Avi Bryant wrote:
> Seconded.  It uses quite a bit more memory than a standard Dictionary,
> but the performance appears to be impressively flat while growing to
> very large sizes, even when using the anemic #identityHash. 

But you are aware that the current behavior is the result of a very 
particular set initialization that is easily fixed, yes? Changing the 
initial set capacity from 3 to 5 will *dramatically* improve the 
behavior ;-)

Besides, I'd be much more interested in a hashtable/dictionary 
implementation that preserves ordering (e.g., if an element is added 
before another it will be enumerated before the other) to preserve 
consistency in a replicated computation. I'll definitely need to look 
into this so if anyone has an implementation I'm all ears...

Cheers,
   - Andreas