Daniel Pittman <daniel@rimspace.net> wrote:
> Elias Oltmanns <oltmanns@uni-bonn.de> writes:
>
>> Daniel Pittman <daniel@rimspace.net> wrote:
>>> Elias Oltmanns <oltmanns@uni-bonn.de> writes:
>>>
>>>> accessing a 25MB nnfolder takes at least twice as much time and
>>>> sometimes even up to four times as much when running No Gnus from
>>>> within emacs-snapshot (Debian packaged cvs version) than running in
>>>> emacs 21.4. The reason I'm reporting this here is that visiting the
>>>> file with C-x C-f directly from emacs does not reveal such a
>>>> difference in speed. Yet, every operation performed by gnus on this
>>>> nnfolder, as entering, reading articles, expiring, etc, challenges my
>>>> patience, to say the least. 
>>>
>>> Well, the best thing to do is probably to profile the operation of
>>> visiting a folder, allowing us to see exactly where time is spent.
>>
>> Here are the profiles. I went about as follows:
>> - Start up emacs2[12],
>> - load elp.el,
>> - M-x set-varialbe RET elp-sort-by-function RET elp-sort-by-average-time RET,
>> - start up gnus,
>> - M-x elp-instrument-package RET gnus RET,
>
> You probably want to instrument 'nn*' here as well

Done now.

[...]
>
>> - enter folder,
>> - M-x elp-results.
>>
>> Note that there are no delays due to disk access as all data had been
>> in cache already.

New profiles attached. There is one striking thing:
nnfolder-existing-articles appears amongst the most time consuming
functions only in emacs 22 so there has to be some construct in the
function definition which behaves differently when executed in the two
emacs versions. So far I managed to rule out search-* and
re-search-*. Still, I have the feeling that it might be one of those
seeking commands, however, I didn't manage to profile goto-char which
was my next guess.

Any ideas?

Elias

---- Profile emacs 21.4.1 ----