Jeff King <peff@xxxxxxxx> writes: > Measuring _just_ the collisions is more like the patch below. In my > measurements it's more like 30ms, compared to 10s for all of the > hashcmps. > > So we really aren't dealing with collisions, but rather just verifying > that our hash landed at the right spot. And _any_ data structure is > going to have to do that. The reverse side of the coin may be if we can shrink the hashtable smaller and load it more heavily without sacrificing performance by making the necessary "have we landed at the right spot" check cheap enough, I guess.