On Fri, Jun 27, 2008 at 11:24 AM, chromatic <[EMAIL PROTECTED]> wrote:
>> @@ -877,6 +951,8 @@
>> PMC_sync(p)->owner = dest_interp;
>> }
>> else {
>> + /* XXX: This error-handling is bad, we should do
>> + something more standard here instead. */
>> /* fprintf(stderr, "BAD PMC: address=%p,
>> base_type=%d\n",
>> p, p->vtable->base_type); */
>
> I'm not sure what else to do than throw an exception. PARROT_ASSERT might be
> useful.
Current behavior is to call PARROT_ASSERT(0), which will cause a
failure but which won't print out any helpful message. Better
alternatives might be real_exception or PANIC, to at least print out a
diagnostic.
This code in question occurs in src/headers.c:fix_pmc_syncs, which
moves shared PMCs to a new interperter once their current interpreter
is destroyed. Prior to this, the pool should be completely swept for
all non-shared objects, and the only objects still alive should be the
shared objects. The panic occurs when a non-shared object is found to
be still alive when shared objects are moving to the interpreter.
At the very least, I feel like this is ignorable. At the most, we
could print out some kind of warning or even throw an exception.
Either way, this is a recoverable error and we shouldn't close down
the whole program here.
--Andrew Whitworth
