Subj : Re: JS_GC question!
To   : richardl <xinyul@prediwave.com>
From : Brendan Eich <brendan@meer.net>
Date : Thu Dec 11 2003 12:11 pm

richardl wrote:

> I embedded spidermonkey to a mini-browser. It works but:
> 
> Having problem when running the following script:
> 
> <SCRIPT>
> function testMem()
> {
>  for (var i=0; i<1000; i++ )
>  {
>   var foo = new Function();
>   delete foo;

First, delete does not free memory allocated by new, as in C++.  It 
removes a property, if possible.  Its result is true if the property 
didn't exist or was removed, false if the property is permanent (can't 
be deleted).  Note that properties created by var declarations are in 
fact permanent, so this delete foo; statement is useless.

>  }
> }
> </SCRIPT>
> <Input type="button" value="click for memory test"  onclick="testMem()">
> 
> After I clicked twice the button, out of memory occured!

See JS_SetBranchCallback.  You should be using it to set a branch 
callback that calls JS_MaybeGC on every 0 mod N branch-counter sample, 
where N is a large power of two (and the counter is unsigned).

> So I have to put  JS_MaybeGC(cx) after each EvaluateScript() call.

You don't have to do that.

> It helps
> to gain back the memory, but I found it is not stable any more.
> 
> It  often causes crash in
> js_FindProperty(JSContext * 0x0126e640, long 19227776, JSObject * *
> 0x00faf630, JSObject * * 0x00faf62c, JSProperty * * 0x00faf5b0) line 1439 +
> 33 bytes
> 
> The statement is      if (!OBJ_LOOKUP_PROPERTY(cx, obj, id, &pobj, &prop))
>   return JS_FALSE;
> 
> obj is not a valid JSObject structure. It looks like the memory is
> corrupted.

Are you using GC roots properly?  See 
http://www.mozilla.org/js/spidermonkey/gctips.html and related docs in 
the same directory.

/be

.