[Dev] Profile of startup

[Nicholas Bastin]

On Apr 5, 2004, at 6:18 PM, John Anderson wrote:

> While fixing a bug, I was doing some hotshot profiling and thought it 
> would be interesting to profile release Chandler startup, when a 
> repository already exists, to see if there was anything interesting 
> that might explain our slow startup time. Surpriseingly, I discovered 
> that the routine that we spent the most time in was warn (warnings.py) 
> and that the vast majority of other routines in the profile listing 
> were repository related. Here's the output for those of you who are 
> interested.
>
> John
>
>
>         631694 function calls (618330 primitive calls) in 40415.812 
> CPU seconds

Seriously, does your chandler startup really take 11 hours, and if so, 
when did you cross-compile to C64?  :-)  I don't think that I need to 
go on a rant on how inaccurate the HotShot stats really are.  HOWEVER, 
in general I've found that the hotshot times are useful in a relative 
sense, so in that respect, you still have a problem.  It seems to me 
that 11000 calls to warn() is a bit excessive, plus warnings.py was a 
source of some discussion at PyCon for optimization, so doing something 
about that would certainly help the startup case.  Do you really need 
warnings, or can you get by with something simpler?  Also, I've been 
looking at rewriting warnings as a builtin to avoid pulling in 
linecache and re (and related sre madness), but maybe you guys will 
want to take that torch and run with it since I really don't have the 
time right now...

--
Nick