Reduce memory usage of `make_pmh_tables`?

[serge-sans-paille]

Reduce memory usage of `make_pmh_tables`?

[serge-sans-paille]

Yeah, the O(n^2) is a n issue. it's only paid at compile time but still it should be avoided. It's tool late for me to think clearly about that, but maybe we don't need that much space when combined with your hash selection criteria: if a bucket is oversize, we could try another seed or something like that? Preallocate N buckets of k (k being a fixed constant) elements and cope with it?

[cbeck88]

Yea we can do some measurements and see. That might be the simplest way to fix. Im only not sure what the max bucket size can be. The second phase can probably tolerate several medium size buckets just not TOO many. If the restriction is too tight we will time out in the do-while loop.