That libc change was intended to be a pure refactoring and should not have changed the behavior at all. That it caused the extra stat is a bug (it's a simple inverted test typo).
This is in the grey area of whether the bug was "user-visible", but a performance regression that bubbles up to perceptibility seems like it qualifies. So please file a bug at https://sourceware.org/bugzilla (you can assign it directly to [email protected]). Unfortunately, I can't think of a way to write a real regression test for this in the libc test suite. We don't really do "external" things like use strace to determine if a test passes. Thanks, Roland
