Making Python 3 more attractive

Making Python 3 more attractive

Fair enough. I don't use Python :)

So I guess the paper is slightly more relevant. But the comment that only a super-optimized RC framework could match a mark & sweep collector simply doesn't square with what the paper actually says. The paper only shows algorithmic equivalency between a) automated RC with cycle collection and b) tracing collectors. In fact, it tends to show that such systems should provide comparable performance when similarly optimized.

"This in turn allowed us to demonstrate that all high-performance garbage collectors are in fact hybrids of tracing and reference counting techniques. This explains why highly optimized tracing and reference counting collectors have surprisingly similar performance characteristics."

This is the Nth time I've seen that paper posted to try to argue that tracing collectors are faster than RC. It's some kind of horrible meme. A couple of other times it's been posted to try to argue that tracing collectors are faster than manual (and minimal, in that the entire graph of objects is rarely subject to independent lifetime management) reference counting in C and similar languages with non-automated memory management. In all cases I don't even think the original posters bothered reading the paper, because the paper 1) only concludes that performance is algorithmically equivalent, and alludes to benchmarks which show equivalent performance in real-world implementations, rather than showing that tracing collector implementations tend to be faster; and 2) is irrelevant to collection frameworks that don't automate as much as tracing collectors.