Re: Thoughts on Message-ID generation

[Per Abrahamsen <abraham@dina.kvl.dk>]

Florian Weimer <fw@deneb.enyo.de> writes:

> Per Abrahamsen <abraham@dina.kvl.dk> writes:
> 
> > I do, anyway, since I use the message-id for sorting away duplicates.
> > Someone else mentioned threading, and mail-to-news gateways also rely
> > on the uniqueness of the mail message-id.
> 
> But message IDs are *not unique* for mail messages.  As soon as you
> crosspost a message to two mailing lists, there are at least two
> copies of the same message being distributed with different headers.

That's just semantics on when a copy of a message is still "the same"
message as the original.  If we count changes in headers, the receiver
will never get "the same" message as the sender send, because of the
"Received:" header.  The exact same happens with news, the message you
post to a server will never be "the same" you read on the server,
because the server changes the "Path:" header.

I believe a definition of sameness that would make two copies of a
messages send to two mailing lists not "the same message" would be
both counter-intuitive and useless.  In generel, the automatic changes
made at the transport layer (this includes mailing list software) does
not make the copy a new message.

A formal definition would probably say that the (mail or news)
injection agent should add a new unique message-id if not already
present, and no other automatic agents may change the message-id.
This will give us a message-id that is the same for most messages that
are intuitively the same, and useful for software purposes.

The intuition will break if the user manually add the same message-id
to messages that are intuitively different.  In this case we can from
a software and standard point of view assume the user know what he
does (i.e. the two messages _should_ be treated like duplicates), and
if not, that it is his own fault.