Re: How will we test version-negotiation?

[Marten Seemann <[e-mail]>]

This seems like a reasonable proposal. Using a virtual version that is
identical to the current draft version make sense to me.
It would also make it easy to add a test case for this to the interop
runner.

On Fri, Apr 24, 2020 at 11:58 PM Christian Huitema <[e-mail]>
wrote:

> I am reading draft-ietf-quic-version-negotiation-00 with an eye on
> implementing the version negotiation, and I wonder how we are going to
> test interoperability of different implementations.
>
> The proposed version negotiation only kicks in if the client sets the
> version number that the server also supports. Then, the server checks
> whether a "better" version is also available and supported by the
> client, and negotiates that. This does not mesh very well with our
> current interoperability tests, which require support for the latest
> draft. In the current tests, every implementation supports draft 27 and
> proposes version 0xFF00001B. There is no version to upgrade to.
>
> Some implementations also support earlier drafts, and could test
> upgrades from draft-25 to draft-27, and we could define a test that way.
> But many don't, and it seems weird to require support of old versions
> just for the sake of testing version upgrade. I don't think that's a
> good idea. I think we need a better proposal, and here is one: let's
> define a "virtual" version that is only used for testing upgrades.
> Better yet, let's define a virtual version that just matches the latest
> draft -- say 0xFFFF001B for draft 27. The interop test could work this way:
>
> 1) Client sets the version number to the latest draft (0xFF00001B) in
> the Initial packet header, and mentions support of 0xFFFF001B in the
> Version Negotiation Transport Parameter.
>
> 2) Server accepts the Initial packet, parses the transport parameters,
> discovers the proposed support for the "better" version.
>
> 3) test succeeds if the server negotiates version 0xFFFF001B and the
> data exchanges succeed.
>
> What do you think? Should we add this test mechanism to the draft, or
> would it be enough to describe that in the interop wiki pages?
>
> -- Christian Huitema
>
>
>
>
>