Library creates conflicts for very short strings

[Trinidus]

I just ran into an issue with some example data where I got like 64 conflicting entries in a sample of about 810.000. This seems to be way too much for what I expected so my questions:

• Is there a minimum string size for this hash function?
• Is there a general problem with the algorithm for very short strings?

Here one of my examples with conflicting identical hashes for different strings:

const hash1 = xxHash32('10-10020567-549308', 0).toString(16) // --> b25eede2
const hash2 = xxHash32('20-20384387-548563', 0).toString(16) // --> b25eede2

[Jason3S]

@Trinidus,

That is to be expected given that the hash is only 32 bits.

Here are a few discussions:

• https://github.com/Cyan4973/xxHash/issues/165
• https://xxhash.com/
• https://stackoverflow.com/questions/45788721/where-can-i-find-xxhash64-and-md5-collision-probability-statistics

Given that xxHash32 is only 32bit, the probability becomes something like:

p(k) = 1 - exp(-k(k-1) / (2 * 2^32))

From the graph, anything over 100k values will have more than a 50% chance of a collision.