intro:
Fnv algo outputs a DWord, other word results while input is arbitrary long.
C-code i found:


confusion:
What is the basis?/How Does it work?
What are FNV limits?/Is there a way to find a collision?
Fasm uses the algo too & fasm limits symbols length to 2xx chars: is it proved no collisions appear ever.
Is FNV limit 2^32?
Can you explain step by step?

Thanks for your attention.

I don't know the algorithm but I think the values 16777619 and 2166136261 are prime so any multiple in the field h (2^32) will not collide. This is similar to the linear congruence method used for PRNG functions, where you choose "good" (usually prime) values. When mixing the key stream into the calculation, XOR is a good choice because, unlike addition or multiplication, values such as 0,1,2^n () cannot skew the accumulation in h towards a collision so easily. It's also important that p is smaller (2^8 ) than h so that the upper 24bits are not affected by the current round and carry over to the next iteration.

Given that h is 2^32, there will certainly be a collision after 2^32 different inputs; there is also a fair chance that if you tried 2^32 different keys you'd find one or more collisions.

Hope that helps.

fasm uses FNV-1a xor-folded down to 24 bits, and it is actually quite easy to get the collision. Quick search gave me collision with "aaigb" and "arsga" texts.

compiles

fasm has a linked list for each used hash value, so you're allowed to have multiple labels with the same hash value, it'll just be slow if there's a whole lot of them. Unless you're making collisions on purpose it won't be a problem.