Couple of comments/questions

after the patch it still goes through strlen characters and computes the hash (at least up to maxlen)
so computing the hash is unlikely to be the issue since it still happens with the patch,
I am not sure about the characteristics of the hash implementation with unruly input, does it limit the chain lengths, or can they extend up to the maximum load factor?
if the chain lengths can grow long then the cost of strcasecmp will dominate since it is slow, having to apply tolower on both strings before comparing, and IIRC SQL is case insensitive.
why the **** is the problematic code searching the hash each character in the first place? Why not scan input until the next end of word and search for that word once?

—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you are subscribed to this thread.