Subj : Re: Help with rooting, JSStrings and JS_DefineFunction
To   : netscape.public.mozilla.jseng
From : Brendan Eich
Date : Wed Dec 01 2004 02:27 pm

Altan wrote:
> When interfacing a C function with JavaScript, JS_DefineFunction is
> used.  Let's assume the following C function is eventually called
> 
> constructInternal(
> 	JSContext *cx, 
> 	JSObject **obj, 
> 	uintN argc, 
> 	jsval *argv, 
> 	jsval *rval)
> {
> 	...
> }
> 
> If one (or more) of the parameters passed in the argv is a JSString
> and the C code will continue to reference this JSString after
> constructInternal returns, do I need to worry about this string being
> garbage collected?  For example, might it be garbage collected if the
> script containing the string that caused constructInternal to be
> executed terminates or under any other conditions?
> 

Sure; consider a caller who says "constructInternal('foo ' + bar)' where 
bar's value is converted to a string if it's not one already, 
concatenated with 'foo ', and passed so that constructInternal receives 
the concatenation in argv[0].  That string is temporary, it need not 
live any longer than the call, since there are no other references to it.


> In general, what are the rules for keeping pointers to arbitrary
> argv's after a C function returns?  Must I use JS_AddRoot in all such
> cases?


That's a sure and safe way (don't forget to JS_RemoveRoot; try to use 
JS_AddNamedRoot for GC_MARK_DEBUG's sake).  There are other ways, such 
as "locking" the GC-thing (JS_LockGCThing), arranging for a root scanned 
by a JSClass.mark hook to protect the string, where the object with that 
JSClass lives as long as you need the string to live, and similar 
techniques.

It helps to think about references forming a directed graph that the GC 
scans, and pondering "what else keeps this string or object from being 
garbage?"

/be
> 
> Thanks in advance for any help.
> 
> ... Altan

.