Re: [hash2] Question about cryptographic guarantees of hash_append_unordered_range

From: Peter Dimov (pdimov_at_[hidden])
Date: 2024-12-09 00:45:20

Darryl Green wrote:
> On Mon, 9 Dec 2024, 7:19 am Peter Dimov via Boost, <boost_at_[hidden]
> <mailto:boost_at_[hidden]> > wrote:
> > Bitwise xor has the problem that xoring two equal values yields 0, so
> > the sequences [1], [1, 1, 1], [1, 2, 2] and so on all hash to the same thing.
> 
> Exactly. Surely this means your other comment that you could add the
> unit64_t[] elements individually, should hopefully fail smhash testing? Add
> without carry on 64 bit chunks is better than add without carry on 1 bit
> chunks but still not ideal?

It's better, yes. It's obviously never worse than adding a single uint64_t.