
> Two (or more) times now we've had a request and discussion about
> requiring formally that there be interoperable implementations of the
> protocols/software being standardized. There have been more than one
> instance where our software/protocols are put into use and we found
> some bad interactions :(
>
> There was some chat at the meeting about this, but I think re-having
> that here is in order.
> Please take some time to consider whether or not this is a healthy
> requirement AND how we may implement it, for instance:
>
>   "Discussion and finalization of the draft happens, before exit of
> WGLC 2 implementations must be prepared and show interoperability
> (document this on sidrops-wiki with an implementation report) before
> exit of WGLC and being sent for publication."

TL;DR - YES, REQUIRE RUNNING CODE :-)

I fully support the working group requiring multiple implementations and
reports on interoperability before sending documents to the wider IETF
and IESG.

I think Geoff rightfully pointed out there is a bit of nuance, but I
think the chairs can navigate that and apply sound judgement. This
requirement should apply to adopted SIDROPS internet-drafts which
specify concepts which are to be implemented in software. So, for
example RFC 9319 didn't describe anything to implement in software, and
thus should be excempt from the requirement. I trust the chairs and
working group participants to be able to discern between such cases.

Another consideration is that for many proposed concepts there are two
sides of the coin: the certificate authorities (signers) and the relying
parties. This to me means that most specifications should strive to
producing at least two signers and two relying parties to demonstrate a
full interoperability matrix.

Proposed text:

    "Before SIDROPS Standards Track internet-drafts can progress to IESG
     review, interoperability must be demonstrated between at least two
     independent implementations for every aspect of the concepts in the
     specification. The chairs may waive this requirement when
     interoperability is of no concern (for example if the document is a
     BCP, problem statement, or requirements document).

Requiring running code has done miracles for the IDR working group, I'm
confident the quality of SIDROPS documents will improve because of this
requirement.

Thanks!

Kind regards,

Job

_______________________________________________
Sidrops mailing list
Sidrops@ietf.org
https://www.ietf.org/mailman/listinfo/sidrops