Re: #536: clarify extensibility for :pseudo header fields

In message <961E96EE-0554-48AD-907C-FD3055584CE2@gmail.com>, David Krauss writes:
>
>On 2014=9607=9602, at 2:07 PM, Mark Nottingham <mnot@mnot.net> wrote:
>
>> Now I'm confused. We're currently talking about whether extra :headers =
>are a hard error when http/2 is in use. What's an ALPN-basis API?

I always assumed that the :headers only were HTTP/2s internal way of
representing the first line of HTTP/1 messages on the wire, not a way
to add an entirely new namespace to HTTP headers/metadata ?

If some extensions need namespaces in HTTP headers, we should add
namespaces which will (or can be made to) work in both HTTP/1,
HTTP/2 and any future HTTP versions, not latch onto an implementation
detail in one particular version of HTTP.

Traditionally namespaces delimiters have been ':' but that will certainly
break too many HTTP/1 implementations to be feasible, but there are plenty
of ASCII characters available that could be used: '=', '/' and so on.

In other words:  Please limit :headers to the ones in 8.1.3.1 and 8.1.3.2
and make attempts to smuggle any other :header through a hard protocol error.

-- 
Poul-Henning Kamp       | UNIX since Zilog Zeus 3.20
phk@FreeBSD.ORG         | TCP/IP since RFC 956
FreeBSD committer       | BSD since 4.3-tahoe    
Never attribute to malice what can adequately be explained by incompetence.

Received on Wednesday, 2 July 2014 06:30:40 UTC