Re: msg-id

[Frank Ellermann <nobody@xxxxxxxxxxxxxxxxx>]

Charles Lindsey wrote:

> That implementation is broken

Indeed, fortunately I found an ISP allowing me to fix it... :-)

> changing our draft cannot fix it.

True, but if you'd say "domain" instead of "id-right" it's more
obvious.  Address literals are already on the border to utter
dubious, with different algorithms for "id-left", different
time zones, and dynamic IPs.  But <whatever@invalid> isn't only
a bad idea, it's broken.

>> Of course the UA is broken, but it's another argument for
>> RHS = domain as specified in 2821.

> Why? You are asking for further changes. Others are asking
> for the minimal change from RFC 2822. Which am I supposed to
> do?

Let's just find a minimal _working_ model for Message-IDs.  In
RfC 2822 you were not interested in these details, you only
wanted some simple syntax covering wild and wonderful things
like mail, s-o-1036, and X.400.

Here and today we know it better, we want something working in
Usenet, in news URLs, in cancel messages, and as many existing
UAs as possible.  IMHO that disqualifies <whatever@invalid>,
and it also doesn't allow NO-WS-CTL.

For the purists like Russ and Bruce the "working model" is most
probably RfC 1036 <unique@full_domain_name>.  For the heretics
like me the "working model" is s-o-1036, essentially the same:

|    message-id          = "<" local-part "@" domain ">"

RfC 2822 is too broad, they wanted a "working model" covering
obsolete and broken software for mail, where Message-IDs aren't
essential.  I'm only guessing why RfC 2822 got it wrong, you
were there:  Why did they replace the old concept of "like an
addr-spec, only unique" by a new "id-left @ id-right" ?

The RfC 2822 "addr-spec" is still fine, it mentions "domain":

| addr-spec  =       local-part "@" domain

But the RfC 2822 Message-ID invents an obscure "id-right" out
of thin air:

| msg-id     =       [CFWS] "<" id-left "@" id-right ">" [CFWS]

I just don't know why you did this in 2822, but it was wrong.
It's incompatible with 1036 and s-o-1036, we can't use it here.

We need the good old "<" unique "@" domain ">" minus RfC 2822
oddities like obs-domain or CFWS.  The RfC 2821 domain syntax
makes sense, it even includes address literals, just forget the
1 in its domain = (subdomain 1*("." subdomain)) / address-literal

Or copy the s-o-1036 definition of domain adding the address
literals as defined in 2821:

| domain        = unquoted-word *( "." unquoted-word )
| unquoted-word = 1*unquoted-char
| unquoted-char = <ASCII printable character except !()<>@,;:\".[]>

It's no big difference from 2822 or your text, and the syntax
still allows an insane <whatever@invalid>, but at least the
intention "domain" is clear and not obscured by an "id-right".

 [dubious <whatever@news-fe-01> and ...!news-fe-01!...]
> it is quite likely to be unique to that site, given that it
> was considered suitable as a path-identity.

It was more likely an error in the configuration of this news
server.

> It is not necessarily wrong.

Show me the UUCP world map entry for news-fe-01, and I consider
it :-)  That's what I said in the de.admin,news.misc.discussion
about it, all others said "unconditionally invalid Message-ID".

> A lot of news (and even some email) is still transported by
> UUCP and other protocols.

s-o-1036 mumbles something about using .uucp in this case, and
to stop this nonsense a.s.a.p.  That was 1994, and the dubious
"domain" and path entry was news-fe-01 (not news-fe-01.uucp) in
2004.

>    <12345678@xxxxxxxxxxxxxxx>
>    <12345678@xxxxxxxxxxxxxxx>

> They are DIFFERENT message-ids. Yes?

Yes.  And I would use the same reasoning for id-left:

     <12345678@xxxxxxxxxxxxxxx>
     <"12345678"@foo.example.com>

Also different.  I'm a KISS extremist.  Some gateways trying
to "fix" the second version while others don't would be hell.

 [NO-WS-CTL in Message-ID]
> persuade someone else on this list to agree to the further
> restrictions you are asking for

If Russ really is a 1036 purist, and Henry still is a s-o-1036
heretic, then I simply assume that they also hate any NO-WS-CTL
in Message-IDs.  Only Bruce loves it, but I'm not sure why.

You have already said that you hate it, because NO-WS-CTL is a
PITA for the news URL.
                      Bye, Frank