Re: That newfangled IMAP thing...

[Ted Zlatanov <tzz@lifelogs.com>]

On Sun, 12 Sep 2010 04:17:13 +0200 Lars Magne Ingebrigtsen <larsi@gnus.org> wrote: 

LMI> Anyway, the splitting methodology would be as follows:

LMI> 1) stream-download all the articles in the INBOX group
LMI> 2) run the mail split rules over each article
LMI> 3) notice what new groups should be created, if any, and create them
LMI> 4) stream out a bunch of "COPY 23242 ding" (etc.) commands, to copy the
LMI>    articles over to the correct groups
LMI> 5) parse the responses
LMI> 6) mark all the messages that were successfully copied as deleted (this
LMI>    would normally be all the articles)
LMI> 7) do an "EXPUNGE"

LMI> It should be a fairly quick process -- the main time spent should be in
LMI> the actual streaming download of the articles in 1).

Can you make (3) and (4) also work for qualified group names outside the
nnimap server, so splitting can work across servers and backends?

On Sun, 12 Sep 2010 04:32:02 +0200 Lars Magne Ingebrigtsen <larsi@gnus.org> wrote: 

LMI> Ted Zlatanov <tzz@lifelogs.com> writes:

>> If Lars doesn't do it I'll keep it in mind.

LMI> So far I've only done a totally synchronous (but streaming) IMAP
LMI> interface.  But to support gnus-async stuff (for article pre-loading),
LMI> I'll have to add some process filters and stuff, and perhaps it would be
LMI> natural to have the async connection return to IDLE-ing on the INBOX
LMI> whenever it doesn't have any work to do?  And then add some callbacks to
LMI> notify.el.

s/notify/notifications/ apparently, but yes.  I wouldn't put too much
work into this.  It's less valuable than the functional improvements
you're making.

LMI> I'm looking forward to using the built-in TLS stuff you're doing for
LMI> Emacs 24, though, because now (with four imaps nnimap servers), Gnus
LMI> startup is overwhelmingly dominated by "openssl s_client" startup (via
LMI> tls.el).  The built-in stuff will be much faster?  Right?  :-)

It will be much faster.  It's a lot of work, though.  There's no "open
this connection and leave me alone" option.  You have chains of
certificates, examples full of "it's easy to make this work right,
if you could move to Greenland" and so on.

I plan to be able to open an unverified SSL or TLS connection before
dropping it into the core.  From that point on the internals may change
but I'll try to keep the API consistent.  Meanwhile, suffer :)

Ted