- From: John Sullivan <jsullivan@velocix.com>
- Date: 2012年5月15日 12:15:02 +0100
- To: Julian Reschke <julian.reschke@gmx.de>
- CC: Willy Tarreau <w@1wt.eu>, Andreas Petersson <andreas@sbin.se>, Amos Jeffries <squid3@treenet.co.nz>, <ietf-http-wg@w3.org>
- Message-ID: <4FB23AB6.4070701@velocix.com>
Julian Reschke wrote: > The HTTP community has been inventing new microsyntaxes for many years, > and as far as I can tell, most header field parsers out there are broken > beyond belief. We need less of them, even if this means that a few edge > cases will be more verbose than necessary. This. Each new syntax risks getting it wrong (even if in this case it ought to be fairly safe) - just look at the horrible horrible mess that is the history of Set-Cookie. In addition to the headers I pointed out yesterday that use close variations on the theme, the definition of Expect appears to be almost *exactly* this syntax - except that it makes the value part ( "=" ( token / quoted-string ) ) optional. This allows for the sending of value-less flags without having to use the somewhat grotty flag="" construct. Perhaps adopting that syntax verbatim would be a good idea? Note also httpbis-p2 S3.1, "Considerations for Creating Header Fields", particularly: Many header fields use a format including (case-insensitively) named parameters (for instance, Content-Type, defined in Section 6.8 of [Part3]). Allowing both unquoted (token) and quoted (quoted-string) syntax for the parameter value enables recipients to use existing parser components. John --
Received on Tuesday, 15 May 2012 11:15:48 UTC