Re: timevar: further work

[Bruno Haible]

Hi Akim,

> Here is my proposal to rely on gnulib for time issues.
> 
> commit 382f32bb5696168a80fbae5c5dc9dd6c1342e68b
> Author: Akim Demaille <address@hidden>
> Date:   Tue Oct 2 06:19:19 2018 +0200
> 
>     timevar: rely on gnulib modules for time portability.
>     
>     * modules/timevar (Depends-on): Add sys_time, sys_times, and times.
>     * m4/timevar.m4: Don't check for clock_t and struct tms,
>     guaranteed by gnulib.
>     * lib/timevar.h: Use extern "C" protection.
>     Include <stdio.h> for FILE.
>     * lib/timevar.c: Include sys/time.h, sys/times.h unconditionally,
>     they are guaranteed by gnulib.
>     Remove uses of clock as (now useless) fallback.

Looks all right to me. You can push it.

> > - Is the 'times' approach preferrable to the 'getrusage' approach in all
> >  cases? I’m thinking of multithreading and of processes that fork() 
> > children.
> 
> On this machine, man times gives:
> 
> > SYNOPSIS
> >      #include <sys/times.h>
> > 
> >      clock_t
> >      times(struct tms *tp);
> > 
> > DESCRIPTION
> >      This interface is obsoleted by getrusage(2) and gettimeofday(2).

This sentence is a personal opinion by whoever wrote the man page.
http://pubs.opengroup.org/onlinepubs/9699919799/functions/times.html
does not say that this interface is obsolete or deprecated.

> So I guess in all cases I should prefer getrusage to times.
> Yet getrusage does not give me the wall clock time, so I
> suppose I should use getrusage + time (not times).

time() gives the wall-clock time with 1-second resolution only.
You should better use gettimeofday() instead of time().

> https://stackoverflow.com/a/12480485/1353549

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.

Bruno