Greg's version's in nondist/sandbox/sets, right? Where's Aric's?
Ah, a C implementation. It seems premature to me to consider such optimization -- for now, it appears, we're still looking around for the right architecture, and that's much more plastic and faster to experiment with in Python. So, I have not studied set.c in detail,
I have, and I'm -1 on it: it's largely a copy-paste-small-edit of massive portions of dictobject.c. If it has to be implemented via massive code duplication, there are less maintenance-intense ways to do that.
just browsed the readme to get an idea of the interface -- and that seems even more peculiar to me than freeze-on-hashing, although generally similar.
freeze-on-hashing was pioneered <wink> in the Python world by kjbuckets. I've used it in my own Set code for years without particular pain. Greg Wilson seemed to hate it, though.
So, for now, I've stuck to Python,