- From: Frank Ellermann <nobody@xyzzy.claranet.de>
- Date: Tue, 21 Dec 2004 00:07:07 +0100
- To: uri@w3.org
Charles Lindsey wrote:
> The trouble is that there are just too many definitions of
> message-id around the place.
Different aspects of the same real thing, all describing the
minimum they could get away with for their special purposes.
> There is going to be strong pressure to keep to RFC 2822
> or at least to a subset of it (which the RFC 1036 definition
> is not).
All 1036 Message-IDs are 2822 Message-IDs, "printable ASCII"
is a proper subset of "ASCII minus white space".
> The alternative is to give a very loose definition, on the
> grounds that URLs are supposed to contain only whatever has
> been used in some existing news article (and the agent that
> generated that article can worry about which standard it
> conformed to). So any string of characters with an '@'
> in the middle is good enough.
That's a good idea, because it's what you really want for the
news URL: If somebody created a Message-ID with NO-WS-CTL,
and a server accepted this, then maybe that's broken, but no
problem of the news URL scheme.
[quoting NNTP]
>| For the purposes of this specification, message-ids are
>| opaque strings
That's fine, garbage in, garbage out, and the same garbage is
the same message and v.v.
>| MUST begin with "<" and end with ">", and MUST NOT contain
>| the latter except at the end.
Not good enough for the news URL, we need the "@", otherwise
it could be a newsgroup name.
> A message-id MUST be between 3 and 250 octets in length.
The minimum is <@> ? LOL, no problem.
>| A message-id MUST NOT contain octets other than printable
>| US-ASCII characters.
See ? No NO-WS-CTL here. That's a sane definition for servers
and user-agents. For UseFor we need a "domain" as RHS (because
that's required for algorithms trying to create new unique IDs),
and the "@", because otherwise RHS makes no sense (right hand
side of what without an "@" ?)
For the news URL we need only an "@" (in theory more than one
is fine). Putting it all together: printable US-ASCII minus
">" and at least one "@" is good enough for the news URL.
>| Other specifications may define two different sequences as
>| being equal because they are putting an interpretation on
>| particular characters.
These other specifications are IMHO erroneous.
>| Note that RFC 1036 [RFC1036] never treats two different
>| strings as being identical. Its draft successor restricts
>| the syntax of message-ids so that, whenever RFC 2822 would
>| treat two strings as equivalent, only one of them is valid
Talking about s-o-1036 probably, but we could use the same
recipe. You already did it, but somehow you forgot to exclude
NO-WS-CTL. But these drafts are yet no RfCs, you can't quote
them in the new memo about a news URL. But the solution of a
"loose definition" is okay, printable with "@" and without ">".
> I just noticed that NNTP will not accept Non-WS-Controls
A fatal error in RfC 2822, don't mention it for the news URL.
> Anyway, the uri@w3c.org list is the proper place to continue
> this discussion.
Okay, I'll post my answer (t)here. Do we need a note about any
Message-ID starting with "//" like a server name ? Something
like "'/' MUST be escaped as %2F if (insert conditions)" ? And
the opposite, '@' MUST not be esacaped, or is it the job of the
user agent to see a %40 ?
Bye, Frank
Received on Monday, 20 December 2004 23:34:33 UTC