Subj : Re: [SpiderMonkey] Confused about class property
To   : netscape.public.mozilla.jseng
From : Peter Paulus
Date : Mon Oct 10 2005 06:02 pm

Brendan Eich wrote:
> Peter Paulus wrote:
> 
>> Brendan Eich wrote:
>>
>>> Peter Paulus wrote:
>>>
>>>>
>>>>
>>>> Not just to start a monologue, I've disassembled the javascript:
>>>>
>>>> var x = ColorSpace.lab;
>>>>
>>>> into:
>>>>
>>>> 00000:   0  defvar "x"
>>>> main:
>>>> 00003:   0  bindname "x"
>>>> 00006:   0  name "ColorSpace"
>>>> 00009:   0  getprop "lab"
>>>> 00012:   0  setname "x"
>>>> 00015:   0  pop
>>>>
>>>> So I guess it must be the interpreter that turns ColorSpace into a 
>>>> Function object, rather than an 'ColorSpace' object.
>>>>
>>>> Are there flags on my JSClass struct or otherwise to manipulate this 
>>>> behavior?
>>>
>>>
>>>
>>>
>>> If you don't want your call to JS_InitClass to bind a constructor to 
>>> the class name in the global object, pass null for the |constructor| 
>>> formal parameter.  Then the class name will denote a prototype object 
>>> of your class, containing the properties and methods (not the static 
>>> properties and methods) you specify.
>>>
>>> /be
>>
>>
>> I'm not sure I want to bind a constructor. I've looked at the 
>> constructor as a convenient place to bind private data to an object 
>> (JS_SetPrivate()).
> 
> 
> 
> Sure.
> 
> 
>> Up till now I've been passing the constructor function to 
>> JS_InitClass() and was still able to handle JSBool 
>> getClassProperty(JSContext*, JSObject*, jsid, jsval*) for any class. 
>> This has worked fine, since every JSClass pointed to it's own static 
>> JSBool getClassProperty(JSContext*, JSObject*, jsid, jsval*) function. 
> 
> 
> 
> That signature looks like JSObjectOps.getProperty, not 
> JSClass.getProperty (note the jsid third argument).  Are you talking 
> about JSClass here, or JSObjectOps?
> 
I'm talking about JSClass. The correct signature is static JSBool 
getClassProperty(JSContext*, JSObject*, jsval, jsval*);. I did it off 
the top of my head and made a mistake in the  third parameter. I'm sorry 
about causing this noise.

> 
>> Implicitely, when this function is called, you know for which class 
>> this is. There was never a need to examine the JSObject* in this case. 
>> Based on the id I could return an appropriate result.
> 
> 
> 
> This is true of instance properties, not of "class static" properties of 
> the constructor.
> 
> 
>> In my current project I'm trying to build classes dynamically in 
>> SpiderMonkey from Adobe InDesign's ExtendScript, in order to use their 
>> native objects from Spidermonkey. In this setup I have exactly one 
>> static method: JSBool getClassProperty(JSContext*, JSObject*, jsid, 
>> jsval*) to serve all classes. Now I was suprised to find that 
>> JSObject* pointed to a "Function" object.
> 
> 
> 
> That's what a constructor object is, though.
> 
> 
>> If I omit the constructor function with JS_InitClass() indeed the 
>> getClassProperty() contains the right kind of object. But then I have 
>> lost my place to augment a JSObject with my private data.
>>
>> Can you advise me of a suitable place to add private data to a JSObject?
> 
> 
> 
> Use constructors for that.
> 
> You need to show actual code.  It's not clear how you were using 
> getClassProperty for all classes, or whether it was a JSObjectOps hook 
> or a JSClass hook, or why you couldn't use the static property specifier 
> argument to JS_InitClass.
> 
> /be
I'm preparing a zip archive which I can hopefully send you tommorrow.

By 'using getClassProperties() for all classes' I mean: On all 
JSPropertySpecs passed to JS_InitClass() the getter functions points to 
a single C++ function. I do not have an getter function specific for a 
certain class on every JSPropertySpec, as would be the usual case. It is 
just a single method across all classes.

With kind regards,
Peter Paulus

.