Re: gnus-group-find-new-groups vs nnimap

[Simon Josefsson <simon@josefsson.org>]

James Cloos <cloos+math_uh-ding@jhcloos.com> writes:

> One issue I've seen with nnimap is that, when it is asked for new
> groups (either at gnus startup or due to an explicit call to
> gnus-group-find-new-groups) it EXAMINEs *every* group returned by
> the call to LSUB, rather than just the new groups.  That means that
> at startup EXAMINE is called on every known group *twice*.
>
> The time loss for doing that is significant for me.  On the order
> of hours.
>
> The code looks like:
>
> ,----( from gnus/lisp/nnimap.el )
> | (deffoo nnimap-request-newgroups (date &optional server)
> |   (when (nnimap-possibly-change-server server)
> |     (with-current-buffer nntp-server-buffer
> |       (gnus-message 5 "nnimap: Listing subscribed mailboxes%s%s..."
> |                     (if (> (length server) 0) " on " "") server)
> |       (erase-buffer)
> |       (nnimap-before-find-minmax-bugworkaround)
> |       (dolist (pattern (nnimap-pattern-to-list-arguments
> |                         nnimap-list-pattern))
> |         (dolist (mbx (imap-mailbox-lsub (cdr pattern) (car pattern) nil
> |                                         nnimap-server-buffer))
> |           (or (catch 'found
> |                 (dolist (mailbox (imap-mailbox-get 'list-flags mbx
> |                                                    nnimap-server-buffer))
> |                   (if (string= (downcase mailbox) "\\noselect")
> |                       (throw 'found t)))
> |                 nil)
> |               (let ((info (nnimap-find-minmax-uid mbx 'examine)))
> |                 (when info
> |                   (insert (format "\"%s\" %d %d y\n"
> |                                   mbx (or (nth 2 info) 0)
> |                                   (max 1 (or (nth 1 info) 1)))))))))
> |       (gnus-message 5 "nnimap: Listing subscribed mailboxes%s%s...done"
> |                     (if (> (length server) 0) " on " "") server))
> |     t))
> `----
>
> As far as I can tell, at the point just before (nnimap-find-minmax-uid
> mbx 'examine) is called, I'd need to search for (mbx) in the current
> buffer and skip the (insert) if it is found, yes?

Does the buffer really contains the same group information twice at that
point?  Note that the function calls (erase-buffer).  So I suspect you
won't find the information that nnimap-find-min-maxuid generates in the
buffer.  You could cache that information in some variable, but then you
run into stale-cache issues.

/Simon