My preference for flag over opcode has to do with "one frame to open streams" while I acknowledge the extra branch in the frame decoder -- though i think it might remove a branch in the session management code :) On Tue, May 28, 2013 at 1:02 PM, Roberto Peon <grmocg@gmail.com> wrote: > I think that the opcode approach is easier (I have a single branch instead > of nested branches), but don't think it is a big deal either way. > > -=R > > > On Tue, May 28, 2013 at 12:43 PM, William Chan (陈智昌) < > willchan@chromium.org> wrote: > >> I have feelings about this bike shed color, but I don't care enough to >> argue why mine is the best color ever. I am satisfied that there is a way >> to convey priority within the same frame as the headers. >> >> >> On Tue, May 28, 2013 at 11:12 AM, Jeff Pinner <jpinner@twitter.com>wrote: >> >>> At the time Roberto made the argument that the number of flags (8) was >>> more sparse than the number of frame types (256), but IIRC this was based >>> on the flags applying to all Control frames. At this point we have (at >>> least implicitly) decided that flags are frame-type specific (see PONG >>> flag), so I don't believe the argument is valid any more. >>> >>> >>> On Tue, May 28, 2013 at 10:50 AM, Martin Thomson < >>> martin.thomson@gmail.com> wrote: >>> >>>> I remember having almost this exact discussion in Tokyo. The only >>>> point that didn't come up this time was an argument Roberto made, >>>> namely: "A frame type is cheaper (fewer bits) than a flag." >>>> >>> >>> >> >
Received on Wednesday, 29 May 2013 01:25:57 UTC