`user-mail-address' for message-ids

`user-mail-address' for message-ids

[Jason R Mastaler]

It would be nice if `message-make-message-id' would use
`user-mail-address' when creating message-ids so that they would match
the user's From address.  Or, at least this should be an option.

Right now, `message-make-message-id' calls `message-make-fqdn' for the
last part of the message-id (after @).  `message-make-fqdn' uses
`user-mail-address' only as a last resort (after consulting
`system-name' and `mail-host-address').

Any objections to moving `user-mail-address' before the other two?  
If so, how about making this optional through a variable setting or
something?

Re: `user-mail-address' for message-ids

[Paul Jarc]

Jason R Mastaler <jason@mastaler.com> writes:
> It would be nice if `message-make-message-id' would use
> `user-mail-address' when creating message-ids so that they would match
> the user's From address.  Or, at least this should be an option.

Why do you care whether it matches the From address?

> `message-make-fqdn' uses `user-mail-address' only as a last resort
> (after consulting `system-name' and `mail-host-address').
> 
> Any objections to moving `user-mail-address' before the other two?  

Message-ID is not meant for human consumption, so you shouldn't care
if it's ugly.  The single most important attribute for it is
uniqueness; anything that gets in the way of that is bad.  If two
Message-IDs are generated on different hosts, but they use the same
fqdn, they might come out identical.  That would be bad.

> If so, how about making this optional through a variable setting or
> something?

What would be the motivation for doing this?


paul

Re: `user-mail-address' for message-ids

[Kai Großjohann]

[[ Paul wants to know why one would want to frob the message id. ]]

Many MUAs put the message id of all received messages into the
References header.  So if I put the token `xyzzy' into the message id
of messages regarding the foo project, then I can recognize incoming
messages regarding foo.

Very useful for mail splitting.

Note that _adding_ the `xyzzy' token into some message ids does not
necessarily make them less unique.

That said, I don't know what was Jason's motivation.

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Paul Jarc]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
> [[ Paul wants to know why one would want to frob the message id. ]]

The domain part, in particular.

> Many MUAs put the message id of all received messages into the
> References header.  So if I put the token `xyzzy' into the message id
> of messages regarding the foo project, then I can recognize incoming
> messages regarding foo.
> 
> Very useful for mail splitting.

Hm, yeah, I guess it would be.  But it should be done prior to the "@"
in order to preserve uniqueness.  Host foo.bar.baz doesn't know
offhand whether there exists a host xyzzy.foo.bar.baz.


paul

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

prj@po.cwru.edu (Paul Jarc) writes:

> Why do you care whether it matches the From address?

Any number of reasons Paul, try using your imagination.  Aesthetics,
to conceal the (system-name) of one's machine, etc..

> Message-ID is not meant for human consumption, so you shouldn't care
> if it's ugly.  

Fiddlesticks.  Message-IDs are referenced all the time by us humans.

> If two Message-IDs are generated on different hosts, but they use
> the same fqdn, they might come out identical.  That would be bad.

I'm not advocating anything new here, just a re-order within
`message-make-fqdn'.  In fact, machines whose (system-name) returns
just "hostname" instead of "hostname.domain.dom" such on Solaris will
end up using `user-mail-address' for the message-id anyway.  I'm
simply requesting this behavior for all machines.  Or, at least the
option to choose it.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 23 Feb 2001, Paul Jarc wrote:

> Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
>> [[ Paul wants to know why one would want to frob the message id. ]]
> 
> The domain part, in particular.

Oh.  I missed something, AGAIN.  Argh.  Must be too tired.

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Paul Jarc]

"Jason R. Mastaler" <jason@mastaler.com> writes:
> prj@po.cwru.edu (Paul Jarc) writes:
> > Why do you care whether it matches the From address?
> 
> Any number of reasons Paul, try using your imagination.  Aesthetics,
> to conceal the (system-name) of one's machine, etc..

Functionality trumps aesthetics.  If you want anonymity, use an
anonymous remailer.  (Message-ID isn't the only place your hostname
would show up in the header.)

> > Message-ID is not meant for human consumption, so you shouldn't care
> > if it's ugly.  
> 
> Fiddlesticks.  Message-IDs are referenced all the time by us humans.

But their primary purpose is to be a universally unique identifier for
messages.  Using the domain from user-mail-address would subvert that
purpose.

> > If two Message-IDs are generated on different hosts, but they use
> > the same fqdn, they might come out identical.  That would be bad.
> 
> I'm not advocating anything new here, just a re-order within
> `message-make-fqdn'.

But that *is* something new.  It would make collisions significantly
more likely.  With collisions, we get problems like:
- News messages would get rejected for no good reason.
- Threaded mailing list archives would break.
- Threading in general would break.

> In fact, machines whose (system-name) returns just "hostname"
> instead of "hostname.domain.dom" such on Solaris will end up using
> `user-mail-address' for the message-id anyway.

Yes, but that's just a last resort.  Message-ID is intended to
uniquely identify a single message in all the world, for all time.
It's clearly not feasible for all hosts to collaborate when generating
Message-IDs, so to make collisions as unlikely as possible, we need a
unique identification of the host generating the Message-ID in the
Message-ID.  Do you think having a unique identifier is not
worthwhile?

> I'm simply requesting this behavior for all machines.  Or, at least
> the option to choose it.

You already know how to make this change for your own Gnus.  Why do
you want to break mine?


paul

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

prj@po.cwru.edu (Paul Jarc) writes:

> Functionality trumps aesthetics.  If you want anonymity, use an
> anonymous remailer.  (Message-ID isn't the only place your hostname
> would show up in the header.)

The point is that just because you may not find this functionality
useful does not mean others do/will not.  And BTW, it is trivial to
make Message-ID the only place your hostname would show up in the
header.

> But their primary purpose is to be a universally unique identifier
> for messages.  Using the domain from user-mail-address would subvert
> that purpose.

No.  Many MUAs use the user's chosen e-mail address (such as Mutt) in
the message-id instead of the system's fqdn.  I say that if the 1st
half of the message-id (before the "@") doesn't guarantee uniqueness,
then we need to rethink the algorithm for `message-make-message-id'.

> You already know how to make this change for your own Gnus.  Why do
> you want to break mine?

Because if done correctly, it won't "break" anything, and it means I
don't have to maintain my own version of `message-make-fqdn' until the
end of days.

Re: `user-mail-address' for message-ids

[Paul Jarc]

"Jason R. Mastaler" <jason@mastaler.com> writes:
> The point is that just because you may not find this functionality
> useful does not mean others do/will not. 

It's more than that: I find it *counter* to usefulness.

> And BTW, it is trivial to make Message-ID the only place your
> hostname would show up in the header.

How are you going to keep it out of the Received field that gets added
*after* the message leaves your computer?  The next SMTP recipient is
typically going to add a Received field with an indication of where
the message came from, and it's not going to trust the sender to
provide that indication; it'll use DNS instead.

> > But their primary purpose is to be a universally unique identifier
> > for messages.  Using the domain from user-mail-address would subvert
> > that purpose.
> 
> No.  Many MUAs use the user's chosen e-mail address (such as Mutt) in
> the message-id instead of the system's fqdn.

I don't know which statement you were trying to refute, but:
- Message-ID's purpose is indeed as I stated.  See RFC822, 4.6.1.
- The behavior you describe does indeed subvert that purpose.  (It
  doesn't guarantee failure, but it encourages it.)

> I say that if the 1st half of the message-id (before the "@")
> doesn't guarantee uniqueness, then we need to rethink the algorithm
> for `message-make-message-id'.

The first part only needs to be unique *on that host*, and
message-make-message-id does a good job of satisfying that
requirement.  But the Message-ID as a whole must be unique
*universally*.  To ensure that, you need to include the host's
identification, and since different hosts may use different methods to
generate the first part, the host's identification must be included in
a universally uniform way, in order to avoid collisions.

> > You already know how to make this change for your own Gnus.  Why do
> > you want to break mine?
> 
> Because if done correctly, it won't "break" anything,

So if hosts don't uniformly embed their own identification into
Message-IDs, and they don't coordinate Message-ID generation, how
would they guarantee uniqueness?  How could it be done "correctly"?


paul

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

If anyone has something constructive and/or useful to add to this
discussion, I'm all ears.  Thanks.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 23 Feb 2001, Jason R. Mastaler wrote:

> No.  Many MUAs use the user's chosen e-mail address (such as Mutt)
> in the message-id instead of the system's fqdn.  I say that if the
> 1st half of the message-id (before the "@") doesn't guarantee
> uniqueness, then we need to rethink the algorithm for
> `message-make-message-id'.

No no no.  Consider people with a gnu.org address, or acm.org, or
hotmail.com.  People who use the default Linux installation will
probably be using an account with the user id 100 (or was it 1000?),
so the user id part that goes into the local part of the message id
doesn't help, either.

You MUST use a rhs part for the msg id which uniquely identifies the
computer you're sitting at.

Savvy ISPs provide a rhs for their dialup customers -- the rhs doesn't
have to be a valid hostname, so something like
<login>.dialup.<provider>.net would do nicely.

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On Fri, 23 Feb 2001, Jason R. Mastaler wrote:

> It would be nice if `message-make-message-id' would use
> `user-mail-address' when creating message-ids so that they would
> match the user's From address.  Or, at least this should be an
> option.

The standard says that the hostname part of a message id must uniquely
identify the host.  Since the hostname part of user-mail-address does
often NOT identify a host (consider @gnu.org, @acm.org, @hotmail.com
mail addresses), the hostname part of user-mail-address is not a good
choice for the right hand side of a message id.

I don't think it is useful for Gnus to provide a default behavior
which is against RFC 822 (or wherever it is).

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Simon Josefsson]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:

> No no no.  Consider people with a gnu.org address, or acm.org, or
> hotmail.com.  People who use the default Linux installation will
> probably be using an account with the user id 100 (or was it 1000?),
> so the user id part that goes into the local part of the message id
> doesn't help, either.

People who use default Linux installations (at least RedHat) and DHCP
will also get a domain part of "localhost.localdomain".

# emacs -q -batch -eval '(princ (concat (system-name) "\n"))'
localhost.localdomain
# hostname -f
sjosefsson-lap

So the only "random" part that make Message-Id unique are time.  Not
good, I think.

(XEmacs 21.2 is the only emacsen on my system that report correct hostname.)

Re: `user-mail-address' for message-ids

[NAGY Andras]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:

> > No.  Many MUAs use the user's chosen e-mail address (such as Mutt)
> > in the message-id instead of the system's fqdn.  I say that if the
> > 1st half of the message-id (before the "@") doesn't guarantee
> > uniqueness, then we need to rethink the algorithm for
> > `message-make-message-id'.
> 
> No no no.  Consider people with a gnu.org address, or acm.org, or
> hotmail.com.  People who use the default Linux installation will
> probably be using an account with the user id 100 (or was it 1000?),
> so the user id part that goes into the local part of the message id
> doesn't help, either.

I think he did not mean the numerical user id, but the local part of
the user's email address.  Say, if joe@hotmail.com uses message ids
like random-stuff-based-on-time.joe@hotmail.com, it will be likely
unique.  (If Joe uses only a single MUA at a given time.)


Andras

Re: `user-mail-address' for message-ids

[Per Abrahamsen]

Jason R Mastaler <jason@mastaler.com> writes:

> Any objections to moving `user-mail-address' before the other two?  

Yes.  It would increase the risk of duplicate message ids, i.e. all
the users of all the machines in our department share the
'dina.kvl.dk' mail host name, but I'm the only one using ssv2.

> If so, how about making this optional through a variable setting or
> something?

The *only* important property of the message-id is that it is unique,
so I do not support any option that would counter that.  Especially
since some misguided people might turn it on in order to get
"prettier" message-ids.

Re: `user-mail-address' for message-ids

[ShengHuo ZHU]

Simon Josefsson <simon@josefsson.org> writes:

> Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
> 
> > No no no.  Consider people with a gnu.org address, or acm.org, or
> > hotmail.com.  People who use the default Linux installation will
> > probably be using an account with the user id 100 (or was it 1000?),
> > so the user id part that goes into the local part of the message id
> > doesn't help, either.
> 
> People who use default Linux installations (at least RedHat) and DHCP
> will also get a domain part of "localhost.localdomain".
> 
> # emacs -q -batch -eval '(princ (concat (system-name) "\n"))'
> localhost.localdomain
> # hostname -f
> sjosefsson-lap
> 
> So the only "random" part that make Message-Id unique are time.  Not
> good, I think.
> 
> (XEmacs 21.2 is the only emacsen on my system that report correct hostname.)

Since message-make-fqdn is not always reliable, should we add a
customizable variable for this?

ShengHuo

Re: `user-mail-address' for message-ids

[Alan Shutko]

ShengHuo ZHU <zsh@cs.rochester.edu> writes:

> Since message-make-fqdn is not always reliable, should we add a
> customizable variable for this?

What's wrong with system-name?

-- 
Alan Shutko <ats@acm.org> - In a variety of flavors!
Being popular is important.  Otherwise people might not like you.

Re: `user-mail-address' for message-ids

[Paul Jarc]

ShengHuo ZHU <zsh@cs.rochester.edu> writes:
> Simon Josefsson <simon@josefsson.org> writes:
> > People who use default Linux installations (at least RedHat) and DHCP
> > will also get a domain part of "localhost.localdomain".

The DHCP client can be configured to call sethostname() and
setdomainname().  If it doesn't do so, and if (system-name)
consequently reports the wrong name, then that's a problem outside
Gnus, which should be solved outside Gnus.

> > # emacs -q -batch -eval '(princ (concat (system-name) "\n"))'
> > localhost.localdomain
> > # hostname -f
> > sjosefsson-lap

How is emacs finding the hostname, in this case?

> Since message-make-fqdn is not always reliable, should we add a
> customizable variable for this?

I think it would do more harm than good.


paul

Re: `user-mail-address' for message-ids

[Denys Duchier]

This discussion reminds me that I have a problem with
message-make-sender which I can only fix with the following hack:

(eval-after-load "message"
  '(fset 'message-make-sender
	 (symbol-function 'message-make-address)))

The problem is that our mail gateway policy forbids the use machine
specific mail addresses.  All addresses must use an official mail
domain.  Usually it doesn't matter, but if there is a problem they'll
just drop your email down the bit bucket without sending back a bounce
message.

It would be really nice if this could be made customizable.

Cheers,

-- 
Dr. Denys Duchier			Denys.Duchier@ps.uni-sb.de
Forschungsbereich Programmiersysteme	(Programming Systems Lab)
Universitaet des Saarlandes, Geb. 45	http://www.ps.uni-sb.de/~duchier
Postfach 15 11 50			Phone: +49 681 302 5618
66041 Saarbruecken, Germany		Fax:   +49 681 302 5615

Sender again again (was: Re: `user-mail-address' for message-ids)

[Per Abrahamsen]

Denys Duchier <Denys.Duchier@ps.uni-sb.de> writes:

> It would be really nice if this could be made customizable.

Gnus should not generate sender by default for mail.

For news, the problem is that sender is sometimes seen as a poor mans
autorization, which is why it is not customizable.  I think it should
just be dropped, but that would make it harder to cancel your own
messages if you are using gnus-posting-styles.

Re: Sender again again (was: Re: `user-mail-address' for message-ids)

[Florian Weimer]

Per Abrahamsen <abraham@dina.kvl.dk> writes:

> For news, the problem is that sender is sometimes seen as a poor mans
> autorization, which is why it is not customizable.  I think it should
> just be dropped, but that would make it harder to cancel your own
> messages if you are using gnus-posting-styles.

I think I'm going to implement the CANLOCK draft for Gnus.  This
solves the problem of recognizing your own articles.

Re: Sender again again (was: Re: `user-mail-address' for message-ids)

[Simon Josefsson]

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

> > For news, the problem is that sender is sometimes seen as a poor mans
> > autorization, which is why it is not customizable.  I think it should
> > just be dropped, but that would make it harder to cancel your own
> > messages if you are using gnus-posting-styles.
> 
> I think I'm going to implement the CANLOCK draft for Gnus.  This
> solves the problem of recognizing your own articles.

Checkout contrib/canlock.el first. :-)

Re: Sender again again (was: Re: `user-mail-address' for message-ids)

[Florian Weimer]

Simon Josefsson <simon@josefsson.org> writes:

> > I think I'm going to implement the CANLOCK draft for Gnus.  This
> > solves the problem of recognizing your own articles.
> 
> Checkout contrib/canlock.el first. :-)

The copyright status of this file seems to be quite unclear.  I doubt
we can incorporate it into Gnus.

Cancel-Locks (was: Re: Sender again again (was: Re: `user-mail-address' for message-ids))

[Per Abrahamsen]

How should the UI be?

I think it would be most useful as a mandatory feature, i.e. Gnus will
always add a cancel lock, and the cancel code will always check the
lock for permission.

Re: Cancel-Locks

[Florian Weimer]

Per Abrahamsen <abraham@dina.kvl.dk> writes:

> How should the UI be?
> 
> I think it would be most useful as a mandatory feature, i.e. Gnus will
> always add a cancel lock, and the cancel code will always check the
> lock for permission.

Yes, I think that's the way to go. (Of course, an undocumented
variable will exist which permits canceling in all cases. ;-) In
addition, the cancel message will be editable, much like a Supersede:
message.

We need to generate a cancel key for each posted article.  I think the
best approach is to generate a secret string once (for example, a hash
over the output of the date, ps, and ls ~ commands, or something like
that, if /dev/[u]random is not available).  A cancel key is the SHA-1
hash of concatenation of this secret string and the message ID of the
article in question.  This way, we don't have to keep a database of
cancel keys, and this system is safe as long as SHA-1 meets its design
goals and the secret string is kept secret.

Comments?

Re: Cancel-Locks

[Per Abrahamsen]

Florian Weimer <fw@deneb.enyo.de> writes:
> (Of course, an undocumented
> variable will exist which permits canceling in all cases. ;-)

I think that should be reserved to those who can hack the code.

> In addition, the cancel message will be editable, much like a
> Supersede: message.

Not by default, I hope.  That should be a separate command.

> I think the
> best approach is to generate a secret string once (for example, a hash
> over the output of the date, ps, and ls ~ commands, or something like
> that, if /dev/[u]random is not available). 

Or "view-lossage" and the content of the *Messages* buffer.

> This way, we don't have to keep a database of cancel keys, and this
> system is safe as long as SHA-1 meets its design goals and the
> secret string is kept secret.

It is not a BIG secret, but it should probably be stored in a separate
unreadable file (e.g. News/cancel-lock) nonetheless.

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

Per Abrahamsen <abraham@dina.kvl.dk> writes:

> Yes.  It would increase the risk of duplicate message ids, i.e. all
> the users of all the machines in our department share the
> 'dina.kvl.dk' mail host name, but I'm the only one using ssv2.

By this argument, using (system-name) for the message-id doesn't
insure uniqueness.  In your case you are the only one using ssv2, but
consider large multiuser servers where hundreds of gnus users may be
using the same machine.

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

Simon Josefsson <simon@josefsson.org> writes:

> People who use default Linux installations (at least RedHat) and DHCP
> will also get a domain part of "localhost.localdomain".

[...]

> So the only "random" part that make Message-Id unique are time.  Not
> good, I think.

Right.  So it still seems that the first part of the Message-ID
(before @) should be enough to guarantee uniqueness if done correctly.
Perhaps if it was a combination of random time, and an aggregate of
username and numeric userid, then it would not matter if
`user-mail-address' was used for the second part of the ID.  Right?

Re: `user-mail-address' for message-ids

[Paul Jarc]

"Jason R. Mastaler" <jason@mastaler.com> writes:
> Per Abrahamsen <abraham@dina.kvl.dk> writes:
> > Yes.  It would increase the risk of duplicate message ids, i.e. all
> > the users of all the machines in our department share the
> > 'dina.kvl.dk' mail host name, but I'm the only one using ssv2.
> 
> By this argument, using (system-name) for the message-id doesn't
> insure uniqueness.

It ensures that this Message-ID won't be the same as any Message-ID
produced on another correctly configured system; this is exactly what
the use of (system-name) is intended to guarantee.

> In your case you are the only one using ssv2, but consider large
> multiuser servers where hundreds of gnus users may be using the same
> machine.

Gnus puts the username in the Message-ID, so this shouldn't be a
problem.  Still, I certainly wouldn't mind if (emacs-pid) were added
into the mix as well.


paul

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

prj@po.cwru.edu (Paul Jarc) writes:

> It ensures that this Message-ID won't be the same as any Message-ID
> produced on another correctly configured system; this is exactly
> what the use of (system-name) is intended to guarantee.

But it is only an intention.  As I mentioned, (system-name) doesn't
return a fqdn on all operating systems (such as Solaris).

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 27 Feb 2001, Jason R. Mastaler wrote:

> Right.  So it still seems that the first part of the Message-ID
> (before @) should be enough to guarantee uniqueness if done
> correctly.  Perhaps if it was a combination of random time, and an
> aggregate of username and numeric userid, then it would not matter
> if `user-mail-address' was used for the second part of the ID.
> Right?

The standard says that the rhs should uniquely identify the host, and
the lhs should uniquely identify the messages from that host.  I
suggest to stick with the standard.  This means to NOT use a rhs like
"hotmail.com". 

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Per Abrahamsen]

"Jason R. Mastaler" <jason@mastaler.com> writes:

> Per Abrahamsen <abraham@dina.kvl.dk> writes:
> 
> > Yes.  It would increase the risk of duplicate message ids, i.e. all
> > the users of all the machines in our department share the
> > 'dina.kvl.dk' mail host name, but I'm the only one using ssv2.
> 
> By this argument, using (system-name) for the message-id doesn't
> insure uniqueness.  

Gnus only use (system-name) if it contains a ".".

In general, the client cannot guarantee unique message-ids on
multiuser machines, but we can minimize the risk.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 27 Feb 2001, Jason R. Mastaler wrote:
> Per Abrahamsen <abraham@dina.kvl.dk> writes:
> 
>> Yes.  It would increase the risk of duplicate message ids, i.e. all
>> the users of all the machines in our department share the
>> 'dina.kvl.dk' mail host name, but I'm the only one using ssv2.
> 
> By this argument, using (system-name) for the message-id doesn't
> insure uniqueness.  In your case you are the only one using ssv2,
> but consider large multiuser servers where hundreds of gnus users
> may be using the same machine.

The lhs should be unique on that host, and the rhs should uniquely
identify the host.  Gnus should make sure that the lhs is unique even
if several people use the same machine.

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Per Abrahamsen]

"Jason R. Mastaler" <jason@mastaler.com> writes:

> Right.  So it still seems that the first part of the Message-ID
> (before @) should be enough to guarantee uniqueness if done correctly.

Nope, there is no way to guarentee that another client will not
generate the same LHS using another algorithm.

> Perhaps if it was a combination of random time, and an aggregate of
> username and numeric userid, then it would not matter if
> `user-mail-address' was used for the second part of the ID.  Right?

Wrong.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 27 Feb 2001, Jason R. Mastaler wrote:
> prj@po.cwru.edu (Paul Jarc) writes:
> 
>> It ensures that this Message-ID won't be the same as any Message-ID
>> produced on another correctly configured system; this is exactly
>> what the use of (system-name) is intended to guarantee.
> 
> But it is only an intention.  As I mentioned, (system-name) doesn't
> return a fqdn on all operating systems (such as Solaris).

Yes.  There is argument about what is a correctly configured system.
Some people say that the FQDN should be in the /etc/hosts file, like
so:

<ip-address>    <host> <host>.<domain.com>

Others say that the implementation of the system-name function in
Emacs is buggy.

I tend to agree with the former group, though I also think that Emacs
could try harder to glean the name, for example from the name server.

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:

> > Right.  So it still seems that the first part of the Message-ID
> > (before @) should be enough to guarantee uniqueness if done
> > correctly.  Perhaps if it was a combination of random time, and an
> > aggregate of username and numeric userid, then it would not matter
> > if `user-mail-address' was used for the second part of the ID.
> > Right?
> 
> The standard says that the rhs should uniquely identify the host, and
> the lhs should uniquely identify the messages from that host.  I
> suggest to stick with the standard.  This means to NOT use a rhs like
> "hotmail.com". 

So in other words, my above assertion is correct, and you now resist
solely because of this "standard"?

I question whether this is really a standard because of how many other
MUAs use the user's defined e-mail address for the rhs of the
message-id.  Mozilla and Mutt immediately come to mind.  I'm sure
there are others as well.  The qmail MTA also allows users to freely
masquerade the hostname of their message-ids to whatever they want.

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:

> Gnus should make sure that the lhs is unique even if several people
> use the same machine.

Or if several people have the same rhs.  This is all I am advocating.

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

Per Abrahamsen <abraham@dina.kvl.dk> writes:

> Nope, there is no way to guarentee that another client will not
> generate the same LHS using another algorithm.

In theory yes, but in reality this is never going to happen.

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

Per Abrahamsen <abraham@dina.kvl.dk> writes:

> > Perhaps if it was a combination of random time, and an aggregate of
> > username and numeric userid, then it would not matter if
> > `user-mail-address' was used for the second part of the ID.  Right?
> 
> Wrong.

Why?  If my `user-mail-address' of "jason@mastaler.com" generates a
rhs of `mastaler.com' and a lhs using time+jason+uid, then how can
there be a conflict?

Re: `user-mail-address' for message-ids

[Per Abrahamsen]

"Jason R. Mastaler" <jason@mastaler.com> writes:

> Per Abrahamsen <abraham@dina.kvl.dk> writes:
> 
> > Nope, there is no way to guarentee that another client will not
> > generate the same LHS using another algorithm.
> 
> In theory yes, but in reality this is never going to happen.

The risk is low as long as the problem is isolated to multiple clients
running on the same machine.  

Re: `user-mail-address' for message-ids

[Per Abrahamsen]

"Jason R. Mastaler" <jason@mastaler.com> writes:

> So in other words, my above assertion is correct, and you now resist
> solely because of this "standard"?

Your assertion wrong.  The reason to resist your suggestion is that it
has absolutely no benefits, has clear practical drawbacks, and on top
of that, violates the standard.

Re: `user-mail-address' for message-ids

[Per Abrahamsen]

"Jason R. Mastaler" <jason@mastaler.com> writes:

> Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
> 
> > Gnus should make sure that the lhs is unique even if several people
> > use the same machine.
> 
> Or if several people have the same rhs. 

It can't.  All it can do is minimize the risk.  If it *must* violate
the standard, it should do so in a way that reduced the risk of
collision, not in a way that increase it.  That is, use something like
"mid.gnus.org" as the right hand side, like Forte Agent does.  In that
case, it would not need to concern itself about conflicts with other
software.

However, I don't think the risk is worth violating the standard for.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 27 Feb 2001, Jason R. Mastaler wrote:

> Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
> 
>> > Right.  So it still seems that the first part of the Message-ID
>> > (before @) should be enough to guarantee uniqueness if done
>> > correctly.  Perhaps if it was a combination of random time, and
>> > an aggregate of username and numeric userid, then it would not
>> > matter if `user-mail-address' was used for the second part of the
>> > ID.  Right?
>> 
>> The standard says that the rhs should uniquely identify the host,
>> and the lhs should uniquely identify the messages from that host.
>> I suggest to stick with the standard.  This means to NOT use a rhs
>> like "hotmail.com".
> 
> So in other words, my above assertion is correct, and you now resist
> solely because of this "standard"?

I didn't try to think about whether it's correct.  But using the user
name or the numeric user id is not going to help much with the many
Linux machines installed with the default settings.  Most probably,
the user name will be "example_user" (or somesuch) and the uid will be
100 (or 101?) -- go look at a Redhat system.

Also, note that normal intuitions about when message ids will be
unique might not take into account the HUGE volume of messages
traversing the Usenet every day.  I wonder how many messages there
are, per second, on an average day?

Then you can see that a dupe in a few million seconds is quite
probable.  A year is 31,536,000 seconds, give or take.  See?

> I question whether this is really a standard because of how many
> other MUAs use the user's defined e-mail address for the rhs of the
> message-id.  Mozilla and Mutt immediately come to mind.  I'm sure
> there are others as well.  The qmail MTA also allows users to freely
> masquerade the hostname of their message-ids to whatever they want.

Ah, right.  Others violate the standard, so let's do it, too.  Right.
Way to go.  Argh!

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 27 Feb 2001, Jason R. Mastaler wrote:

> Why?  If my `user-mail-address' of "jason@mastaler.com" generates a
> rhs of `mastaler.com' and a lhs using time+jason+uid, then how can
> there be a conflict?

Suppose one client does time+uid+name and another does time+uid+Xname
and there is foo@hotmail.com as well as Xfoo@hotmail.com.  Then it's
sufficient for the uid to be equal and the time, and -- clash!

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 27 Feb 2001, Jason R. Mastaler wrote:
> Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
> 
>> Gnus should make sure that the lhs is unique even if several people
>> use the same machine.
> 
> Or if several people have the same rhs.  This is all I am
> advocating.

On a single machine, Gnus can write `fsf' into the lhs and hope that
no other client is going to use that string.  But how do two Gnusae on
two different systems make sure that there won't be a clash?  They'd
have to write the system name into the lhs.

The standard says to write it into the rhs, which is just as good.

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Per Abrahamsen]

"Jason R. Mastaler" <jason@mastaler.com> writes:

> Why?  If my `user-mail-address' of "jason@mastaler.com" generates a
> rhs of `mastaler.com' and a lhs using time+jason+uid, then how can
> there be a conflict?

Nothing guarantees that another client won't generate the same lhs
using a different algorithm.

Re: `user-mail-address' for message-ids

[Simon Josefsson]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:

> I tend to agree with the former group, though I also think that Emacs
> could try harder to glean the name, for example from the name server.

I think this is what's happening -- it uses uname() or gethostname()
to get a name, and if that value doesn't conain `.' it proceeds with a
gethostbyname(). `init_system_name' in sysdep.c.

Re: `user-mail-address' for message-ids

[Paul Jarc]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
> On a single machine, Gnus can write `fsf' into the lhs and hope that
> no other client is going to use that string.

Actually, Gnus already does better than this, by encoding the uid into
the lhs.  But this doesn't protect against two Gnusae running as the
same user at the same time.  Should we be worried?  It's an easy fix;
here's a patch, along with some comment cleanups.  ChangeLog:
	* message.el (message-unique-id): Even uniquer!  Use (emacs-pid).

--- message.el~	Tue Feb 27 14:04:55 2001
+++ message.el	Tue Feb 27 14:10:00 2001
@@ -3158,11 +3158,11 @@
 ;; or string boundary), or modify the "fsf" string.
 (defun message-unique-id ()
   ;; Don't use microseconds from (current-time), they may be unsupported.
-  ;; Instead we use this randomly inited counter.
+  ;; Instead we use this randomly initialized counter.
   (setq message-unique-id-char
 	(% (1+ (or message-unique-id-char (logand (random t) (1- (lsh 1 20)))))
 	   ;; (current-time) returns 16-bit ints,
-	   ;; and 2^16*25 just fits into 4 digits i base 36.
+	   ;; and 2^16*25 just fits into 4 digits in base 36.
 	   (* 25 25)))
   (let ((tm (current-time)))
     (concat
@@ -3176,6 +3176,8 @@
 			       (lsh (% message-unique-id-char 25) 16)) 4)
      (message-number-base36 (+ (nth 1 tm)
 			       (lsh (/ message-unique-id-char 25) 16)) 4)
+     "-"
+     (message-number-base36 (emacs-pid) -1)
      ;; Append the newsreader name, because while the generated
      ;; ID is unique to this newsreader, other newsreaders might
      ;; otherwise generate the same ID via another algorithm.


paul

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:

> Suppose one client does time+uid+name and another does
> time+uid+Xname and there is foo@hotmail.com as well as
> Xfoo@hotmail.com.  Then it's sufficient for the uid to 
> be equal and the time, and -- clash!

Yes, but the username will not be equal (name != Xname) 
and therefore the Message-IDs will differ.

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

Per Abrahamsen <abraham@dina.kvl.dk> writes:

> Nothing guarantees that another client won't generate the same lhs
> using a different algorithm.

By another "client", do you mean a piece of software other than gnus?

Re: `user-mail-address' for message-ids

[Paul Jarc]

"Jason R. Mastaler" <jason@mastaler.com> writes:
> Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
> > Suppose one client does time+uid+name and another does
> > time+uid+Xname and there is foo@hotmail.com as well as
> > Xfoo@hotmail.com.  Then it's sufficient for the uid to 
> > be equal and the time, and -- clash!
> 
> Yes, but the username will not be equal (name != Xname) 
> and therefore the Message-IDs will differ.

They'll be the same: user foo uses algorithm 1, which spits out
"X"+username ("Xfoo"), along with other stuff.  User Xfoo uses
algorithm 2, which spits out username ("Xfoo"), along with other
stuff, which happens to be the same other stuff.  Both algorithms,
with their respective inputs, produce the same output.

We can ensure that the LHS would be unique if everyone used the same
algorithm, but we can't make everyone use the same algorithm.


paul

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="", Size: 519 bytes --]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:

> Ah, right.  Others violate the standard, so let's do it, too.
> Right.  Way to go.  Argh!

But you haven't provided any proof of this "standard" of requiring the
fqdn (and nothing else) of the running system to make up the rhs.

I may be wrong, but it doesn't seem likely to me that several distinct
and intelligent groups of software developers (mutt, mozilla, qmail,
etc..) would have goofed so badly if this is an established standard.

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

prj@po.cwru.edu (Paul Jarc) writes:

> They'll be the same: user foo uses algorithm 1, which spits out
> "X"+username ("Xfoo"), along with other stuff.  User Xfoo uses
> algorithm 2, which spits out username ("Xfoo"), along with other
> stuff, which happens to be the same other stuff.  Both algorithms,
> with their respective inputs, produce the same output.
> 
> We can ensure that the LHS would be unique if everyone used the same
> algorithm, but we can't make everyone use the same algorithm.

I'm not following this argument.  We are talking about gnus users here
which using the same software will be using the same algorithm for
message-id generation.  Therefore gnus users foo and Xfoo won't be
producing the same output.
 

Re: `user-mail-address' for message-ids

[Paul Jarc]

"Jason R. Mastaler" <jason@mastaler.com> writes:
> But you haven't provided any proof of this "standard" of requiring the
> fqdn (and nothing else) of the running system to make up the rhs.

See RFC1036, 2.1.5.

> I may be wrong, but it doesn't seem likely to me that several distinct
> and intelligent groups of software developers (mutt, mozilla, qmail,
> etc..) would have goofed so badly if this is an established standard.

They don't goof, in themselves.  They just make it easier for you to
make them goof.  It's possible to use, e.g., qmail's idhost to change
the domain in Message-IDs in such a way that uniqueness is still
guaranteed, although this might technically be a violation of the RFC.
It's also possible to misuse these features, and misuse would be much
more common than benign use in the case of user-mail-address.


paul

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 27 Feb 2001, Jason R. Mastaler wrote:

> Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
> 
>> Suppose one client does time+uid+name and another does
>> time+uid+Xname and there is foo@hotmail.com as well as
>> Xfoo@hotmail.com.  Then it's sufficient for the uid to 
>> be equal and the time, and -- clash!
> 
> Yes, but the username will not be equal (name != Xname) 
> and therefore the Message-IDs will differ.

No, if one user is named `Xfoo' and the user agent does time+uid+name,
then you might get a clash with a user named `foo' and a user agent
which does time+uid+Xname.
                    ^

I admit this is a contrived example.

But note that, as the number of people increases, the probability of a
clash goes up.  Dramatically.

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 27 Feb 2001, Jason R. Mastaler wrote:

> Per Abrahamsen <abraham@dina.kvl.dk> writes:
> 
>> Nothing guarantees that another client won't generate the same lhs
>> using a different algorithm.
> 
> By another "client", do you mean a piece of software other than
> gnus?

Yes.  Please note that, amongst the people using an @hotmail.com mail
address (say), there will be many different clients.

If, however, you include the hostname in the rhs of a message id, then
the probability of a clash goes down, since on the same host not so
many different user agents will be used.

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Paul Jarc]

"Jason R. Mastaler" <jason@mastaler.com> writes:
> We are talking about gnus users here which using the same software
> will be using the same algorithm for message-id generation.

There's the confusion: we have to talk abaout *all* users, with *all*
software, on *all* machines, or else Message-ID becomes useless.  An
identifier that's unique only among messages that came from Gnus isn't
useful.

Hosts shouldn't have to coordinate Message-ID generation, so they put
their FQDNs in Message-IDs, so that Message-IDs generated by different
hosts won't collide.

User agents on the same host shouldn't have to coordinate, so they
(hopefully) put an identifier such as ".fsf" in Message-IDs, so that
Message-IDs generated by different user agents won't collide.

Similarly for different users using the same agent on the same host,
different processes of the same user using the same agent on the same
host, etc.


paul

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 27 Feb 2001, Simon Josefsson wrote:

> Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
> 
>> I tend to agree with the former group, though I also think that
>> Emacs could try harder to glean the name, for example from the name
>> server.
> 
> I think this is what's happening -- it uses uname() or gethostname()
> to get a name, and if that value doesn't conain `.' it proceeds with
> a gethostbyname(). `init_system_name' in sysdep.c.

By `trying harder' I mean that it should recognize
`localhost.localdomain' as a very improbable FQDN and try to ask the
name server, instead.

It used to be that Emacs used the output of the `domainname' command
(well, the appropriate equivalent library function or system call),
but that was the NIS domain and not the DNS domain...

My idea is that Emacs should look if there is /etc/resolv.conf and if
there is a nameserver entry in it, and if so, it should ask the name
server for the FQDN of the local host.  (Unless the FQDN was found
already in /etc/hosts.)

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 27 Feb 2001, Paul Jarc wrote:

> Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
>> On a single machine, Gnus can write `fsf' into the lhs and hope
>> that no other client is going to use that string.
> 
> Actually, Gnus already does better than this, by encoding the uid
> into the lhs.

I meant that the `fsf' is there to distinguish Gnus from, say, slrn or
tin.  The uid is there to distinguish one Gnus user from another.

I like your `emacs-pid' suggestion.  I only wonder if it is portable
between Emacs and XEmacs and different operating systems.

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:

> No, if one user is named `Xfoo' and the user agent does
> time+uid+name, then you might get a clash with a user named `foo'
> and a user agent which does time+uid+Xname.

But remember that gnus also adds a `.fsf' to the lhs.

Re: `user-mail-address' for message-ids

[Paul Jarc]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
> My idea is that Emacs should look if there is /etc/resolv.conf and if
> there is a nameserver entry in it, and if so, it should ask the name
> server for the FQDN of the local host.

I'm a bit wary of having to talk to the network for this purpose.  I'd
almost prefer signaling an error and giving up, except that would hurt
non-root users on systems with lazy admins.

> (Unless the FQDN was found already in /etc/hosts.)

If a system is locally misconfigured such that gethostname() doesn't
return something useful, I doubt /etc/hosts will be any more helpful.


paul

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

Per Abrahamsen <abraham@dina.kvl.dk> writes:

> Your assertion wrong.  The reason to resist your suggestion is that it
> has absolutely no benefits, has clear practical drawbacks, and on top
> of that, violates the standard.

You are certainly entitled to your opinion, but two of the above three
arguments are just that.  There are benefits to others besides
yourself, and I'm still not convinced that my suggestion has any
*practical* drawbacks or adds any risks.

However, since it technically violates the RFC, I will throw in the
towel.  You win.

Re: `user-mail-address' for message-ids

[Paul Jarc]

"Jason R. Mastaler" <jason@mastaler.com> writes:
> Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
> > No, if one user is named `Xfoo' and the user agent does
> > time+uid+name, then you might get a clash with a user named `foo'
> > and a user agent which does time+uid+Xname.
> 
> But remember that gnus also adds a `.fsf' to the lhs.

A different algorithm might happen to append that same string,
possibly as part of its randomness.  There's no standard (AFAIK) for
user agents to avoid stepping on each others' toes here, so there's no
way for Gnus to generate a Message-ID that it know will never be
produced by a different agent.

"But this means the uniqueness problem is intractable!" you may say,
and you'd be right: to absolutely guarantee uniqueness, we'd need
either a single facility for generating Message-IDs systemwide, or a
standard to keep the different generators out of each others' way.  We
don't have those, so we don't have a perfect solution.  But we do have
a solution that, while imperfect, is not too bad, and we can avoid
making it worse.


paul

Re: `user-mail-address' for message-ids

[Florian Weimer]

Per Abrahamsen <abraham@dina.kvl.dk> writes:

> Nothing guarantees that another client won't generate the same lhs
> using a different algorithm.

If there are, say, 64 bit of entropy in the message ID, this is not
going to happen, regardless of the scheme being used.

The '%s.mid.gnus.org' solution seems to be a good idea.  For cancel
locks, we need an entropy collector anyway, so we can use its output
to generate a unique identifier for the RHS as well.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset=us-ascii, Size: 1102 bytes --]

On 27 Feb 2001, Jason R. Mastaler wrote:
> Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
> 
>> Ah, right.  Others violate the standard, so let's do it, too.
>> Right.  Way to go.  Argh!
> 
> But you haven't provided any proof of this "standard" of requiring
> the fqdn (and nothing else) of the running system to make up the
> rhs.

The standard does not require the fqdn.  The standard requires the rhs
to uniquely identify the host.

For dialup users, it might be convenient to use a rhs of the form
johnsmith.dialup.provider.net even though this `host name' is not a
valid name in the DNS.  But the host master of the provider.net domain
guarantees that a rhs of this form always uniquely identifies the
dialup host of the user `johnsmith'.

I hope you now understand the intent of the rhs of a message id.

I've been so vague in saying `the standard' because I knew it's a
pertinent RFC, but I forgot the number.  Thankfully, Paul said it's
RFC 1036.  Okay, now you can replace `the standard' with `RFC 1036' in
my previous posts... :-)

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Samuel Padgett]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:

> There is argument about what is a correctly configured system.  Some
> people say that the FQDN should be in the /etc/hosts file, like so:
> 
> <ip-address>    <host> <host>.<domain.com>

What if I connect to the Internet using DHCP, and I have a dynamic IP
address?

Sam
-- 
What the large print giveth, the small print taketh away.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 27 Feb 2001, Jason R. Mastaler wrote:

> You are certainly entitled to your opinion, but two of the above
> three arguments are just that.  There are benefits to others besides
> yourself, and I'm still not convinced that my suggestion has any
> *practical* drawbacks or adds any risks.

The problems I envision stem from lots of people using `@hotmail.com'
message ids.

As the number of people with the same right hand side increases, the
risk of having a clash in the left hand side increases more than
linearly.  (Not sure if it's exponentially.)  Even though there are
365 days in a year, just a handful of people (maybe 7, maybe 13, maybe
14, not sure) are enough to give a more than 50% chance of two of them
having the same birthday (maybe in different years).  You might want
to use this for little bets at parties...

The chance of two people having the same birthday is one in 365.  The
chance of any two people in three having the same birthday is 3 in
365 (one in 365 for Peter and Paul, 1/365 for Paul and John, and 1/365
for Peter and John).  The chance for such a pairing in four people is
... hm.  There is a 3/365 chance for Mary to have the same birthday as
one of Peter, Paul, or John, and a 3/365 chance for any two of the
guys having the same birthday.  So now we're up to 6/365.  And then
12/365 (for five), and 24/365 (for six), and 48/365 (for seven).  So
nine or ten people should be enough.  Any error in my calculation?

So the chance really does grow exponentially.

So even though the chance of a clash between two specific Hotmail
users is small, the chance of a clash between _any_ two Hotmail users
is quite large because there are so many Hotmail users.

Message ids need to be unique for two years.

Also consider the huge volume of messages passing through Usenet every
day. 

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 27 Feb 2001, Paul Jarc wrote:

> Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
>
>> (Unless the FQDN was found already in /etc/hosts.)
> 
> If a system is locally misconfigured such that gethostname() doesn't
> return something useful, I doubt /etc/hosts will be any more
> helpful.

I think gethostname() looks in /etc/hosts, no?

I was proposing to do something else if gethostname() fails.

It is normal for Solaris machines to not have the FQDN in /etc/hosts
or in the NIS database.  Therefore on a typical Solaris machine you
have to ask DNS.

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:

> The problems I envision stem from lots of people using
> `@hotmail.com' message ids.

Practically speaking, I still don't feel there are any problems with
this.  We already established that different gnus users @hotmail.com
won't be clashing.  Now, even if a different user-agent produces a lhs
that is identical to gnus' "time+username+uid+pid", the `.fsf' which
is unique to gnus will differentiate them.  In practice, no user-agent
other than gnus is going to produce a lhs that is identical to gnus'
"time+username+uid+pid+.fsf", period.

All this said, I already agreed to drop my case because of the RFC
violation.

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

prj@po.cwru.edu (Paul Jarc) writes:

> > But remember that gnus also adds a `.fsf' to the lhs.
> 
> A different algorithm might happen to append that same string,

In practice this just isn't going to happen.  To reduce the
possibility even further however, I'd replace `.fsf' with `.gnus'.

Index: lisp/message.el
===================================================================
RCS file: /usr/local/cvsroot/gnus/lisp/message.el,v
retrieving revision 6.65
diff -u -r6.65 message.el
--- lisp/message.el     2001/02/19 01:13:41     6.65
+++ lisp/message.el     2001/02/27 23:48:54
@@ -3155,7 +3155,7 @@
 ;; If you ever change this function, make sure the new version
 ;; cannot generate IDs that the old version could.
 ;; You might for example insert a "." somewhere (not next to another dot
-;; or string boundary), or modify the "fsf" string.
+;; or string boundary), or modify the ".gnus" string.
 (defun message-unique-id ()
   ;; Don't use microseconds from (current-time), they may be unsupported.
   ;; Instead we use this randomly inited counter.
@@ -3179,7 +3179,7 @@
      ;; Append the newsreader name, because while the generated
      ;; ID is unique to this newsreader, other newsreaders might
      ;; otherwise generate the same ID via another algorithm.
-     ".fsf")))
+     ".gnus")))
 
 (defun message-number-base36 (num len)
   (if (if (< len 0)

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

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

> > Nothing guarantees that another client won't generate the same lhs
> > using a different algorithm.
> 
> If there are, say, 64 bit of entropy in the message ID, this is not
> going to happen, regardless of the scheme being used.

Thanks for expressing mathematically what I've been trying to get
across all day.

> The '%s.mid.gnus.org' solution seems to be a good idea.  For cancel
> locks, we need an entropy collector anyway, so we can use its output
> to generate a unique identifier for the RHS as well.

I'm all for a better RHS than (system-name), but why stick "gnus.org"
in there.  This looks too much like a fqdn.

why `system-name' is bad for Message-IDs (was Re: `user-mail-address' for message-ids)

[Jason R. Mastaler]

Alan Shutko <ats@acm.org> writes:

> What's wrong with system-name?

For one, it encourages SPAM/UCE.  I constantly get UCE sent to
"message-id@system-name" which ends up in my mailbox because of the
blindforwarding I've setup for my domain.  This alone is a good
enough reason to use something else for the RHS IMO.

Re: `user-mail-address' for message-ids

[Paul Jarc]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
> I think gethostname() looks in /etc/hosts, no?

No, it's a system call; the hostname is maintained in memory by the
kernel.  You tell the kernel what the hostname is using sethostname()
as part of bootup procedure; depending on the system, this may involve
looking at /etc/hosts, but gethostname() itself doesn't know about any
particular files.

(This is on Linux, but my Linux man page says it conforms to SVr4 and
4.4BSD, among others.)

> It is normal for Solaris machines to not have the FQDN in /etc/hosts
> or in the NIS database.  Therefore on a typical Solaris machine you
> have to ask DNS.

Once at boot, maybe, to tell the kernel.  After that you just ask the
kernel.


paul

Re: `user-mail-address' for message-ids

[Paul Jarc]

"Jason R. Mastaler" <jason@mastaler.com> writes:
> We already established that different gnus users @hotmail.com
> won't be clashing.

How so?


paul

Re: why `system-name' is bad for Message-IDs (was Re: `user-mail-address' for message-ids)

[Paul Jarc]

"Jason R. Mastaler" <jason@mastaler.com> writes:
> Alan Shutko <ats@acm.org> writes:
> > What's wrong with system-name?
> 
> For one, it encourages SPAM/UCE.  I constantly get UCE sent to
> "message-id@system-name" which ends up in my mailbox because of the
> blindforwarding I've setup for my domain.  This alone is a good
> enough reason to use something else for the RHS IMO.

No, that's a good enough reason to fix your mail forwarding.


paul

Re: `user-mail-address' for message-ids

[Lloyd Zusman]

"Jason R. Mastaler" <jason@mastaler.com> writes:

> Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
> 
> > No, if one user is named `Xfoo' and the user agent does
> > time+uid+name, then you might get a clash with a user named `foo'
> > and a user agent which does time+uid+Xname.
> 
> But remember that gnus also adds a `.fsf' to the lhs.

But any other user agent might randomly decide to do that, too.  We
cannot control what the other agents do.


-- 
 Lloyd Zusman
 ljz@asfast.com

Re: why `system-name' is bad for Message-IDs (was Re: `user-mail-address' for message-ids)

[Jason R. Mastaler]

prj@po.cwru.edu (Paul Jarc) writes:

> No, that's a good enough reason to fix your mail forwarding.

BS.  This is an extremely widespread technique for virtual mail
domains.  Again, just because you don't find it useful doesn't mean
that others don't or that it's wrong.

Re: `user-mail-address' for message-ids

[Florian Weimer]

"Jason R. Mastaler" <jason@mastaler.com> writes:

> I'm all for a better RHS than (system-name), but why stick "gnus.org"
> in there.  This looks too much like a fqdn.

So-called standards and common practice mandates a FQDN there.  In
addition, if you post to some groups with a semantically incorrect
message ID, you're going to be flamed to death.

Sorry, the FQDN issue is not open to discussion.

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

prj@po.cwru.edu (Paul Jarc) writes:

> > We already established that different gnus users @hotmail.com
> > won't be clashing.
> 
> How so?

These are your own words Paul:

  "We can ensure that the LHS would be unique if everyone used the same
   algorithm, but we can't make everyone use the same algorithm."

Gnus users share the same algorithm which ensures that the LHS is
unique which means they won't be clashing.

Re: why `system-name' is bad for Message-IDs (was Re: `user-mail-address' for message-ids)

[Florian Weimer]

"Jason R. Mastaler" <jason@mastaler.com> writes:

> For one, it encourages SPAM/UCE.  I constantly get UCE sent to
> "message-id@system-name" which ends up in my mailbox because of the
> blindforwarding I've setup for my domain.

That's a problem with your personal setup.

Re: why `system-name' is bad for Message-IDs (was Re: `user-mail-address' for message-ids)

[Paul Jarc]

"Jason R. Mastaler" <jason@libertine.org> writes:
> prj@po.cwru.edu (Paul Jarc) writes:
> > No, that's a good enough reason to fix your mail forwarding.
> 
> BS.  This is an extremely widespread technique for virtual mail
> domains.  Again, just because you don't find it useful doesn't mean
> that others don't or that it's wrong.

I'm not saying you should *stop* forwarding, just that you can solve
this problem by *changing* your forwarding.  Since you're in control
of this domain, you can set your system's hostname to a name that has
an appropriate A record, but which you don't accept mail for.


paul

Re: `user-mail-address' for message-ids

[Paul Jarc]

"Jason R. Mastaler" <jason@mastaler.com> writes:
> prj@po.cwru.edu (Paul Jarc) writes:
> > > We already established that different gnus users @hotmail.com
> > > won't be clashing.
> > 
> > How so?
> 
> These are your own words Paul:
> 
>   "We can ensure that the LHS would be unique if everyone used the same
>    algorithm, but we can't make everyone use the same algorithm."
> 
> Gnus users share the same algorithm which ensures that the LHS is
> unique which means they won't be clashing.

Note that I didn't say what it would take to provide that guarantee.
You'd have to encode the FQDN on the LHS, I think; your suggestion
didn't include that, so I can't see how your suggestion provides
uniqueness.


paul

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

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

> So-called standards and common practice mandates a FQDN there.  In
> addition, if you post to some groups with a semantically incorrect
> message ID, you're going to be flamed to death.
> 
> Sorry, the FQDN issue is not open to discussion.

OK boss, but using ".*gnus.org" in the FQDN is not open for discussion
either unless you happen to be Lars.

Re: why `system-name' is bad for Message-IDs (was Re: `user-mail-address' for message-ids)

[Jason R. Mastaler]

prj@po.cwru.edu (Paul Jarc) writes:

> I'm not saying you should *stop* forwarding, just that you can solve
> this problem by *changing* your forwarding.  Since you're in control
> of this domain, you can set your system's hostname to a name that has
> an appropriate A record, but which you don't accept mail for.

What if I want my system to directly accept mail?  Why should I have
to incur unnecessary SPAM simply because of my message-id?

Re: why `system-name' is bad for Message-IDs (was Re: `user-mail-address' for message-ids)

[Jason R. Mastaler]

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

> That's a problem with your personal setup.

You call it a "problem", but many people call it a useful feature.

Re: why `system-name' is bad for Message-IDs (was Re: `user-mail-address' for message-ids)

[Florian Weimer]

"Jason R. Mastaler" <jason@libertine.org> writes:

> Florian Weimer <fw@deneb.enyo.de> writes:
> 
> > That's a problem with your personal setup.
> 
> You call it a "problem", but many people call it a useful feature.

What feature?  The feature that your blindforwarding setup does not
filter out addr-specs which look like message IDs (and the local-parts
'fsf' and 'fsf_-_')?

Re: `user-mail-address' for message-ids

[Florian Weimer]

"Jason R. Mastaler" <jason@libertine.org> writes:

> OK boss, but using ".*gnus.org" in the FQDN is not open for discussion
> either unless you happen to be Lars.

Yes, but the entire discussion might become pointless as soon as Lars
is involved because in the past, Lars has strongly opposed running
Gnus on machines which do not have an official FQDN.

Re: `user-mail-address' for message-ids

[Florian Weimer]

"Jason R. Mastaler" <jason@mastaler.com> writes:

>   "We can ensure that the LHS would be unique if everyone used the same
>    algorithm, but we can't make everyone use the same algorithm."

If the local-part includes a reasonable amount of entropy, the
probability of a Message-ID collision is negligible (compared to, say,
the whole universe vanishing in a quantum fluctuation).

OTOH, it is extremly bad style to use someone else's FQDN in your
message IDs.

Re: `user-mail-address' for message-ids

[Christoph Conrad]

    > As the number of people with the same right hand side increases,
    > the risk of having a clash in the left hand side increases more
    > than linearly. (Not sure if it's exponentially.) Even though

It's the birthday paradox, see

<http://www.howstuffworks.com/question261.htm>

for a good explanation.


>From <http://www.io.com/~ritter/ARTS/BIRTHDAY.HTM>:

========================================================================
CUT HERE - CUT HERE - CUT HERE - CUT HERE - CUT HERE - CUT HERE - CUT HE
========================================================================
                            N!
       Pd(N,s) = 1  -  ------------
                        (N-s)! N^s

where s is the expected number of samples needed, N the size of the
population being sampled, and p the given probability.
========================================================================
CUT HERE - CUT HERE - CUT HERE - CUT HERE - CUT HERE - CUT HERE - CUT HE
========================================================================

A little bit later in the article is a approximation formula.

A simple approximation formula for big N is sqrt( 2 * N ) for the 50%
chance of having a collision.

Best regards,
cu, -cc-
-- 
=> GNU Emacs Webring @ <http://www.gnusoftware.com/WebRing/> <=
Look Ma, this man can twist his fingers as if they were made of rubber,
isn't that amazing? -- Not really, he's been using emacs for years...!

Re: `user-mail-address' for message-ids

[Paul Stevenson]

P Jarc writes:

> 	* message.el (message-unique-id): Even uniquer!  Use (emacs-pid).
> 
> --- message.el~	Tue Feb 27 14:04:55 2001
> +++ message.el	Tue Feb 27 14:10:00 2001


I have a sillier way of making Message-ID repetition even more unlikely:

(setq pauls-random-word-file "/usr/dict/words")
(defun message-unique-id ()
  (save-excursion
    (setq message-unique-id-char
          (% (1+ (or message-unique-id-char 
                     (logand (random t) (1- (lsh 1 20)))))
             (* 25 25)))
    (let ((tm (current-time)))
      (concat
       (if (memq system-type '(ms-dos emx vax-vms))
           (let ((user (downcase (user-login-name))))
             (while (string-match "[^a-z0-9_]" user)
               (aset user (match-beginning 0) ?_))
             user)
         (message-number-base36 (user-uid) -1))
       (message-number-base36 (+ (car   tm)
                                 (lsh (% message-unique-id-char 25) 16)) 4)
       (message-number-base36 (+ (nth 1 tm)
                                 (lsh (/ message-unique-id-char 25) 16)) 4)
       "."
       (progn
         (save-excursion
           (let ((word-buffer (find-file-noselect pauls-random-word-file))
                 (number-of-lines)(random-word))
             (set-buffer word-buffer)
             (widen)
             (goto-char (point-min))
             (setq number-of-lines (count-lines (point-min) (point-max)))
             (random t)
             (forward-line (random number-of-lines) )
             (let ((here (point)))
               (end-of-line)(narrow-to-region here (point)))
             (setq random-word (current-word))
             (kill-buffer word-buffer)
             random-word)))))))

Re: `user-mail-address' for message-ids

[Per Abrahamsen]

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

> If the local-part includes a reasonable amount of entropy, the
> probability of a Message-ID collision is negligible (compared to, say,
> the whole universe vanishing in a quantum fluctuation).

This is the strategy Forte Agent uses.

However, given that there is no standard algorithm for generating the
LHS, and thus no guarantee that the algorithm contains a sufficient
amount of entropy, it makes sense to insist on the rules that exists
for the thr RHS, which at least localizes the problem.

> OTOH, it is extremly bad style to use someone else's FQDN in your
> message IDs.

Which is why Forte has registeret a domain specifically for Agent
message-id's.  It violates the letter of the standard, but at least it
provides a guarentee against collisions with message-id's generated by
other agents.

Re: `user-mail-address' for message-ids

[Per Abrahamsen]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:

> I tend to agree with the former group, though I also think that Emacs
> could try harder to glean the name, for example from the name server.

There was a period where the came a lot of bug reports from Linux
users about Emacs being absurdly slow (even more than usual) at
upstart.  I don't remember the details, but the problem involved Emacs
waiting for a nameserver query to time out, and the solution involved
adding something to /etc/hosts.

Re: `user-mail-address' for message-ids

[Florian Weimer]

Per Abrahamsen <abraham@dina.kvl.dk> writes:

> Florian Weimer <fw@deneb.enyo.de> writes:
> 
> > If the local-part includes a reasonable amount of entropy, the
> > probability of a Message-ID collision is negligible (compared to, say,
> > the whole universe vanishing in a quantum fluctuation).
> 
> This is the strategy Forte Agent uses.
> 
> However, given that there is no standard algorithm for generating the
> LHS, and thus no guarantee that the algorithm contains a sufficient
> amount of entropy,

As long as the Gnus algorithm provides enough entropy, message IDs
generated by Gnus will not cause collisions.

Re: why `system-name' is bad for Message-IDs (was Re: `user-mail-address' for message-ids)

[Alan Shutko]

"Jason R. Mastaler" <jason@mastaler.com> writes:

> Alan Shutko <ats@acm.org> writes:
> 
> > What's wrong with system-name?
> 
> For one, it encourages SPAM/UCE.  

1.  Deal with it.  If your forwarding is such that people sending to
    message-ids at your domain gets to you, fix the forwarding.

2.  Read the whole post.  My question was in reference to having a
    variable to set the fqdn.  I was wondering why system-name (a
    variable that already exists, which strangely enough does exactly
    that) isn't adequate.

Personally, if you've got such a problem about having the host name in
your message ids, you have the source.  Fix it yourself rather than
trying to convince other people that they should disregard the
standard because it bothers you on your system.

-- 
Alan Shutko <ats@acm.org> - In a variety of flavors!
Everything bows to success, even grammar.

Re: why `system-name' is bad for Message-IDs (was Re: `user-mail-address' for message-ids)

[Paul Jarc]

"Jason R. Mastaler" <jason@libertine.org> writes:
> What if I want my system to directly accept mail?  Why should I have
> to incur unnecessary SPAM simply because of my message-id?

Have your system's hostname be, say, host.domain.tld; have it accept
mail for domain.tld and reject mail for host.domain.tld.  Then you'll
get your mail, and the spam will bounce.


paul

Re: why `system-name' is bad for Message-IDs (was Re: `user-mail-address' for message-ids)

[Kai Großjohann]

On Wed, 28 Feb 2001, Alan Shutko wrote:

> 2.  Read the whole post.  My question was in reference to having a
>     variable to set the fqdn.  I was wondering why system-name (a
>     variable that already exists, which strangely enough does
>     exactly that) isn't adequate.

Hm.  Some people might wish to use a different rhs than their fqdn.
For example, a dialup user Joe might be able to use
joe.dialup.provider.net as the rhs of a message id, but this need not
be a fqdn known to DNS.

On the one hand, this makes it clear that the rhs of a message id is
not always the same as the fqdn.  On the other hand, these dialup
users probably don't HAVE a fqdn...  And if they did, they might as
well use that.  Hm.

Thoughts?

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 27 Feb 2001, Kai Großjohann wrote:

> The chance of two people having the same birthday is one in 365.
> The chance of any two people in three having the same birthday is 3
> in 365 (one in 365 for Peter and Paul, 1/365 for Paul and John, and
> 1/365 for Peter and John).

This is bogus.  Hm.  The probability of Peter and Paul having a
different birthday is 364/365, same for Paul and John, and same for
Peter and John.  The probability for no clash is that Peter and Paul
have a different birthday, and Paul and John, and Peter and John, so
the probability for no clash is 364/365 * 364/365 * 364/365 =
(364^3)/(365^3).  So the probability for a clash is 1 -
(364^3)/(365^3) = 0.008 = .8%.

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 28 Feb 2001, Florian Weimer wrote:
> "Jason R. Mastaler" <jason@libertine.org> writes:
> 
>> OK boss, but using ".*gnus.org" in the FQDN is not open for
>> discussion either unless you happen to be Lars.
> 
> Yes, but the entire discussion might become pointless as soon as
> Lars is involved because in the past, Lars has strongly opposed
> running Gnus on machines which do not have an official FQDN.

He he.  But you need Lars' cooperation to be able to use mid.gnus.org :-)

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 27 Feb 2001, Paul Jarc wrote:
> Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
>> I think gethostname() looks in /etc/hosts, no?
> 
> No, it's a system call; the hostname is maintained in memory by the
> kernel.  You tell the kernel what the hostname is using
> sethostname() as part of bootup procedure; depending on the system,
> this may involve looking at /etc/hosts, but gethostname() itself
> doesn't know about any particular files.

Ah, right.  When booting, a Solaris machine does "hostname `cat
/etc/hostname.le0`" or similar.  It is normal for a Solaris machine to
have a node name there, rather than a FQDN.

I think there was a problem with netgroups being too long or something
like this, I don't quite remember where this convention comes from.

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 27 Feb 2001, Samuel Padgett wrote:

> Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:
> 
>> There is argument about what is a correctly configured system.
>> Some people say that the FQDN should be in the /etc/hosts file,
>> like so:
>> 
>> <ip-address>    <host> <host>.<domain.com>
> 
> What if I connect to the Internet using DHCP, and I have a dynamic
> IP address?

When asked, the same people reply that you should obtain a FQDN from
dyndns.org or some similar service.

I'm not saying it's my own opinion.

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Toby Speight]

0> In article <puid7c4nh80.fsf@cyan.acl.lanl.gov>,
0> Jason R. Mastaler <URL:mailto:jason@mastaler.com> ("Jason") wrote:

Jason> I question whether this is really a standard because of how
Jason> many other MUAs use the user's defined e-mail address for the
Jason> rhs of the message-id.  Mozilla and Mutt immediately come to
Jason> mind.  I'm sure there are others as well.

"Other MUAs are broken, let's break Gnus too."  No, thanks.

Re: `user-mail-address' for message-ids

[Toby Speight]

0> In article <vafofvnohqx.fsf@lucy.cs.uni-dortmund.de>,
0> Kai Großjohann <URL:mailto:Kai.Grossjohann@CS.Uni-Dortmund.DE> ("Kai") wrote:

Kai> As the number of people with the same right hand side increases,
Kai> the risk of having a clash in the left hand side increases more
Kai> than linearly.  (Not sure if it's exponentially.)

It's O(n²), IINM.

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 28 Feb 2001, Per Abrahamsen wrote:

> There was a period where the came a lot of bug reports from Linux
> users about Emacs being absurdly slow (even more than usual) at
> upstart.  I don't remember the details, but the problem involved
> Emacs waiting for a nameserver query to time out, and the solution
> involved adding something to /etc/hosts.

Yep.  Hm.  So maybe it's good that Emacs doesn't ask the name server
anymore.  Hm.

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[Toby Speight]

0> In article <vafofvnohqx.fsf@lucy.cs.uni-dortmund.de>,
0> Kai Großjohann <URL:mailto:Kai.Grossjohann@CS.Uni-Dortmund.DE> ("Kai") wrote:

Kai> As the number of people with the same right hand side increases,
Kai> the risk of having a clash in the left hand side increases more
Kai> than linearly.  (Not sure if it's exponentially.)

0> In article <s83dcybylu.fsf@lanber.cam.eu.citrix.com>,
0> Toby Speight <URL:mailto:streapadair@gmx.net> ("Toby") wrote:

Toby> It's O(n²), IINM.

IWM.  Doh!

Re: `user-mail-address' for message-ids

[Norbert Koch]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:

Hi!

> This is bogus.  Hm.  The probability of Peter and Paul having a
> different birthday is 364/365, same for Paul and John, and same for
> Peter and John.  The probability for no clash is that Peter and Paul
> have a different birthday, and Paul and John, and Peter and John, so
> the probability for no clash is 364/365 * 364/365 * 364/365 =
> (364^3)/(365^3).  So the probability for a clash is 1 -
> (364^3)/(365^3) = 0.008 = .8%.

Isn't this the probability of Peter, Paul, and Mary err John having
birthday on the same day?

It's been a long time since my last stastistical lesson in school,
though. 

norbert.

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

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

> OTOH, it is extremly bad style to use someone else's FQDN in your
> message IDs.

I totally agree, which is why I objected to using "*.gnus.org"
globally as the RHS.

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

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

> Yes, but the entire discussion might become pointless as soon as
> Lars is involved because in the past, Lars has strongly opposed
> running Gnus on machines which do not have an official FQDN.

I will be interested to hear what Lars has to say about this long
winded discussion.

Re: `user-mail-address' for message-ids

[Florian Weimer]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:

> On 28 Feb 2001, Florian Weimer wrote:
> > "Jason R. Mastaler" <jason@libertine.org> writes:
> > 
> >> OK boss, but using ".*gnus.org" in the FQDN is not open for
> >> discussion either unless you happen to be Lars.
> > 
> > Yes, but the entire discussion might become pointless as soon as
> > Lars is involved because in the past, Lars has strongly opposed
> > running Gnus on machines which do not have an official FQDN.
> 
> He he.  But you need Lars' cooperation to be able to use mid.gnus.org :-)

We could use gnus.gnu.org (or even gnus.enyo.de). ;-)

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:

> He he.  But you need Lars' cooperation to be able to use
> mid.gnus.org :-)

Regardless, as the gnus.org maintainer, I object to this strongly, so
please let's arrive at the solution in a different manner.

Re: `user-mail-address' for message-ids

[Jason R. Mastaler]

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

> As long as the Gnus algorithm provides enough entropy, message IDs
> generated by Gnus will not cause collisions.

Can you expand on this a bit further?  In particular, is it feasible
to arrive at the sufficient amount of entropy with the LHS only?

Re: `user-mail-address' for message-ids

[Florian Weimer]

"Jason R. Mastaler" <jason@libertine.org> writes:

> > As long as the Gnus algorithm provides enough entropy, message IDs
> > generated by Gnus will not cause collisions.
> 
> Can you expand on this a bit further?  In particular, is it feasible
> to arrive at the sufficient amount of entropy with the LHS only?

If only entropy is considered, yes.  If we use 130 bits of entropy
(which would result in 26 characters in base32), and there are 2^40
message IDs to be assigned under a single domain, the probability of
at least one collision among all IDs is around 10^-15 (at least I
think so).  If there 2^50 message IDs, it goes up to 10^-9, if there
are only 2^30, it's approximately 10^-21.  If you add four more
characters (or 20 bits), the probability in the 2^50 case is around
10^-15 again.

If a special domain part is used, only collisions among Gnus users
have to be avoided, so if we assume that the random string in the
domain part is reused for each message once it has been generated, and
there are 2^20 users, with 80 bits of entropy (16 characters), the
probability of a collision is around 10^-12.  Together with the rest
of the domain part and the local-part, we have around 35 characters in
a complete.

On the other hand, If we put all the entropy goes into the local-part,
the domain part is probably rather short (people who have FQDN are
going to use the old scheme, I think), and we come out around 35
characters as well.

So if my calculations are correct, there is no purely technical reason
for putting the entropy in the domain part.  However, it's probably
out of question to obtain 130 bits of entropy per posted message.

Re: `user-mail-address' for message-ids

[Paul Jarc]

Florian Weimer <fw@deneb.enyo.de> writes:
> So if my calculations are correct, there is no purely technical reason
> for putting the entropy in the domain part.  However, it's probably
> out of question to obtain 130 bits of entropy per posted message.

There's a ready source of 128 bits per message: an MD5 hash of the
message itself.  Maybe this is an exaggeration - I wouldn't be too
hasty to advocate use of MD5 hashes alone as unique identifiers, given
the rate of message generation on the net, and the less-than-random
content of messages.  Food for thought, though.


paul

Re: `user-mail-address' for message-ids

[Kai Großjohann]

On 28 Feb 2001, Jason R. Mastaler wrote:

> Can you expand on this a bit further?  In particular, is it feasible
> to arrive at the sufficient amount of entropy with the LHS only?

Of course, you could postulate that the lhs be random stuff followed
by `#' followed by something which identifies the host.

Then you could do away with the rhs.

But in essence, you have just replaced `@' with `#' which is kind of
pointless... 

kai
-- 
Be indiscrete.  Do it continuously.

Re: `user-mail-address' for message-ids

[NAGY Andras]

Not following the thread, so someone may have already posted
this... if not, check out http://www.jwz.org/doc/mid.html


Andras

Re: `user-mail-address' for message-ids

[Samuel Padgett]

Kai.Grossjohann@CS.Uni-Dortmund.DE (Kai Großjohann) writes:

> On 27 Feb 2001, Samuel Padgett wrote:
>
> > What if I connect to the Internet using DHCP, and I have a dynamic
> > IP address?
> 
> When asked, the same people reply that you should obtain a FQDN from
> dyndns.org or some similar service.

Hey, cool.  Didn't know that such a service existed.  Thanks.

Sam
-- 
What the large print giveth, the small print taketh away.