time(1): add an option to use the monotonic clock

[asomers]

Summary

time(1): add an option to use the monotonic clock

POSIX-compliance requires use of the real-time clock, but we have a non-default
option to use the monotonic clock.

Test Plan

Tested "time sleep 5" with and without -m, while stepping the system
clock forward and backward by one hour.

[asomers]

usr.bin/time/time.c

303 ↗

(On Diff #38403)

This "real" is a little confusing when used with -m. It probably should've said "elapsed" from the very beginning. But I can't change it now without breaking tons of awk scripts. I thought it would be stupid to conditionally change it to "monotonic", because that word doesn't describe what you're measuring (elapsed time), but how you're measuring it. So I left it as is. We should all just think "elapsed" when we see "real" here.

[jilles]

Hmm, it is remarkable to see this after the monotonic clock has existed for so long. Also, GNU time and mksh's keyword appear to use CLOCK_REALTIME without configurability.

I'm not sure preserving the letter of POSIX is important enough to add this option. time is not meant as a high-accuracy benchmarking tool with portable results. Also, relative time services with CLOCK_REALTIME like nanosleep() are not supposed to be affected by clock jumps (but they are affected by speedup/slowdown such as from ntp_adjtime(2)). So changing the clock used by time to CLOCK_MONOTONIC without configurability may be an option as well.

Note that various shells (but not our sh) have a time keyword, which allows timing builtins, pipelines and functions. This keyword tends not to support any options and uses a different implementation. Depending on the shell, some form of quoting or env may or may not be required to call the external time with options.

[asomers]

Remove the -m option and use the monontonic clock unconditionaly

[ken]

This looks good. I think jilles is right, implementing it with CLOCK_MONOTONIC by default may be better than making it optional. Most people won't look at the man page, and this gives it more of a "do what I mean" behavior.

You might want to run it by bde and see what he thinks about it.