Subj : JS_AddNamedRoot() and finalizer
To   : netscape.public.mozilla.jseng
From : Matthew Mondor
Date : Mon Aug 23 2004 10:00 am

Hi,

Using SpiderMonkey from C, I am creating a test class called API which
has to be almost immutable (although not totally as immutable as a
scealed one, since there are a few voluntary read-write properties as
part of it). This class object is protected from GC collection by
using JS_AddNamedRoot(), and normally then unlinked using
JS_RemoveRoot().

However, in an attempt to clean up the program, and using a function
similar to JS_CreateStandardClasses() to create this class, I
attempted to also include the JS_RemoveRoot() as part of a custom
finalizer method of that class, instead of needing a corresponding
function which only calls JS_RemoveRoot(), or calling it explicitely,
before calling JS_DestroyContext() and JS_DestroyRuntime() before
application exit. However, this causes the following message to be
output:

JS API usage error: the address passed to JS_AddNamedRoot currently holds an
invalid jsval.  This is usually caused by a missing call to JS_RemoveRoot.
The root's name is "API_Prototype".
Assertion failure: root_points_to_gcArenaPool, at jsgc.c:972
Abort (core dumped) 

So it seems that the finalizer method of a class may not use
JS_RemoveRoot()... I suspected that perhaps the two parameters passed
to the finalizer aren't exactly the expected ones, causing the call to
fail, but then realized that the finalizer function never gets called
at all.

Would the finalizer not be invoked because of the fact that the class
has JSPROP_SHARED properties? Or is it simply never normally triggered
by JS_DestroyContext() the objects are on, but only called by the GC
as required at runtime?

In any case is there another recommended method to "register" a
finalizer function to be called at context shutdown? If this can't be
done it isn't the end of the world however since I can simply have
internal implementation of required finalization code calling
JS_RemoveRoot() for required objects before shutdown, but the above
tests made me wonder...

Thanks,
Matt

.