Hi Bruno!
> Le 4 oct. 2018 à 18:51, Bruno Haible <br...@clisp.org> a écrit :
>
> […]
>
> I would conclude that either times() or getrusage()+gettimeofday()
> provides the necessary functionality. Which one to choose? I think
> it's useful to compare the results on various platforms (regarding
> precision and behaviour regarding children processes), to see
> which one is better.
With respect to wall clock:
- nothing in getrusage. You suggest gettimeofday, but it’s
not monotonic, clock_gettime(CLOCK_MONOTONIC,…) is a better
option. It has portability issues, but gnulib to the rescue.
- times returns a clock_t, in system dependent clock ticks.
Resolution is unclear, but I think I understood from various
place that it’s typically 1µs.
https://www.gnu.org/software/libc/manual/html_node/Processor-Time.html
With respect to what you called multiplicity (i.e., time spent
in subprocesses or threads):
- getrusage offers two choices:
- RUSAGE_SELF: the calling process and its threads.
- RUSAGE_CHILDREN: main and subprocesses that have been wait’ed.
- Linux offers RUSAGE_THREAD, which is just about the current
thread. I don’t think we care about this.
(https://linux.die.net/man/2/getrusage)
- times offers the same options:
- tms_utime/tms_stime: user/sys CPU time of the calling process.
- tms_cutime/tms_cstime: user/sys sum of the tms_stime/tms_cstime
of wait’ed child processes.
(http://pubs.opengroup.org/onlinepubs/9699919799/functions/times.html)
With respect to precision:
- getrusage: 1µs on modern platforms, possibly 1ms
(https://stackoverflow.com/a/12480485/1353549)
- times: clock ticks (as for its usr/sys times), so I believe, 1µs
in practice
With respect to portability:
- getrusage is handled by gnulib
- times too, but AFAICT, only for Windows (which is probably
the sign that’s it’s more portable than getrusage.
It seems that both choices are fairly equivalent. That’s what I
can see looking at macOS and GNU/Linux. I have no idea about the
other platforms.
