On Wed, 11 Jan 2023 15:33:47 GMT, Claes Redestad <redes...@openjdk.org> wrote:
>> I agree with the fact that is less intrusive, but certainly not potentially >> quicker, since the complete recalculation of cardinality requires linear >> time in the value of wordsInUse > > A perhaps slightly less racy way would be to only read and write from `words` > once: > > final long newValue = words[wordIndex] ^ bitMask; > cardinality += (newValue & bitMask) != 0 ? 1 : -1; > words[wordIndex] = newValue; > > .. but the fact remains that `BitSet` is not thread-safe, so I think we > shouldn't complicate things to avoid or eliminate potential races. (Using a > local might help the compiler to avoid the back-to-back array read, but > probably doesn't matter) I think the same of @cl4es. ------------- PR: https://git.openjdk.org/jdk/pull/11837
