Re: http/2 & hpack protocol review

On 5 May 2014 16:43, Julian Reschke <julian.reschke@gmx.de> wrote:
> If HTTP/2 is inappropriate for a case where HTTP/1.1, that's a problem we
> need to solve. This WG is chartered to define a new protocol version that
> can *replace* HTTP/1.1.
>
> So I'd really like to see something more concrete than "it's too complex".
> After all, correctly parsing text messages is complex as well, and that part
> is gone in HTTP/2.
>
> Best regards, Julian

I think that ship has sailed.

It's been my understanding and expectation that HTTP/2 is simply not
going to replace HTTP/1.1 wholly and completely. The complexity
increase in terms of quantity of code in any language that has basic
text-parsing functionality built-in (i.e. not C) from HTTP/1.1 to
HTTP/2 is so astronomical that it's staggering. Logic is added for
maintaining multiple sets of independent state (connection state,
stream state, per-stream header compression state, per frame
compression state). There are plenty of excellent gains to be achieved
by this increase in complexity, but we shouldn't be pretending that
HTTP/2 can be used everywhere HTTP/1.1 can.

HTTP/2 is also harder to debug than HTTP/1.1, even if we exclude the
fact that most HTTP/2 connections will be over TLS-with-PFS and will
therefore be painful to intercept and trace.

Certainly, in the world of client side HTTP libraries we're expecting
that HTTP/2 will not supplant HTTP/1.1. We're not even sure that
HTTP/2 will become the default.

Received on Tuesday, 6 May 2014 09:09:49 UTC