With look(1) you can already search the file in significantly less than one second, but it’s cool to see projects that cut down time even more. Thanks for sharing!
I don’t think this author is benchmarking correctly. It’s not possible to use an on-disk B-tree to look up things in 49 microseconds. A single disk read is like 10 ms. A single read on a fast SSD takes like 200 microseconds.
It would be interesting to compare it against sqlite, inserting the hashes in 256 tables (for the first byte).
If space usage and access time are similar, we wouldn’t need a new binding on our projects and insertion/update would probably be solved easily