This article felt like a bait-and-switch. It sounded interesting until the very end, when finding out it’s immutable-only and, as such, not an online-capable algorithm. That’s a pretty big caveat and a huge difference when basing the article on bloom and cuckoo comparisons. :(

Neither Bloom nor cuckoo filters are online algorithms. It is all fixed capacities.

Any comparison to vacuum filter (vldb19?)

Xor filters are going to be faster, smaller; the implementation far simpler (~200 lines).

CF nerd here, thanks for the paper I'll make sure to read it carefully. In the meantime, two nitpicks: 1. Cuckoo's aren't that hard, I have a <250 LOC implementation that doesn't cut corners. github.com/kristoff-it/zi…

Your implementation is great and quite simple. I don't think I ever wrote that this wasn't possible. I will write however that your implementation of an xor filter would be simpler. At least the query part...

Nice implementation! The space savings are impressive. Have you considered comparing against cache-sectorized and register-blocked Bloom filters? vldb.org/pvldb/vol12/p5… github.com/peterboncz/blo…

This repo contains reproducibility instructions: github.com/peterboncz/blo…

When I experimented with random acyclic graphs, murmur was quite bad on some inputs, look at how many tests are commented out: github.com/alnsn/rgph/blo…

Murmur is quite good. Are you sure that your implementation is correct?