Mac clock shim not needed after 10.12

[cellularmitosis]

Summary

This PR:

• changes the Mac clock shim to only be used for Macs older than 10.12.
• moves the AvailabilityMacros.h include from os.c to util.c
• removes the unused mach/clock.h and mach/mach.h includes from os.c

Details

In util.c, the Mac clock shim was included for all Macs:

if defined(JANET_APPLE)

However, clock_gettime() was added in 10.12, so the clock shim isn't needed from that point forward:

• https://opensource.apple.com/releases/
• https://opensource.apple.com/source/Libc/Libc-1158.1.2/gen/clock_gettime.c.auto.html

This PR changes the preprocessor logic to only use the clock shim for Macs running OS X older than 10.12:

/* clock_gettime() wasn't available on Mac until 10.12. */
#elif defined(JANET_APPLE) && !defined(MAC_OS_X_VERSION_10_12)

Update: I also noticed AvailabilityMacros.h was included in os.c, but is only used in util.c, so I moved the include.

Update2: Also noticed that mach/clock.h and mach/mach.h were included in os.c, but are only needed for the shim. I'd guess that the clock shim stuff used to live in os.c before being pulled out into util.c.

[cellularmitosis]

It looks like clock_gettime is about 100 times faster than the shim on my M1 running BigSur 11.6.

bench1.c:

#include <time.h>
#include <stdio.h>

int janet_gettime(struct timespec *spec) {
    return clock_gettime(CLOCK_REALTIME, spec);
}

int main(int argc, char** argv) {
    struct timespec spec;
    int i;
    for (i = 0; i < 100000000; i += 1) {
        janet_gettime(&spec);
    }
    printf("%ld\n", spec.tv_sec);
    return 0;
}

bench2.c:

#include <time.h>
#include <stdio.h>

#include <mach/clock.h>
#include <mach/mach.h>
int janet_gettime(struct timespec *spec) {
    clock_serv_t cclock;
    mach_timespec_t mts;
    host_get_clock_service(mach_host_self(), CALENDAR_CLOCK, &cclock);
    clock_get_time(cclock, &mts);
    mach_port_deallocate(mach_task_self(), cclock);
    spec->tv_sec = mts.tv_sec;
    spec->tv_nsec = mts.tv_nsec;
    return 0;
}

int main(int argc, char** argv) {
    struct timespec spec;
    int i;
    for (i = 0; i < 1000000; i += 1) {
        janet_gettime(&spec);
    }
    printf("%ld\n", spec.tv_sec);
    return 0;
}

cell@indium$ gcc -Wall -o bench1 bench1.c && time ./bench1 
1647914319

real	0m1.385s
user	0m1.328s
sys	0m0.002s

cell@indium$ gcc -Wall -o bench2 bench2.c && time ./bench2 
1647914324

real	0m1.536s
user	0m0.279s
sys	0m1.206s

So I can call clock_gettime 100 million times in about the same runtime as calling the shim 1 million times.

(I also tried -O2 which didn't make any difference).

[cellularmitosis]

Looks like the tests failed on windows:

Starting suite 9...
✘ "localname should match peername: msg=\"127.0.0.1 57922 127.0.0.1 8000\", buf=@\"57922\"": false
✘ "localname should match peername: msg=\"8000\", buf=@\"57923\"": false
✘ "localname should match peername: msg=\"57923\", buf=@\"8000\"": false
✘ "localname should match peername: msg=\"8000\", buf=@\"57924\"": false
✘ "localname should match peername: msg=\"57924\", buf=@\"8000\"": false

Hmm, that doesn't feel related to the PR. Is this just a CI burp?

[bakpakin]

Yeah, that is just some unreliable networking for windows - an open bug that I've been unsuccessful at diagnosing.