Logging unknown TransportParameters

[LPardue]

Back in #176 I said

The spec would be improved by having a well-defined "extension bucket" field that TPs can go in, an array of integers tuples would be fine, for example.

and we addressed that with a structured solution using group sockets in #417.

However, I think one original problem still stands. As a receiver, if I receive an unknown TP, there's not a standard way to log the codepoint and value. H3 already has a solution to this via the SETTINGS frame but QUIC TPs are different and it isn't so straightforward.

@hawkinsw has this problem right now. He is doing research on the Accurate ECN extension and very kindly proposed some en/decode to quiche. And while a small patch, adding code for an I-D we don't implement (and might want to track all I-D revisions that change the code point) adds overhead. So it got me wondering could we use qlog instead and require zero changes, leading back here.

For example, by extending parameters_set as below, we could easily externalise the code that checks supported TPs to log tooling

QUICParametersSet = {
  // all the existing stuff
  
  unknown_parameters: [* UnknownTransportParameter]
  
}

UnknownTransportParameter= {
    id: uint64,
    value: uint64
}

[hawkinsw]

@LPardue This sounds like an excellent idea! In fact, it sounds like something that I would have fun implementing! Would you allow me a crack at it?

[rmarx]

@LPardue Looking it over, it's a bit of a mess :)

For H3 we indeed have the option to log as unknown in the H3SettingsFrame, but not in the h3:parameters_set event.
In QUIC, we don't have the equivalent to H3SettingsFrame (not seriously considering adding TLS events though ;)) and so indeed this kind of falls through the cracks a bit.

So in general the proposed solution of adding unknown_parameters to quic:parameters_set looks good to me, and let's do that, but I'm wondering if the h3:parameters_set discrepancy is a problem in practice or not (probably not, but still... it's an annoying asymmetry).

Also: hey @hawkinsw, great to see you're still doing QUIC stuff and wandering over into qlog country :)

[hawkinsw]

@rmarx I am just trying to keep up with you and @LPardue !! I will take a crack at doing some "good" here and let you know how I progress. I won't get a chance to start on it until tomorrow AM. Thank you for both being so welcoming!

[LPardue]

@rmarx yeah I don't like the asymmetry much either. Although I think some of that can be put down to H3 extensions being a little more loosy goosy than QUIC. Adding a bucket to h3 params set sound fine to me.

@hawkinsw I'm not sure if your volunteering to update the qlog spec, or to experiment with quiche, or both. But whatever you'd like to volunteer is appreciated!

[hawkinsw]

@LPardue and @rmarx Thank you both for your support. If the PR (referenced above) looks like it is on the right track, I will take a whack and opening something for the qlog spec!

Thanks again!

[rmarx]

Discussed in editors meeting. No clear preference to add symmetric option for this in h3:parameters-set event as well, so the one who creates the PR gets to choose what it'll be :)

[hawkinsw]

@rmarx It looks like you might have beaten me to the punch. I'm sorry I was slow. I tried my best.