Subj : Re: differnce btwn JS_AddRoot and JS_LockGCThing???
To   : netscape.public.mozilla.jseng
From : Aaron Reed
Date : Thu May 27 2004 04:38 pm

Aaron Reed wrote:
> Hey,
> 
> Ok, we compile a function using scriptContext->CompileFunction and then 
> we want to prevent the returned funcObj from being garbage collected. If 
> we us JS_LockGCThing, it is protected just fine.  If we use JS_AddRoot 
> (since JS_LockGCThing is deprecated), this throws an assertion in 
> gc_root_marker.  gc_root_marker is checking to see if the item was 
> allocated from the gcArenaPool.  However what it is checking isn't the 
> object that we rooted, but rather the return value from 
> JSVAL_TO_GCTHING(v).  And it isn't marking the obj that we rooted 
> either, it is marking the return value from JSVAL_TO_GCTHING(v).
> 
> So why the difference?  Why doesn't gc_lock_marker care about the JSVAL? 
>      Should we just go ahead and use JS_LockGCThing?  If we should use 
> JS_AddRoot, how do we ensure that we are marking the right thing?
> 
> Thanks for your help,
> --Aaron

If you are interested, looks like JS_AddRoot takes a pointer to the 
value that you really want to keep around.  JS_LockGCThing takes the 
value that you want to keep around.  Make sure if you use JS_AddRoot 
that you keep the pointer around (alloc it, don't use local stack value) 
or you'll probably take an exception in gc_root_marker later on when 
that memory is reused.  That is where I was screwing up.

.