Subj : Re: questions of js_GC
To   : lyg <liyg@tech.coretek.com.cn>
From : Brendan Eich <brendan@meer.net>
Date : Wed Mar 24 2004 11:43 am

lyg wrote:
> Hi,
>    In my app, I implemented a Global object named GlobObj.
> and use
> JS_AddRoot(cx, &cx->globalObject);

For the last time, stop including header files other than jsapi.h.  You 
should *not* be including jscntxt.h.  Worse, there is no need to add a 
root for cx->globalObject -- it's already a root-by-definition that the 
GC scans!

> 
> In the following,
> I use JS_NewObject(....) to create a fieldObj, and then
> 
> JS_SetProperty(cx,GlobObj, "field", &(jsval)fieldObj).

Stop casting till your code compiles.  Use the right typed variable, and 
use the JSVAL macros.

(Apparently you are using C++.)

> optionsObj = JS_NewArrayObject(.......)
> JS_SetProperty(cx, FieldObj, "options", &(jsval)optionsObj)
> 
> In this case, can I protect the optionsObj not to be collected by js_gc() in
> using js_addroot() function?

There is no need to do that if FieldObj is well-connected to the live 
object graph.  You don't root every object, you root only those that are 
not named by properties in other well-connected objects.

> In my app, why the optionsObj->map->slots[0] and slots[1] ... are changed to
> 0xcdcdcdcd or 0x10101010 after js_gc()?

Because you have rooting bugs.

/be

.