Subj : Re: Equivalent of JS_InitClass for global object?
To   : "John J. Lee" <jjl@pobox.com>
From : Brendan Eich <brendan@meer.net>
Date : Fri Aug 22 2003 05:24 pm

John J. Lee wrote:

>From the point of view of the JS interpreter, my "usual machinery"
>does this:
>
>1. malloc a JSClass for the proxy.
>

Are you doing this for each and every proxy?  I hope not.  Why can't you 
use a static JSClass shared among all proxies, or at most one JSClass 
for each set of proxies of the same Python class?

>2. Initialise the JSClass struct with stub methods, except for
>   get/setProperty (which delegate to the Python interpreter).
>3. JS_InitClass(
>       cx, global_object, NULL, jsclass, constructor_stub, 0,
>       NULL, NULL, NULL, NULL);
>

Do you need step 3?  What does constructor_stub do?

>4. JS_DefineObject(cx, global_object, name, jsclass, NULL,
>                   JSPROP_READONLY);
>

Do all proxies have unique names in the global object?

>Obviously, 3 doesn't carry over to the case where the Python object
>I'm trying to bind to JS is the JS global object itself.
>

Neither would 4.

>So, what should I do instead?  I tried the following in place of 3 and
>4 above:
>
>3b. global_object = JS_NewObject(cx, jsclass, NULL, NULL);
>

This will work -- you need not call JS_InitClass to use a JSClass 
struct.  You need to call JS_InitClass only if you have a class 
constructor, and want it to be named by the class name, and to have a 
class prototype; or, if you pass null for the constructor, to have the 
class name denote the class prototype object (e.g., Math).

>4b. JS_DefineProperty(cx, global_object, name,
>                      OBJECT_TO_JSVAL(global_object), NULL, NULL, 0);
>

Why would you make a synonym in the global referring to itself?  You 
would if you want DOM level 0 properties such as "window" and "self" to 
work (and "top" in the topmost window in a frameset/frame hierarchy).  
So that must be what's going on here, but it seems like a different case 
from the original step 4.

>So, to summarise, instead of:
>
>global_object = make_global_object(cx);
>JS_InitStandardClasses(cx, global_object);
>
>
>I've now got steps 1, 2, 3b, 4b, followed by JS_InitStandardClasses,
>as usual.
>

Do you check the return value of JS_InitStandardClasses?  Is it perhaps 
failing?

>When I do that, though, binding another Python class (steps 1 to 3
>above) stops working -- eg. if I bind a Python class named "foo" and
>then evaluate this:
>
>var f = new foo();
>
>
>spidermonkey complains:
>
>TypeError: foo is not a constructor
>

What is foo?  Print or otherwise dump typeof foo, uneval(foo), etc. to 
get an idea.

>It seems my JS_InitClass on foo doesn't work when the global object is
>created with NewObject.
>

No, that's not the problem.   The problem may lie in your proxy class's 
getProperty and setProperty hooks, especially in setProperty.  What do 
those look like?

>  Another symptom is that, though the global
>object has a "window" property, it doesn't delegate to Python --
>rather, I get:
>
>TypeError: window has no properties
>

For what expression do you get that error?  You must be dereferencing 
window.foo or some such, and window's value is undefined or null.  
Again, I need to see what the proxy class's getProperty and setProperty 
hooks look like.

>I don't understand that, because window was JS_DefineProperty'd to the
>global object, whose class has get/setProperty functions that call
>back to Python.
>

That's ok, but if your getProperty hook clobbers *vp with JSVAL_VOID, 
you would get the above outcome.

>What am I doing wrong?
>  
>

Insufficient data, as the computers on old sci-fi movies would say; I 
need to see the relevant code to say more.

/be

.