Re: Another CONTINUATION Proposal

> My understanding is that the problem with PUSH_PROMISE modifying the stream state isn't just that it's changing the state of a stream, but also because it's changing the hpack state. In other words, moving the headers to a different frame type doesn't solve the issue that a frame carrying hpack header blocks changes shared state.

Correct, we'd have to remove the header block form PUSH_PROMISE so
that it only changed stream state. Push streams would now look like:

PUSH_PROMISE 1*HEADERS SYN_STREAM 1*HEADERS 0*DATA (DATA 0*HEADERS)

where the first set of HEADERS frames contained the request header
fields, the second set the response header fields, and the third
optional set the response triailers

Received on Tuesday, 8 July 2014 21:51:07 UTC