libc-coord - Re: Relation between C timespec_get time bases and POSIX clockid

Follow @Openwall on Twitter for new release announcements and other news

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <87sfco8b0r.fsf@oldenburg.str.redhat.com>
Date: Tue, 25 Apr 2023 13:19:16 +0200
From: Florian Weimer <fweimer@...hat.com>
To: Jonathan Wakely <jwakely@...hat.com>
Cc: libc-coord@...ts.openwall.com
Subject: Re: Relation between C timespec_get time bases and
 POSIX clockid_t values.

* Jonathan Wakely:

> C11 added timespec_get and the constant TIME_UTC. POSIX extends this
> by requiring TIME_UTC to use the same clock as
> CLOCK_REALTIME. However, TIME_UTC is required to be non-zero, and
> CLOCK_REALTIME is zero on existing implementations.

Uh-oh.

| If the timespec_get function is successful it returns the nonzero
| value base; otherwise, it returns zero.

Here, “base” is TIME_UTC, so that non-zero assumption is baked pretty
strongly into the API.  This is rather unfortunate.

> C23 adds TIME_MONOTONIC, and (optionally supported) TIME_ACTIVE and
> TIME_THREAD_ACTIVE.  Presumably a future version of POSIX will say
> those use the same clocks as CLOCK_MONOTONIC, CLOCK_PROCESS_CPUTIME_ID
> and CLOCK_THREAD_CPUTIME_ID respectively.
>
> It would be helpful if POSIX guaranteed that every C time base can be
> converted to a POSIX clockid_t by adding the constant value (TIME_UTC
> - (int)CLOCK_REALTIME). This would allow programs using both
> timespec_get and clock_gettime to use a single value for a given
> clock, and easily convert to the other domain. For example, a program
> using timespec_get in one module (which might be written in portable
> ISO C) might want to pass a clock to pthread_cond_timedwait in another
> module (written in POSIX C). This could also be relevant to a proposed
> C++ extension that would expose the underlying clock used by C++
> clocks such as std::chrono::system_clock and std::chrono::steady_clock
> (see https://lists.isocpp.org/std-proposals/2023/04/6379.php for
> details).

I think the C++ extension should aim at supporting POSIX (plus vendor
extensions) exclusively, and ignore the ISO C clocks.

C libraries don't know what kind of incompatibilities WG14 will come up
next, so we cannot give you any promises for the future behavior of
timespec_get.  You probably should ignore it.

Thanks,
Florian

Confused about mailing lists and their use? Read about mailing lists on Wikipedia and check out these guidelines on proper formatting of your messages.