Meeting:

• would make it easier to port raw QUIC apps to WebTransport that don't have an obvious URL
• what about https://myurl/protocol
• but e.g. a https://myorigin/stream?id=queryargs
• can be added later
• QUIC provides ALPN, so if it's not available in WebTransport then MOQ will need to do something else.
• ALPN is like port numbers for TLS. Seems a bit dangerous to expose if we don't have to.

ICYMI: https://datatracker.ietf.org/doc/html/rfc7639

I noted the ALPN header RFC during the call, I think the discussion us summarized under the final point Jan-Iver captured

Meeting:

• Do we strictly needed? No. Would it be useful? Yes.
• Request is from MOQ transport.
• Proposal is you offer a list of strings and the server has to pick one of those
• Would require a new header in the IETF.
• Or could put stuff in the URL, except there's no such channel server side
• Do we want it?
• Use case: versioning negotiating in MOQ we want it to negotiate protocol version, same way HTTP/3 used to negotiate them. Typically people organize their stacks to receive ALPN, look up some code to handle specific protocol. Handler determination logic. This would let you port from QUIC to WebTransport. In WebTransport we already require you to read headers before payloads.
• There is utility of being able to receive this data before you have any application stream/datagram handling.
• API would be constructor options to add a protocol strings.
• Why is doing it in the app-layer insufficient? (if you're designing the application layer)
• Use case 2: CDN wants to stand up generic services. Which handler should the server invoke when the connection is made? Parse it as MOQ? A CDN only controls one side of the application

New header is available from the IETF side in this PR, so we should be fine to start offering a constructor that takes this list of protocol strings.

So something like:

const wt = new WebTransport(url, {protocols: ["moqt", "loc"]});
try {
  await wt.ready;
  console.log(wt.protocol); // "moqt" or "loc"
} catch (e) {
  if (e.name != SomeUniqueError) throw;
  // client did not provide a suitable protocol
}

Meeting:

• A PR would need to validate length limits for string inputs based on Add subprotocol negotiation ietf-wg-webtrans/draft-ietf-webtrans-http3#144 ...
• ...which BTW also says "Servers MAY reject the request if the client did not include a suitable subprotocol."
• If the server doesn't reject over unsupported protocol, should the API still reject?
• E.g. either
  • It could succeed and wt.protocol is null, OR
  • wt.ready rejects anyway
• Found ref "Parsers must support tokens of at least 512 characters."
• Maybe look at WebSocket for mitigations?

If the server doesn't reject over unsupported protocol, should the API still reject?

So, the reason we changed protocol from non-nullable to nullable is that this provides a way for applications that have not used subprotocols previously to start using them; it would make sense for the API to follow that logic.

Found ref "Parsers must support tokens of at least 512 characters."

I think we should pick some number and make sure that we support it, and reject everything above (the server has to pick one of the client-offered tokens, so if we say "reject above N bytes", that will be the de-facto limit).