Date: Sat, 9 Jul 1994 14:51:22 -0400
From: "Steven D. Majewski" <sdm7g@elvis.med.virginia.edu>
Message-Id: <199407091851.AA23653@elvis.med.Virginia.EDU>
To: Rob Austein <sra@epilogue.com>
Subject: Re: sequence numbers ( was: (long) sketch of proposed imap: URL syntax ... )
On Jul 9, 12:36, Rob Austein wrote:
>
> Date: Wed, 6 Jul 1994 13:54:22 -0400
> From: "Steven D. Majewski" <sdm7g@elvis.med.virginia.edu>
>
> There hasn't been enough comment to reliably project a consensus, but
> the early returns seem to soundly reject sequence numbers in an imap
> URL.
>
> I'd phrase it even more strongly than that. The only reason I haven't
> agitated for completely removing sequence numbers from IMAP4 (ie,
> using nothing but UIDs) is to be nice to the people who worry about
> backwards compatability with IMAP2.
>
> In my world view, sequence numbers are an artifact of the user
> interface, and have no place in the protocol. [ ... ]
>
> -- End of excerpt from Rob Austein <sra@epilogue.com>
I *completely* agree with your sentiments with respect to IMAP{2,4},
Rob. It was a mistake in specifying IMAP2 to let the implementation
show thru by using sequence numbers. There should have been an
internal mapping of permanent id's to sequence numbers on the
server ( if needed for effeciency ) that was kept hidden from
clients. It's a mistake that we reluctantly have to live with now.
( I'm not trying to pick on Mark or anyone else by calling it a
"mistake". I wear bifocals, myself, and only have 20/20 vision
in hindsight! :-)
Wrt: sequence numbers in the URL's: I'm willing to deprecate them even
_further_. I still require them in my implementation, but I think their
status should be similar to the notion of partial URL's, which the WWW
servers depend upon quite heavily in actual practice, but the draft
ietf standards have more or less disowned. ( _Partial_Universal_
Resource Locator being considered somewhat oxymoronic, I guess. :-)
I'm still trying to figure out how to hide them from clients, though.
Another possibility suggests itself:
Since I *am* making guarantees of permanence of sequence numbers
by imposing certain constraints in my application ( All folders
will be APPEND access and nothing will be deleted, etc. ) then,
I can call them UID's for the purpose of my gateway, which will
map "UID"s to sequence numbers on the server (which doesn't
support UID's itself. )
However, this means that access that bypasses the gateway
may not be able to access the message with the same URL.
Actually, this points out a more major flaw in using UID's
as part of the locators that is almost as bad as using sequence
numbers. Unless I'm missing something, the UID is guaranteed
across sessions, but ONLY for that particular server. If I
upgrade to a newer version of imap server software, or switch
from washington.edu's server to JGM's, I don't believe I have
any reasonable expectation that UID's from one server will be
conserved on the other.
Without such a guarantee, URL's based on UID selectors are
a bit more permanent than those based on sequence numbers
( and the URL's are NOT defined to be permanent ) but they
are still too unreliable. ( Following JGM's ftp analogy
of "the third file in that directory" : you can usually
install a new ftpd without it changing the names of all
the files on the server! )
So that leaves Message-Id's as, despite their non
guarantee of uniqueness, the best candidate for a
cannonical message selector in an imap URL.
I would, however, unlike sequence numbers, which I
agree should be deprecated to the point, perhaps, of
not even mentioning them, define a UID selector, but
post the proper caveats and warnings.
( I don't quite consider them the "loaded gun" that
someone called sequence numbers, but they still fall
into the "attractive nusiance" catagory! )
- Steve Majewski (804-982-0831) <sdm7g@Virginia.EDU>
- UVA Department of Molecular Physiology and Biological Physics