Re: Working Group Last Call The ORIGIN HTTP/2 Frame

On Wed, Sep 27, 2017 at 8:47 PM, Martin Thomson <martin.thomson@gmail.com>
wrote:

> On Thu, Sep 28, 2017 at 10:37 AM, Patrick McManus <pmcmanus@mozilla.com>
> wrote:
> > so is ORIGIN_SUPPORT the equivalent of sending an empty ORIGIN frame in
> the
> > previous draft?
> >
> > What's the benefit of additional moving pieces? It doesn't close the race
> > condition as the client doesn't need to wait for settings to start
> writing
> > requests..
>
> But the server HAS to send SETTINGS.  So at least the waiting is
> deterministic.


deterministic might be a little strong. The server has to send it as the
first TCP bytes, but it doesn't have to send it right away.. I guess I
could mumble something about congestion controllers that measure in packets
instead of bytes giving a motivation for such behavior. I guess in general
I'm deeply skeptical of any algorithm that expects waiting.

I'm not going to object to the setting - it just seems it doesn't really
address the fact that the client is going to see both 7540 rules and ORIGIN
rules at some point on the same connection so there's not a lot of point to
it imo.


>   And in TLS 1.3, the server speaks first, so there
> shouldn't be any waiting around there.  I agree that this is not ideal
> for TLS 1.2, but you could simply decide not to wait and suffer the
> consequences if you are wrong.
>