sem_timedwait3(..., ..., clockid_t)

[Eric van Gyzen]

Would anyone else be interested in me adding a sem_timedwait3() function?

int sem_timedwait3(sem_t *sem, const struct timespec *abs_timeout,
clockid_t clock_id);

The clock_id would specify the reference clock for the absolute timeout.
The standard sem_timedwait() API uses CLOCK_REALTIME, which does not
play nicely with large clock adjustments.

I realize it would be non-standard. I also realize that there are
standard APIs that should probably be used instead, such as Pthread
condition variables with pthread_condattr_setclock(3). On the other
hand, the diff is [currently] tiny, so I thought I would offer it.

Cheers,

Eric

[Eric van Gyzen]

Post by Eric van Gyzen
Would anyone else be interested in me adding a sem_timedwait3() function?
int sem_timedwait3(sem_t *sem, const struct timespec *abs_timeout,
clockid_t clock_id);
The clock_id would specify the reference clock for the absolute timeout.
The standard sem_timedwait() API uses CLOCK_REALTIME, which does not
play nicely with large clock adjustments.
I realize it would be non-standard. I also realize that there are
standard APIs that should probably be used instead, such as Pthread
condition variables with pthread_condattr_setclock(3). On the other
hand, the diff is [currently] tiny, so I thought I would offer it.

kib@ suggested that I use the _np suffix, which is an obviously good
idea, so the proposal is now:

int sem_timedwait3_np(sem_t *sem, const struct timespec *abs_timeout,
clockid_t clock_id);

Eric

[Ed Schouten]

Hi Eric,

Post by Eric van Gyzen
int sem_timedwait3_np(sem_t *sem, const struct timespec *abs_timeout,
clockid_t clock_id);

Could we please make the argument order a bit more consistent to
clock_nanosleep() (i.e., putting the clockid_t in front of the
timespec)? Should we also provide support for the TIMER_ABSTIME flag?

[Sebastian Huber]

Post by Ed Schouten
Hi Eric,

Post by Eric van Gyzen
int sem_timedwait3_np(sem_t *sem, const struct timespec *abs_timeout,
clockid_t clock_id);

Could we please make the argument order a bit more consistent to
clock_nanosleep() (i.e., putting the clockid_t in front of the
timespec)? Should we also provide support for the TIMER_ABSTIME flag?

Maybe it makes sense to talk also with the glibc and POSIX people

http://www.opengroup.org/austin/
http://austingroupbugs.net/main_page.php
--
Sebastian Huber, embedded brains GmbH

Address : Dornierstr. 4, D-82178 Puchheim, Germany
Phone : [phone]
Fax : [phone]
E-Mail : ***@embedded-brains.de
PGP : Public key available on request.

Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG.

[Eric van Gyzen]

Post by Sebastian Huber

Post by Ed Schouten
Hi Eric,

Post by Eric van Gyzen
int sem_timedwait3_np(sem_t *sem, const struct timespec *abs_timeout,
clockid_t clock_id);

Could we please make the argument order a bit more consistent to
clock_nanosleep() (i.e., putting the clockid_t in front of the
timespec)? Should we also provide support for the TIMER_ABSTIME flag?

Maybe it makes sense to talk also with the glibc and POSIX people
http://www.opengroup.org/austin/
http://austingroupbugs.net/main_page.php

It certainly makes sense, at some point. I'll get some agreement within the
FreeBSD community, then take it to glibc and then to POSIX.

Thanks for the suggestions.

Eric

[Eric van Gyzen]

Post by Ed Schouten
Hi Eric,

Post by Eric van Gyzen
int sem_timedwait3_np(sem_t *sem, const struct timespec *abs_timeout,
clockid_t clock_id);

Could we please make the argument order a bit more consistent to
clock_nanosleep() (i.e., putting the clockid_t in front of the
timespec)?

Sure. I had made them consistent with sem_timedwait, but your suggestion makes
sense, especially since...

Post by Ed Schouten
Should we also provide support for the TIMER_ABSTIME flag?

That sounds useful. I would also need to add a parameter for returning the
remaining time.

How does this look?

https://reviews.freebsd.org/D9656

I have not yet tested it, so don't spend time on a line-by-line review yet.

Eric