Re: [Pyobjc-dev] Re: 1.0?

[Ronald O. <ous...@ci...>]

On Wednesday, Jun 18, 2003, at 22:10 Europe/Amsterdam, Ronald Oussoren 
wrote:

>
> On Wednesday, Jun 18, 2003, at 21:55 Europe/Amsterdam, Jack Jansen 
> wrote:
>
>>
>> On woensdag, jun 18, 2003, at 20:21 Europe/Amsterdam, Ronald Oussoren 
>> wrote:
>>> We already build minimally initialized classes, the problem is that 
>>> [NSBundle bundleForClass:] seems to be very expensive. That is at 
>>> least what I remembered from the previous time this came up, I think 
>>> bundleForClass did a lot of I/O for every class we look at. At the 
>>> time there were more important issues, but I think this is the major 
>>> technical issue with the bridge at the moment.
>>
>> How many bundles are we talking about, usually? Cocoa and the main 
>> application bundle? If so, then could we keep those two handy, and 
>> use [bundle classNamed: xxx] as a quick test to see whether the class 
>> is in there?
>>
>> That does assume that classNamed: is cheaper than bundleForClass:.
> It is worth testing if classNamed: is cheaper. If it is cheaper we can 
> use this method instead of bundleForClass.

It isn't. I saw no noticable improvement when switching to this method. 
I did think of an alternative:
I've checked in a patch that adds an argument to the function that 
actually loads frameworks. This argument is a list of classnames. If a 
class is in the list it is assumed to be in the framework, and won't be 
subjected to an expensive lookup. This seems to help a little on my box.

Please let me know if this is an improvement.

Ronald