Thanks for addressing this so fast. Annoyingly, I suspect it will not help the original case that led me to finding the slowdown (I had some code that was translating from 56 latin-1 Romance characters with diacritics to the equivalent ASCII characters, so it was 1-1, but it was always mapping from non-ASCII to ASCII, and therefore won't benefit from a change that only caches code points 0-127).

That said, every *other* time I've used str.translate has been in an ASCII->ASCII scenario, where this *will* help. So again, thank you.