Why do we always find the subtle bugs just before a release?
Looks like I won't be releasing 3.0 today, just because Hugh found this incredibly subtle pathname lookup bug. We have a patch, we understand the problem, and it looks ObviouslyCorrect(tm), but I don't think I want to release 3.0 just a couple of hours after applying it.
Of course, the bug is so hard to see that Hugh needs weeks to reproduce it even with his stress test, so we could just ignore it and backport the fix later. But I really hate making releases with known issues even if it's almost certainly a "nobody will ever hit this in practice" issue.