unsigned int luaS_hash (const char *str, size_t l, unsigned int seed,
size_t step) {
unsigned int h = seed ^ cast_uint(l);
for (; l >= step; l -= step)
h ^= ((h<<5) + (h>>2) + cast_byte(str[l - 1]));
return h;
}
From this code it is clear that the subset of characters sampled does not depend on the seed but only from the length of the string and the step.
Therefore if you build strings that only differs in characters that are not sampled, you are going to get same number of collisions. Only the buckets were the collisions happen will change, depending on the seed.
What is the purpose of the seed? it seems that the only effect is to randomize performance from one run to another...
