Subj : [SpiderMonkey] prototypes and millions of objects
To   : netscape.public.mozilla.jseng
From : Brian Barnes
Date : Sat May 15 2004 02:29 am

A couple of people are starting to use my 3D engine (which uses 
SpiderMonkey for scripting.)  They've gone way beyond my original 
demos, and they have thousands of objects loaded into the engine.

This, of course, takes forever to create and eats up a lot of runtime 
memory.

Each "item" in the map has a corresponding js object, which has a bunch 
of children objects.  If you have 500 items in a map, you could easily 
have 5000 objects in the engine.

So, since each of these items has the exact same object model, 
prototypes would be good.

Question #1: My engine depends on private data from the objects to work 
right.  If I have a prototype, and I get a set or get property call, 
will I get the private data from the object or from the prototype (the 
property will be on the prototype.)  If it's the prototype, then I 
can't use them :(

Question #2:  As I'm experimenting, I noticed something.  I do this 
little trick where each class I use I override the "add property" stub, 
and if a flag tells me I'm in creation, then to let adds go through.  
Otherwise, fail them.  This turns off people's ability to add to my 
objects.  This doesn't work when using prototypes because if I use the 
same class for the prototype as I do the real object, the real object 
triggers this add before actually checking the prototype for the 
property!  Is this some optimization?

Question #3: When creating the "real" object (using JS_DefineObject) 
from the prototype, do I need to really set any class since that class 
is all set in the prototype?  Could I use a stubed class (or null, for 
that matter?)  that might solve #2.

[>] Brian

.