Benjamin M. Schwartz

Note that this is the upstream read, so network misbehavior has no effect. Any app that retries its UDP packet at least once every 30 seconds will keep the NAT...

Personally, I don't see a material advantage to the FINAL_DATA+ERROR or END_DATA proposals. In the FINAL_DATA proposal, if we imagine a client that receives DATA, DATA, CONNECTION_METADATA, (end stream), it...

> Though you linked https://github.com/httpwg/http-extensions/pull/2997 and that seems unrelated, did you mean another PR? You're right, I meant #2949. Corrected.

@kazuho @PiotrSikora You both expressed interest in an explicit RST signal in #2949. Are you still interested in that, or should we move forward with #2949 as-is?

> There was a question about what to do with the TCP connection with HTTP/1.1. https://stackoverflow.com/questions/38507586/reset-a-tcp-socket-connection-from-application might help. This StackOverflow answer strikes me as evidence that a large fraction of...

> The drawback of [#2949](https://github.com/httpwg/http-extensions/pull/2949) as it stands is that you can't relay a TCP RST without using RST_STREAM This is not correct. A stream that ends gracefully but without...

I think we are all discussing a simple model with three states for each leg: open, complete, and truncated. There are only two allowed transitions: open->complete (FIN) and open->truncated (RST)....

I've significantly updated #2949 based on feedback from @wtarreau. #2949 now enables this distinction without adding any significant mandatory requirements. Please review #2949. I'll plan to merge it in 48...

Merged. Thanks all for your input on this topic.

When writing DNS software related to SVCB/HTTPS records, there are many cases where wants to perform type-independent processing or generation of these record types, so having a unified type would...