Re: org-mode on sloooow computer

[Carsten Dominik <carsten.dominik@gmail.com>]

On Apr 25, 2009, at 12:26 AM, Eric S Fraga wrote:

> Carsten Dominik writes:
>> Hmmm, `org-up-heading-all' still gets called, this should no longer
>> happen.
>> I wonder why, when.
>
> There must have been an error in recompilation after pulling the
> updates with git.  Although I asked emacs to recompile the directory,
> it must have done something strange.  I've cleared out the elc files
> and compiled the whole directory and this is what I get now (which
> indicates yet another improvement!):
>
> org-agenda                             1           1.114424       
> 1.114424
> org-agenda-list                        1           0.889913       
> 0.889913
> org-dblock-write:clocktable            7           0.586827       
> 0.0838324285
> org-get-clocktable                     1           0.427968       
> 0.427968
> org-update-dblock                      1           0.3902330000   
> 0.3902330000
> org-prepare-agenda-buffers             2           0.267907       
> 0.1339535
> org-agenda-get-day-entries             6           0.255334       
> 0.0425556666
> org-agenda-get-restriction-and-command 1           0.224329       
> 0.224329
> org-clock-sum                          6           0.195653       
> 0.0326088333
> org-agenda-get-scheduled               6           0.145536       
> 0.024256
> org-prepare-agenda                     1           0.128176       
> 0.128176
> org-refresh-category-properties        12          0.0830839999   
> 0.0069236666
> org-get-tags-at                        8           0.059022       
> 0.00737775
> org-up-heading-safe                    21          0.0560520000   
> 0.0026691428
> org-get-entries-from-diary             1           0.042899       
> 0.042899
> ...
>
>> If you want to help:
>>
>> 1. reload uncompiled-code:
>>
>>    C-u C-c C-x r
>
> Ummm, this doesn't work for me?  C-c C-x r is undefined.

You need to be in an org-mode buffer for this..

>
>> 2. Arm the function
>>
>>   M-x debug-on-entry RET org-up-heading-all RET
>>
>> 3. TUrn on debugging on error (needed, but hard to explain here)
>>
>>   You can best do this from the "Options" menu, "Enter Debugger on
>> Error"
>>
>> 4. Produce your agenda.  You will get a backtrace buffer, which I
>> would like to see...
>
> I don't get a backtrace buffer, which I guess now makes sense since
> org-up-heading-all doesn't seem to get invoked any more.

yes..

>
>> if you want to hunt for more improvements, looking at
>> this table, much can be gained by turning off the
>> clocktable in the agenda and only turning it on
>> interactively with "R" when you really need it.
>
> Yep, this helps as well:
>
> org-agenda                             1           0.963539       
> 0.963539
> org-agenda-list                        1           0.739349       
> 0.739349
> org-finalize-agenda-entries            1           0.286326       
> 0.286326
> org-agenda-highlight-todo              8           0.282616       
> 0.035327
> org-agenda-get-day-entries             6           0.2574450000   
> 0.0429075000
> org-agenda-get-restriction-and-command 1           0.224016       
> 0.224016
> org-agenda-get-scheduled               6           0.147105       
> 0.0245175
> org-prepare-agenda                     1           0.1283210000   
> 0.1283210000
> org-prepare-agenda-buffers             1           0.1202870000   
> 0.1202870000
> org-get-tags-at                        8           0.059197       
> 0.007399625
> org-up-heading-safe                    21          0.056139       
> 0.0026732857
> org-get-entries-from-diary             1           0.043058       
> 0.043058
> ...
>
> However, I may actually leave the clock display on by default as it
> has a psychological effect that is worthwhile for my working
> practices; but at least I know that turning it off might be worthwhile
> on the Asus.

Of course, this i what I meant.

>
> In any case, I am impressed: from 4.5 seconds to less than a second in
> one day!  Can you do the same for my 2+ day optimisation runs I
> require in my research? ;-)

:-)  Someone needs to write a profiler for that.  In fact, clocking your
work and looking at the biggest time consumers may help :-)

Thanks, this was a fun day.


- Carsten