Can anyone else who reads this enlighten me about how this algorithm is
useful?

By my read, it recursively collects an object and any objects it references
into a "region". The only way I can see two regions can exist is if two
different global variables point to two totally disjoint graphs of objects.
I don't understand how this is at all useful.

Most programs I can think of would end up with all objects in one big
region, degenerating this solution into an inefficient ARC+cycle finding
algorithm.

Further, instead of merely having the bad thread-contention of reference
counts on objects, this system requires a count per region of all
references into the region, which is going to be even more heavily
contended.

If anyone can offer an explanation as to where this algorithm would be
useful, I'm really curious what that might be.
_______________________________________________
bitc-dev mailing list
[email protected]
http://www.coyotos.org/mailman/listinfo/bitc-dev

Reply via email to