bug#58472: [PATCH] Make `message-unique-id' less prone to collisions

[Stefan Kangas]

Paul Eggert <eggert@cs.ucla.edu> writes:

> If the goal is to avoid collisions, though, wouldn't it be better to
> avoid the time entirely? time is not very random in the high order bits.

The main goal is to avoid collisions, but using the time also gives an
idea of when the message was sent, which is kind of nice.  Time also
guarantees a somewhat unique value even if the user has happened to set
the random seed.

If we limit the length of the time string to 12 characters, and the
total length to 25 characters (including the ".gnu" part), we still have
a guaranteed 9 characters of random data, or 46 bits of entropy.

    ⌊log₂ 36⁹⌋ = 46

If we want to guarantee 62 bits of entropy, we could do a total length
of 28 characters instead, but that might be overkill.

> Also, the comment about ".fsf" and other newsreaders is wrong [...]
> And if we're changing the algorithm perhaps we should change ".fsf" to
> something else.

Changing it to ".gnu" makes sense to me.

> Something like this, perhaps, where you can choose LEN as you like:
>
>    (concat
>       (let ((len 18))
>         ;; Pass LEN, not -1, to message-number-base36 so that it never
>         ;; returns "" which would make the message-ID nonconforming.
>         (message-number-base36 (random (expt 36 len)) len))
>       ;; Append ".gnu" to advertise that we're GNU Emacs.
>       ".gnu")

I think making sure it is always the same length is a good idea.

How about the attached?

0001-Make-message-unique-id-less-prone-to-collisions.patch
Description: Text Data