Subj : Re: Garbage Collection Woes
To   : Stephen Jones <Stephen@Jones-net.com>
From : Brendan Eich <brendan@meer.net>
Date : Fri May 02 2003 01:21 pm

Stephen Jones wrote:

>Hi Brendan, thanks for the feedback.
>Iterating over all contexts, I am definately seeing different objects GC'ed
>for each context I pass in.
>Should I assume:
>
>1. Objects are being incorrectly GC'ed when I pass in different contexts, or
>2. Objects that are being GC'ed when passing through one context, release
>other objects that are GC'ed when the next context is iterated.
>
>I'm definately passing the address of JSObjects to be rooted. I have one
>base C++ class that handles this.
>  
>

See dbradley's followup.

>I've temporarily solved this problem by ref-counting my objects and calling
>RemoveRoot when the ref-count is zero, but I'd like to figure out a real
>solution.
>  
>

If you have a JSObject * member of some data structure, let's call it 
myDataStruct->mJSObject, you must pass it to JS_AddNamedRoot by reference:

     JS_AddNamedRoot(cx, &myDataStruct->mJSObject, 
"myDataStruct->mJSObject");

If you have several calls to this function for the same address of a 
JSObject pointer, then you are wasting cycles on all but the first call, 
and indeed, you need a reference count to avoid removing the root until 
the count goes to zero.  But you should be calling JS_AddNamedRoot 
exactly once per unique address of a stable JSObject pointer, and 
JS_RemoveRoot before that storage becomes unstable (goes free, gets 
recycled).

/be

>Thanks,
>- S
>
>"Brendan Eich" <brendan@meer.net> wrote in message
>news:3EB1C5E9.60802@meer.net...
>  
>
>>Stephen Jones wrote:
>>
>>    
>>
>>>One more piece of information . . . this only seems to happen to objects
>>>created in contexts other than the global context  . . . am I doing
>>>something wrong when executing GC? This is the code I am using . . .
>>>
>>>while( ( theCurrentContext = JS_ContextIterator( m_JSRunTimePtr,
>>>&theContextIterator ) ) != NULL )
>>>{
>>>   JS_GC( theCurrentContext );
>>>}
>>>
>>>Is the "live-thing" graph different for different contexts?
>>>
>>>
>>>      
>>>
>>No.
>>
>>Therefore you don't need to call the GC more than once per runtime,
>>using any context created in that runtime.
>>
>>You shouldn't add a bunch of JS_AddNamedRoot calls without knowing why
>>you need each one.  Remove all that you don't believe you need.
>>
>>Make sure you aren't passing object pointers or jsvals by value to
>>JS_Add*Root.  You have to pass the address of a JSObject* or jsval to
>>register a root.
>>
>>/be
>>
>>    
>>
>
>
>  
>

.