(In reply to Richard Palethorpe from comment #8) > Well, I think I now know the change which stopped the test from passing, > from the same file with fresetlockfiles(): Hmm if that indeed changed anything something else is going on. As noted fresetlockfiles only worked for locks that weren't explicitly created by the user, so it's new position (now conditional under multiple_threads) shouldn't make a difference for the test (because it, well, does create the lock explicitly :) ). But yeah, asking POSIX for clarification would be a good idea IMHO.