[EAI] OT: Message-IDs vs. hashes

[Frank Ellermann <[e-mail]>]

On 3 October 2011 23:51, Julien ÉLIE wrote:

 [FWIW]
> I do not think there would be any major issue with INN (a news
> server) because message-IDs are internally "hashed" after having
> parsed them.  They can be retrievable.

That is an implementation detail.  At least in theory Message-IDs
are "globally unique forever", and a finite set:  The unique part
(LHS) and the RHS have known limitations such as LDH labels in the
RHS.  It is not required to hash the Message-IDs, they can be used
"as is" for their purposes.

OTOH any hash has collisions, by definition.  For various reasons
this WG allegedly decided (= as noted by the Chairs) to modify the
Message-ID syntax.  One reason I'm aware of is that "unique part"
always happened to have the same syntax as "local part", only the
semantics was slightly different.

That could create two similar messes:  Mailbox names (local parts)
with different Unicode "spellings", ditto Message-IDs.  The "KISS"
solution to avoid this potential mess for Message-IDs was rejected:

Folks claimed that nobody would get this right, IOW, it will be a
mess anyway, therefore the old "same syntax" approach will be good
enough.  Normally I'd prefer one big mess (= WG proposal) instead
of two different smaller messes.  But actually I don't think that
nobody would get this right (= only one smaller mess for the local
parts not affecting unique parts).

As long as you treat Message-IDs as opaque and don't try to modify
any Unicode points the new "EAI-IDs" have the same features as the
NetNews message-IDs.  Notably they are not supposed to be hashes.

-Frank