Closed AlexanderSaydakov closed 4 months ago
Perhaps I could explain in the test. The sketch reaches capacity for the first time at 2 K 15/16, but at that point it is still in exact mode, so the serialized size is not the maximum (theta in not needed in the exact mode). So we need to catch the second time, but some updates will be ignored in the estimation mode, so I updated more than enough times keeping track of the maximum. Perhaps I should have figured out the exact number of updates given this particular sequence, but not assuming that might be even better (say, in case we change the load factor or just out of principle not to rely on implementation details too much).
Yeah, the test is fine. It just feels sort of overkill to serialize after every update just to check. Not quite an ideal design for quick tests but with lgK=10 it should be ok in practice.
Pull Request Test Coverage Report for Build 9037130030
Details
💛 - Coveralls