Subj : Re: automatic variable that holds JSAtom* in JS_DefineFunction()
To   : itaj sherman <myngs@hotmail.com>
From : Brendan Eich <brendan@meer.net>
Date : Sun Mar 09 2003 11:32 am

itaj sherman wrote:

>i tried the patch, and it solves this specific case, but other cases i
>described in the previous post "SpiderMonkey crashes on invalid JSAtom*"
>still occur.
>

That sounds like a different bug, then.

>that is: the program runs 5 threads that run the main JS script in a loop.
>the main script invokes a native function that compiles and executes another
>script.
>there's also a GC thread that calls JS_GC() once every second.
>after a short while it will crash while executing the inner script trying to
>use an invalid JSAtom* read from the bytecode.
>

What's an "inner script"?  Sorry if I missed it in a previous post.  If 
you are compiling while the GC thread tries to run, you still should be 
safe, because the compile disables the GC.  But as a workaround and to 
test things, you might try changing your JS_GC call to js_GC (you'll 
have to include jsgc.h) and pass the GC_KEEP_ATOMS flag.

>even if the inner script only defines a function and does nothing else it
>crashes in js_Interpert() JSOP_DEFFUN.
>

What roots the inner script?  Do you make a script object for it and 
root a pointer to that object, or a jsval referring to it (or otherwise 
protect, e.g., via JS_LockGCThing)?

>i wanted to send you that testing program - i started making the changes in
>the code so it works with NSPR thread API instead of our regular library. if
>you need it i can finish it tomorrow and send it.
>

Please send any testing program, now and when you've finished the NSPR port.

/be

>
>thanks
>itaj
>
>
>
>  
>

.