Unique? Message-Id generation by Gnus

Unique? Message-Id generation by Gnus

[Matt Armstrong]

Why does Gnus add a Message-Id header instead of letting the news/mail
server do it?

With people setting user-mail-address to the base domain name of their
company/ISP/mail-forwarder, it is possible to get a lot of
<blahblah.fsf@domain.com> message ids even though the logic that
generates the "blahblah" part doesn't necessarily guarantee
uniqueness.
E.g. <blahblah.fsf@bigfoot.com>

It just seems like the news/mail servers are more centralized, and so
could guarantee uniqueness more effectively.

I realize this may be a theoretical and not a practical problem -- I'm
just imagining the day _everybody_ uses Gnus.  ;-)

-- 
matta

Re: Unique? Message-Id generation by Gnus

[Matt Armstrong]

Stainless Steel Rat <ratinox@peorth.gweep.net> writes:

> -----BEGIN PGP SIGNED MESSAGE-----
> 
> >>>>> "MA" == Matt Armstrong <matta@geoworks.com> writes:
> 
> MA> Why does Gnus add a Message-Id header instead of letting the news/mail
> MA> server do it?
> 
> Because it is strongly recomended that Message-Id strings be
> generated by the MUA, rather than the MTA.  The reason...<snip>

I'll accept that at face value.


> MA> With people setting user-mail-address to the base domain name of
> MA> their company/ISP/mail-forwarder, it is possible to get a lot of
> MA> <blahblah.fsf@domain.com> message ids even though the logic that
> MA> generates the "blahblah" part doesn't necessarily guarantee
> MA> uniqueness.
> 
> Actually, it does.  The "blahblah" part is a hash (sort of) derrived
> from your login name and the current system time.  As it is
> practically impossible for you (that is, your login) to generate two
> distinct messages at *exactly* the same time, the hash is
> effectively guaranteed to be unique.

It actually defaults to using the UID, not the text of the user's
login.  This doesn't work under NT -- message.el/message-unique-id
should probably add 'windows-nt to the list of system types for which
the UID is not a good thing to use.

I realize now that this is really a worry of mine over
message-unique-id's use of user-mail-address to get the domain name of
the machine.  Say all my messages are of the form:

From: matt_armstrong@bigfoot.com
Sender: matta@quark.geoworks.com

The Message-Id should be generated from what is in the Sender: field,
not the user-mail-address.  My UID isn't guaranteed to be unique under
the bigfoot.com domain.


-- 
matta

Re: Unique? Message-Id generation by Gnus

[Hallvard B Furuseth]

Matt Armstrong <matta@geoworks.com> writes:

> a worry of mine over message-unique-id's use of user-mail-address to
> get the domain name of the machine.

It does that only as a backup.

> The Message-Id should be generated from what is in the Sender: field,

It is, indirectly.  The sender field is generated from (system-name).
See (message-make-sender) in message.el.  (message-make-fqdn), which
generates the domain part of the message ID, tries:
    first  (system-name)      if that has at least 2 components,
    then   mail-host-address  if that has at least 2 components,
    then   user-mail-address' domain.

> not the user-mail-address.  My UID isn't guaranteed to be unique under
> the bigfoot.com domain.

Note that your message's ID is <uvhtixv1b.fsf@geoworks.com>.  It does
not fall through to using mail-user-address.

-- 
Hallvard