Re: HTTP/2 vs. proxies ?

On 24 June 2014 03:34, Mike Bishop <Michael.Bishop@microsoft.com> wrote:

> I'll note that while changing the semantics of your setting to supplement,
> rather than replace, previously-sent values of the setting is perfectly
> legal (it's an extension, so you can change semantics if you need to), I
> really don't like having one setting that behaves completely differently
> than all of the others, and it doesn't seem critical to your goals.  That's
> why I made the setting a bitmask when I pulled your proposal into my
> draft's appendix.  However, a bitmask loses the ability to specify weights,
> which is needed for a strict replacement of HTTP/1.1 T-E.
>
> Another option might be an (informative) frame type which gives
> weightings, but given the relative scarcity of frame types versus settings
> in the accepted version, I understand the desire to keep this in the
> setting space.
>
>
Another alternative is one setting per encoding, although that's a bit
silly. (Come to think of it, that was how it ended up in the HTTP2 draft
before being yanked -- one setting for GZIP).

In any case I think it's appropriate to make an effort, at least for the
first few extension-type discussions, to try and set the tone right. If we
think there's more value in promoting the idea that settings should keep
the "replace" semantic, I'll try to work my draft to fit.

​Incidentally, this still isn't a strict replacement for T-E because I
can't stack encodings, and because 256 rank levels is fewer than 1001, and
I have no support for other parameters, etc. It might be worth me adding an
explicit acknowledgement of the differences.


-- 
  Matthew Kerwin
  http://matthew.kerwin.net.au/