Magic

[mnot]

We need some magic that clearly indicates that HTTP/2 is being spoken on the wire. It should fast fail when sent to a reasonable set of HTTP/1.1 servers.

[mnot]

In the draft, so closing (although we might tweak the string as time goes on).

[phluid61]

If the goal of the connection header magic string is to emulate a HTTP/1.x request in a way that is both as transparent as possible (can be parsed as HTTP/1.x) and unambiguously differentiated (could NOT be understood as HTTP/1.x), I understand the criteria to be:

When interpreted as a HTTP/1.x request:

• the HTTP-Version must be "HTTP/2.0";
• the Request-URI must be "*", since this message is not resource-specific;
• the Method should not coincide with any extant HTTP/1.x methods, since a poorly-implemented device might, for example, only inspect the minor part of the HTTP-Version and misinterpret the header as a HTTP/1.0 request;
• the Method should be exactly three or four bytes in length, since there are known implementations in the wild that have this expectation.

To fit in a 24-byte structure, using the given pattern, we would have to use either a three-byte Method and a two-byte entity (before the final "\r\n\r\n"), or a four-byte Method and a one-byte entity. I think 3-2 is better, aesthetically.

I also think it's better to avoid encoding a metasyntactic variable into the standard. "FOO" is universally used as a placeholder for user-supplied data, and it may be confusing for the spec to require implementers to send or expect the literal value "FOO". Additionally there is a legitimate chance that there are implementations in the wild that actually use "FOO" as a method, e.g. for developmental purposes.

I propose "CON" and "go" as the Method and message entity parts. I.e. 434f4e202a20485454502f322e300d0a0d0a676f0d0a0d0a or "CON * HTTP/2.0\r\n\r\ngo\r\n\r\n"

[mnot]

Hmm. "CON" is a prefix of "CONNECT"; not saying that implementations will be matching prefixes, but I get a bit worried.

On-list I proposed "STA" and "RT".

[phluid61]

STA works for me; I can't think of any reasons it might go wrong.

(I would die happy if it were "HELO" and ".")