That's true - see my other comment below the original one. I'd remembered that one from faint memory as one of the canonical examples of optimizations gone wron

That's true - see my other comment below the original one. I'd remembered that one from faint memory as one of the canonical examples of optimizations gone wrong. And think it might have been a case where that access comes AFTER a test. I don't really remember, though - and I've never had serious problems with optimizations myself.