> I've read about hierarchical garbage collection, where objects start

Very interesting. I actually started to browse a little... :)

> I'd really rather not see us moving to reference counting--one of the
> joys of Ruby is the ease with which extensions are written.

I agree. But I'm not sure it really hurts a lot if one could reference count
and be blazinly fast. If GC could be tuned not to be order-of-magnitude (or
couple) slower, then there's, of course, no need to think alternatives.

My colleague anyway had a good laugh when I told him about this marvellous
language, and the comparison of the perl and ruby version speed :).

He told, btw., that he has experienced same kind of problems with Java GC.
Much of the problems went away when they just had enough memory and *set* it
explicitly in use in some (probably JVM specific) configuration files.

I wonder if we should add same kind of thing, to enable one to say "let's
start with 300 MB". I guess that my program generates short-living objects
in such rate that it wouldn't help me much, but might be some idea anyway.
And maybe those constants in gc.c could be normal attributes of class GC.

	- Aleksi