[sugar] Re: pygtk performance issue

[Marco Pesenti Gritti]

Mike Hearn wrote:
> On Wed, 06 Sep 2006 12:43:07 +0100, Mike Hearn wrote:
>   
>> My own profiles show most of the time  ...
>>     
>
> Erm, that is my own profiles on my own 1ghz Duron desktop machine. I don't
> have an OLPC board.
>
>   

Running on the OLPC makes a big difference, since open is 10 times slower.

> I get 765 NOENTs on my own box but almost all of these look
> simple to eliminate ... most are from Python or the dynamic linker
> searching various paths for things. Keeping the linker search path as
> small as possible and making Python smarter about finding modules could
> fix it easily.
>   
I'm sure it can easily reduced, I'm less convinced it can be easily fixed.

> But again I wouldn't bother unless you can get an oprofile trace showing
> that it's losing all its time in the kernel file open routines ...
>   

The reduced test case I attached to the mail starting this thread is 
doing only "open".

On the OLPC:

Total "import gtk" time : 0.9 secs
500 opens with my testcase: 0.35 secs
(import gtk cause around 500 opens)

Note that I'm ignoring all the "stat". Still, this seem to be a good 
part of the startup time.

Marco