Re: [GSWHackers] Future of GnusStep WebObject

[Helge Hess]

On Mar 2, 2004, at 4:27 PM, NeXT wrote:
> > It will be a significantly slower, but is going to provide all the 
> > KVC features available in gstep-base.
>
> Why should it be "significantly" slower? It could not be cached, or 
> KVC itself is so slow?

Well, KVC only works with objects, which has a significant speed hit if 
you do a lot of KVC on base types, which is very common with WO. Of 
course KVC itself is not inherently slow.

For example a conditional used in a repetition on thousands of objects:
Component.wod:
  Condition: WOConditional {
    condition = showImage;
  }
Component.m:
  - (BOOL)showImage {
    return YES;
  }
WOConditional.m:
...
  if ([[self->condition valueInComponent:[context component]] 
boolValue])

Since there is no -boolValueForKey:, the naive implementation of 
WOKeyPathAssociation will lookup the method, turn the (char) result 
into a NSNumber object, then in the dynamic element will in turn 
convert the number into a char (/BOOL) for the actual check (notably 
this is also hit by the fact that we can't detect BOOL here which 
allows for further optimizations). This is about 2-10 times slower than 
directly accessing the bool result, do this in loops and see ;-)
One of the optimizations in WOKeyPathAssociation, but there are others.

Some years of profiling already went into SOPE.

> PS)
> I'm going to test more extensively (with some real load) SOPE and 
> GSWeb, I'm searching for all default options and documentations on 
> both of them. Does anybody know good source of information? (Yes, 
> source is very good source of information, but you know :-)

On SOPE:
  http://www.opengroupware.org/en/devs/sope/
  http://docs.opengroupware.org/Members/helge/SOPE/

Not really much documentation though. But better than nothing ;-)

regards,
  Helge
--
http://docs.opengroupware.org/Members/helge/
OpenGroupware.org