discuss-gnustep
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [GSWHackers] Future of GnusStep WebObject


From: David Ayers
Subject: Re: [GSWHackers] Future of GnusStep WebObject
Date: Fri, 05 Mar 2004 10:02:46 +0100
User-agent: Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.6) Gecko/20040113

Helge Hess wrote:
On 03.03.2004, at 14:34, David Ayers wrote:

Interesting... I actually prefer the approach that gnustep took in solving the main issue within -base/NSNumber which returns cached versions common NSNumber instances. This solves the root of this particular problem for a wider range of applications. Yet optimization within GSWeb and GDL2 is needed.


This is a different optimization only solving a small part of the problem (actually its not really related to the issue that base types must be piped through objects). It still has significant method call overheads:
a) it has to call a method to create an object or retrieve a cached one for
   a base type
b) it has to call the actual method
c) it has to turn the object back into a base type
At least three times slower for a very common case (in practice you need to add cache management overhead etc).

This is only one of the association optimizations done in SOPE, there are various others. Feel free to look into the source, though fast code is not necessarily readable (but well worth the effort the more complex code due to the high usage of this code).

I think you counted a method invocation or two too much there. But like I said, we still need optimization. And thanks, I'll probably take you up on the offer to look a some of the techniques you used, once I'm at the stage of optimizing GSWeb. Currently my priorities are oriented at higher API compatibility wrt WO4.5.

Cheers,
David





reply via email to

[Prev in Thread] Current Thread [Next in Thread]