For our purposes, the sem_t code is better, so we've preferred it since commit c7ab816af110ea08809e1aabd8bf1e08967c6d53 (5.7). Olivier wrote in that commit's message, "It makes tst_QMutex::contendedQMutex with no msleep 8 times faster". That's true because QMutex didn't lock the underlying pthread_mutex_t in QMutex::lock(), as we used the inlined atomic code for the uncontended case. It is probably possible to merge the qmutex_mac.cpp and qmutex_unix.cpp code now (both are based on semaphores), but I won't do that for two reasons: 1) At best, the PThread functions are going to be thin-wrappers around the code we already have, like FreeBSD's are around usem (see [1]) 2) Darwin has a private API that resembles futexes so we may want to go that way eventually (see [2]) [1] https://github.com/freebsd/freebsd-src/blob/main/lib/libc/gen/sem_new.c [2] https://codereview.qt-project.org/335849 Change-Id: Idd5e1bb52be047d7b4fffffd175369b13ba47bed Reviewed-by: Mårten Nordheim <marten.nordheim@qt.io> Reviewed-by: Marc Mutz <marc.mutz@qt.io>