A seed-independent hash collision in GxHash

For a pangenome deduplication tool I’m working on (pandedup), I’m using 128-bit hashes as “unique” identifiers of substrings.

Sadly, it turns out that gxhash128 has a hash collision between the two strings of length 88 below that only differ in 5 positions, regardless of the value of the seed.

Looking at the strings, this clearly isn’t a random collision. Those could be expected when doing \(2^{64}\) lookups in a dictionary of \(2^{64}\) keys, but my data is nowhere near that large.