Severe performance regression in NoGnus

Severe performance regression in NoGnus

[James Cloos]

Comparing NoGnus (from gnus' cvs) and v5.x (from emacs' cvs) I see a
massive regression in startup time.  It takes No about *two hours* to
start up, whereas 5.x takes only a few tens of minutes.

I did a restart with functions matching '^gnus' and '^nn' instrumented
and found that several functions were called more often than should be
required.

Each of nnmh-request-list, gnus-request-scan, nnmh-request-scan and
nnmail-get-new-mail were called thrice; gnus-request-newgroups was
called twice.  The full list of calls that took over a minute are:

,----
| Function Name                     Call Count  Elapsed Time  Average Time
| ================================  ==========  ============  ============
| nnmh-request-list-1               23136       17625.744430  0.7618319687
| gnus-group-restart                1           8920.153097   8920.153097
| gnus                              1           8918.801146   8918.801146
| gnus-1                            1           8916.793688   8916.793688
| gnus-setup-news                   1           8894.381629   8894.381629
| nnmh-request-list                 3           8872.079321   2957.3597736
| gnus-find-new-newsgroups          1           5563.511609   5563.511609
| gnus-ask-server-for-new-groups    1           5563.461238   5563.461238
| gnus-request-newgroups            2           5554.888336   2777.444168
| nnmh-request-newgroups            1           5553.105174   5553.105174
| gnus-read-active-file             1           3321.319827   3321.319827
| gnus-read-active-file-1           2           3321.3077359  1660.6538679
| gnus-request-scan                 3           1671.766675   557.25555833
| nnmh-request-scan                 3           1671.766453   557.25548433
| nnmail-get-new-mail               3           1671.766402   557.25546733
| nnmail-activate                   1           1671.51756    1671.51756
| gnus-request-list                 1           1647.456813   1647.456813
| nnheader-directory-files          23136       1099.5671810  0.0475262439
`----

(There are 7703 directories under ~/Mail/Folders.)

I'm working on a better setup (using the filesystem as a database is
getting long in the tooth, much as it did with inn(8) some years ago).

Until then, however, it would be cool to be able to restart w/o such a
long delay.

Diffing the two trees hasn't yet led me to the culprit.  

Going back and forth between the two version is also a problem; in any
group where anything was cached by 5.x, No is unable to open any but
the first such article shown in the *Summary* buffer.  Any attempt to
pull up or operate on any such article jumps to the first one instead.
Even things like removing the marks.

Also, any unread and unmarked mail disappears when moving to No.

So I'm highly motivated to stay with No and not bounce between them.

Does anyone have any thoughts on this?

-JimC
-- 
James H. Cloos, Jr. <cloos@jhcloos.com>

Re: Severe performance regression in NoGnus

[Lars Magne Ingebrigtsen]

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

> Comparing NoGnus (from gnus' cvs) and v5.x (from emacs' cvs) I see a
> massive regression in startup time.  It takes No about *two hours* to
> start up, whereas 5.x takes only a few tens of minutes.

Has this cleared itself up by now?

> | nnmh-request-list-1               23136       17625.744430  0.7618319687

[...]

> (There are 7703 directories under ~/Mail/Folders.)

Wow.  :-)

-- 
(domestic pets only, the antidote for overdose, milk.)
  larsi@gnus.org * Lars Magne Ingebrigtsen