I am just arguing why not be safer than sorry when so much is at stake? Why would you recommend adding unnecessary risk even where you think you are omniscient and there is no risk?
I only need good statistical properties of the hash function, not cryptographic security.
In fact siphash-2-4 is already overkill for my purposes, with siphash-4-8 being
close to cryptographically secure.
Your attack is about as conceivable to me as P being equal to NP,
in which case there is no cryptographically secure hash function anyway.
What I am saying is that the edge trimming eliminates most of the nodes from consideration and then you have a hash table representing a sparse array for the remainder of the nodes which aren't the pruned/trimmed leaf edges.
Yes, totally correct. But in my view the term "bucket" denotes a fixed size unordered container,
and my current algorithms use no such thing.
Well my wink is about using maximum nonce counts, i.e. edge fraction M/N > 1/2.
My algorithms break down in that case, since the expected number of cycles becomes non-constant,
causing the basic algorithm to overlook an increasing fraction of them,
and edge-trimming fails altogether to remove a large fraction of edges.