I tried with the strings in all_constants(). I was only looking at the variance of bucket size. So this was basically using ptr >> 2. I can recollect the code and put together some statistics with different keysets to make a start.
On Sat, 19 Mar 2011, Martin Stjernholm, Roxen IS @ Pike developers forum wrote:
Interesting. I imagine integers can be tough in general since one cannot say anything about their distribution.
What about power-of-two tables where the hashes based on pointers (shifted down a couple of bits)? That's the most common case in pike.
/.../ I think it would be reasonable to conduct some more experiments...
Yes, that'd definitely be a good start.