Glibc change exposing bugs

My reaction here is the same as HJ's: if a function as speed-critical as memcpy() is made faster by a change, and it breaks overlapping memcpy()s, the fault is the overlapper for being bloody stupid. (However, if Linus is right that this isn't actually speeding anything up except perhaps in microbenchmarks, then the large size of the 'fast' memcpy() implementation becomes an issue.)