Re: [quicwg/base-drafts] Use a "stream" for transmitting CIDs (#1826)

[Marten Seemann <[e-mail]>]

> In my mind, using an actual stream (with STREAM and MAX_STREAM_DATA frames) eliminates the need for new frame types. Imagine a special flow-controlled CID stream. For simplicity, assume CIDs are 8 bytes long. If the flow control window is 64 bytes, writing 32 bytes to the stream (4 CIDs) will trigger a MAX_STREAM_DATA from the peer. This is the signal to write more CIDs to the stream. If the peer is not consuming the CIDs, MAX_STREAM_DATA is not sent and no CIDs are emitted.

That seems like a misuse of streams to me. You could end up with CID frames split up in the middle. I'm not saying that this wouldn't work, it just doesn't seem like a nice way to engineer a solution to me.

> The original objections to this was that the transport layer does not own the streams. I don't know how to respond to this. It seems that adding new frame types and protocol-like specifications around them for every new thing that we think to add to QUIC is not the most pragmatic thing to do. Take the CRYPTO frames, for example: chances are, they are just wrappers on the stream mechanism anyway.

They're not, in my implementation.
To me, using frames and protocol-like specification for new protocol features seems like exactly the way things should be.

-- 
You are receiving this because you are subscribed to this thread.
Reply to this email directly or view it on GitHub:
https://github.com/quicwg/base-drafts/issues/1826#issuecomment-426422980