unix/mphal: use CLOCK_MONOTONIC for ticks_ms/us

[mzakharocsc]

Using gettimeofday for ticks_ms/us is not guaranteed to be increasing, which is contrary to the documentation. Using CLOCK_MONOTONIC fixes this.

[pfalcon]

While gettimeofday() isn't guaranteed to be increasing, clock_gettime() isn't guaranteed to exist at all.

[mzakharocsc]

While gettimeofday() isn't guaranteed to be increasing, clock_gettime() isn't guaranteed to exist at all.

Is there an alternative? gettimeofday() coupled with NTP on a system can cause random unintended sideffects for code that assumes ticks_ms() is always increasing. clock_gettime() is available on Linux/FreeBSD/Mac OSX since 10.12. What other environments does the codebase need to support?

[pfalcon]

gettimeofday() coupled with NTP on a system can cause random unintended sideffects for code that assumes ticks_ms() is always increasing.

What specific issue are you facing?

clock_gettime() is available on Linux/FreeBSD/Mac OSX since 10.12. What other environments does the codebase need to support?

Unix port of MicroPython is intended to support as wide selection of Unix-like OSes as possible. E.g. that XENIX or AIX versions of 1985 or something.

[mzakharocsc]

What specific issue are you facing?

In our application, we use ticks_ms() to measure time since last heartbeat from a child processes. If NTP changes system time into the future, this triggers heartbeat timeout and forces us to restart as a recovery measure.

[pfalcon]

Thanks,

So:

Is there an alternative?

The alternative is to select gettimeofday() vs clock_gettime() via #ifdef.

[mzakharocsc]

The alternative is to select gettimeofday() vs clock_gettime() via #ifdef.

How should that #ifdef look like?

[pfalcon]

It should allow to use either gettimeofday() of clock_gettime() for implementation of those functions.

[dpgeorge]

It might work to use the following to select the implementation (untested):

#if _POSIX_TIMERS > 0 && defined(_POSIX_MONOTONIC_CLOCK)
// use clock_gettime
#else 
// use gettimeofday
#endif 

[dpgeorge]

@mzakharocsc I'm happy to make the #if changes and merge if you aren't able to.

[mzakharocsc]

Updated with #ifdefs

[dpgeorge]

Thanks for updating the PR. Merged in ced340d