Array marking

Mon May 1 15:46:00 GMT 2000

Godmar Back wrote:
> I also don't see how you'd avoid marking classes unless you store a
> it in the object that says whether its class is collectable or not
> and therefore must be marked or not.

Well... we can do that easy enough using the bitmap marking, if we assume
that compiled classes are never collectable. But I still think the idea of
(eventually) being able to collect dynamically loaded classes is appealing,
though, so I think classes should be unconditionally marked. The issue at the
moment is that the collector has to scan vtables in order to find the class
pointer, which might be slowing it down? It would be nice if the collector
knew to mark the first word of the vtable, but not scan the rest.
In a JIT compiler, we might need to mark the actual methods themselves, but
we still don't need to scan the vtable beacuse they are also reachable via
Class->methods[i]->ncode.
regards
 [ bryce ]