Subj : Re: Spidermonkey: parameters to 'construction' function
To   : Peter Paulus <peter@neroc.nl>
From : Brendan Eich <brendan@meer.net>
Date : Fri Apr 16 2004 12:01 pm

Peter Paulus wrote:

> // Class initialization function
> void Document::initialize(::javascript::core::Context& context)
> {
> _prototype = context.registerClazz(Node::_prototype, &clazz, 
> &createDocument, 0, (JSPropertySpec*) &objectProperties, 
> (JSFunctionSpec*) &objectMethods, (JSPropertySpec*) &classProperties, 
> (JSFunctionSpec*) &classMethods);


Why all the casts?  Are you casting away const?


> As far as I understand this my Document::createDocument() method is 
> invoked by the following JavaScript statement:
> 
> 1) var document = new w3cDOM·Document();
> 
> aswell as
> 
> 2) var document = w3cDOM·Document();
> 
> In case 1 both the JSObject* 'object' parameter and the jsval* 'result' 
>  parameter both point to the same instance of an w3cDOM·Document class.
> 
> I've verified this as follows:
> 
> JSClass* objectClass = JS_GetClass(object);
> cout << objectClass->name << "\n";


Right.  You can use JS_IsConstructing(cx) to distinguish the two cases, 
by the way.


> In case 2 the JSObject* 'object' parameter and jsval* 'result' differ. 
> In this case the 'object' parameter points to the 'global' object that 
> got registered with the runtime for this context.
> 
> The 'result' parameter seems to be a JS_IS_VOID parameter.


You mean JSVAL_VOID.


> Now JS_SetPrivate(...) fails with the assertion:
> Assertion failed: OBJ_GET_CLASS(cx, obj)->flags & JSCLASS_HAS_PRIVATE, 
> at jsapi.c:1404. This is correct since my 'global' object has no flags 
> specified in JSClass.flags.
> 
> In both cases JS_IsConstructing(context) is true. (When will this be 
> false??).


Not so.  JS_IsConstructing(cx) returns false for case 2.  Please show a 
*minimal* testcase, preferably based on js.c, demonstrating otherwise. 
But there is no such bug in the engine; many built-in constructors, such 
as Date and Array, test the same predicate that JS_IsConstructing tests 
in order to do something different when called as a function.


> Would it be a correct approach the define a 'constructor' function as 
> follows:


No.  You are writing unnecessary code to work around a bug in your 
embedding, or in the version of the engine you are using.  There is no 
need to test JSVAL_IS_VOID(*rval) once you know JS_IsConstructing(cx) 
evaluates to true.

/be

.