Damien Zammit, le ven. 21 août 2026 08:44:24 +0000, a ecrit: > In glibc pt-timedblock.c: > > if ((MSG_OPTIONS & MACH_RCV_INTERRUPT) != 0 && err == > MACH_RCV_INTERRUPTED) > return EINTR; > > But MSG_OPTIONS is defined as 0, so this makes __pthread_timedblock > uninterruptible?
Yes, and that is expected, see the standard for pthread_mutex_timedlock / cond_timedwait: These functions shall not return an error code of [EINTR]. pthread locking functions are not interruptible. Only sem_wait is interruptible. > I am getting mach_msg_trap after mach_msg is sent and then a hang, if > you look at thread 11 in the previous backtrace. But the question is: what was supposed to interrupt your timed block? Is the condition of wake-up valid or not? Is the timeout value meaningful? > I can't figure out if pt-timedblock-intr.c is selected or not at compile > time in glibc. > Both versions seem to be in the htl/Makefile. Both are compiled, yes. pt-timedblock-intr.c simply define the interruptible __pthread_timedblock_intr version, which was probably used before it got reimplemented into the current version. Samuel
