Subj : Re: [SpiderMonkey] Serializing JSObject
To   : ggadwa@charter.net
From : Brendan Eich
Date : Wed Jun 23 2004 02:50 pm

ggadwa@charter.net wrote:
> Brendan Eich wrote:
> 
>> Since you are dealing with an implementation of the language here, 
>> namely SpiderMonkey, you might rather use JSVAL_IS_INT, 
>> JSVAL_IS_DOUBLE, etc., than JS_TypeOfValue.
> 
> 
> Ah ... okey dokey.  I use JS_TypeOfValue to filter out the things I 
> don't want to serialize (i.e., functions, and currently, objects, though 
> that will change.)  I use the _IS_ to filter it down further.  That 
> makes sense to me.


You would save cycles by skipping the JS_TypeOfValue filtering.


>> Generally, serialization is domain-specific.
> 
> 
> If you did it a compact way, it would be.


No, I meant it requires knowledge of JSClass-specific private data, 
other JSObjectOps variations on object structure, and in general, issues 
to do with scope chain and other runtime parameters.


> I'm basically generating 
> strings that are ... to a point ... human readable. 


Have you looked at uneval, which is built on toSource for object types?


> Yeah, but everything seems to be more complex then it needs to be :)  
> Surely, there's good reason, but what I think is useful is just two 
> function calls that take a JSObject, return a string, and the opposite.  
> Nothing else to be done.


This sounds awfully like uneval and eval to me.

/be

.