El siguiente código simple (C ++ 11) se ejecutará solo en Mac OS y Linux:

#include <thread>
#include <chrono>
#include <iostream>

void threadFunction() {
    for (int cc=0; cc < 10000000; ++cc) {
        if (cc%1000000 == 0) {
            std::this_thread::sleep_for(std::chrono::milliseconds(200));
        }
    }
}


int main(int argc, char **argv) {
    std::thread threads[10];
    for (int tt = 0; tt < 10; ++tt) {
        threads[tt] = std::thread(threadFunction);
    }
    // Wait for the threads to complete
    for (int tt = 0; tt < 10; ++tt) {
        printf("About to join %d\n", tt);
        std::cout.flush();
        threads[tt].join();
        printf("Joined %d\n", tt);
        std::cout.flush();
    }
    return 0;
}

Sin embargo, finalice la ejecución en Valgrind:

valgrind --leak-check=full --show-reachable=no --track-fds=yes --error-exitcode=1 --track-origins=yes ./theexecutable

... y ahora funciona en Linux, pero no en Mac OSX, fallando con:

==47544== Process terminating with default action of signal 11 (SIGSEGV)
==47544==  Access not within mapped region at address 0x700008F31C3E
==47544==    at 0x10052E37F: _pthread_find_thread (in /usr/lib/system/libsystem_pthread.dylib)
==47544==    by 0x100530D52: _pthread_join_cleanup (in /usr/lib/system/libsystem_pthread.dylib)
==47544==    by 0x100530C63: pthread_join (in /usr/lib/system/libsystem_pthread.dylib)
==47544==    by 0x100057E94: std::__1::thread::join() (in /usr/lib/libc++.1.dylib)
==47544==    by 0x100001BB1: main (in ./vgtest)

¿Alguna idea de por qué esto está fallando (y luego solo en OSX)?

Valgrind-3.11.0, OSX 10.11.3