Re: [quicwg/base-drafts] Consider a new stream type to act like MAX_PUSH_ID (#2418)

[Mike Bishop <[e-mail]>]

@LPardue, QPACK encoder data was originally in HEADERS frames on the control stream; the reason for moving to unidirectional streams was the insight that the blocks of updates weren't themselves significant; it was really the state of individual updates that needed to be tracked.  Because they were an ordered sequence with no semantics to the division into frames, we decided that moving to a stream would be simpler to process.

In this case, the semantics are a little different -- you really only ever care about the most recent value.  In the transport and in H2, we carry such things in frames because that's what we have.  An application on top of QUIC doesn't really have an equivalent mechanism until DATAGRAM exists.  Putting the current frame in a DATAGRAM would make the most sense.  In the meantime, @martinthomson is correct that only deltas are needed, and that's actually true even with the frame -- they're on the control stream in order, so a delta would be sufficient.  The current design does rather straddle the two options, delivering full updates in order.  The main argument for keeping the current design seems like avoiding the need to change the structure when we are able to adopt DATAGRAM.

I don't think moving it to a unidirectional stream makes it any more extension-like than defining a frame type -- extensions can do either.

-- 
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/2418#issuecomment-460813704