Subj : Re: Threading question - (50 bucs if you help me)
To   : netscape.public.mozilla.jseng
From : nick.longinow@brio.com (Nick Longinow)
Date : Tue Feb 18 2003 05:51 am

Brendan Eich <brendan@meer.net> wrote in message news:<3E4D8989.3070100@meer.net>...
> Nick Longinow wrote:
> 
> >Just kidding on the 50 Ohio State Univ. players thing.  
> >Did it get you to read this ? Please help!!
> >
> >I have a single service that creates the JS engine, and 
> >then each user request that hits it creates its own
> >runtime and context.  I find that this works fine except
> >for threads that hit at the roughly same time.  
> >
> >Specifically, I AddRoot several items right after the
> >context is created for a new thread, and shortly after
> >do a MaybeGC.  So far so good, as long as the threads
> >hit far enough apart.  
> >
> >But if they occur close enough together, then the 
> >gc_root_marker check will find that at least one of
> >those items I did AddRoot on fail the check for
> >root_points_to_gcArenaPool = JS_TRUE;  
> >
> 
> That means the GC scanned a root you registed by its address and found 
> it contained a pointer (jsval or plain pointer) that did not point 
> within the GC heap.  You either failed to initialize the root, or you 
> let it turn into garbage on the malloc heap (get recycled, and filled in 
> with new bits that did not form a valid GC-thing pointer) without first 
> unregistering the root with JS_RemoveRoot.
> 
> /be
> 
> >
> >The 1.5 jsgc.c code has a comment that says this is due to
> >some root not being removed properly. But, I am still 
> >using those objects, and so wouldn't *want* to do a 
> >remove root on them.  
> >
> >Any ideas what the collision here is all about ?
> >  
> >
Makes sense.  It looks like the new threads are trying to
reinitialize roots that were already added, using static
addresses.  I'll modify them to non-static, and let the thread
destructor properly remove them as roots.  Thanks for the good
description of this rooting mechanism.

.