On Friday, 22 April 2016 at 18:51:57 UTC, David Nadlinger wrote:
As long as elements are not removed too frequently (what do your numbers say?), the performance impact of doing a full linear search in those cases shouldn't be too bad.
Note that a (properly tuned) probabilistic data structure like a Bloom filter allows you to avoid having to keep a full second copy of the list around, hopefully making it possible to keep the optimisation on by default without regrets. (For tiny modules with only a small number of members, you can still do the linear search to skip the hash lookup.)
— David
