> The reason that this works is because many platforms implement sleep(2)
> as:
>
> alarm(seconds);
> sigsuspend(...);
>
>
> This is implementation is done in Solaris, etc. I have implemented my
> own sleep function because when using threads under Solaris, messing with
> signals is a bad proposition. So:
Ummm... I don't know what version of Solaris you're using, but
as of 2.5, sleep(3c) is MT-Safe. So long as you're not playing with
SIGALRM yourself, it works as expected in multi-threaded as well as
single-threaded applications.
> void mysleep(int seconds)
> {
> struct timeval tv;
>
> tv.tv_sec=seconds;
> tv.tv_usec=0;
>
> select(0,NULL,NULL,NULL,&tv);
> }
>
> The code above should be portable to every platform that supports the
> standard select(2) semantics. It allows for subsecond precision too.
> This implementation isn't subject to signal dainbrammage either.
This, on the other hand, has the problem that the sleep can
terminate prematurely if any signal is delivered - select(3c) will
exit with EINTR in that case. Unfortunately, select is not
guaranteed to tell you how much time is remaining, so you must
keep track of when to wake up and keep sleeping until that time
arrives.
Note that neither sleep(3c) nor select(3c) is a system call on Solaris.
As you pointed out, sleep() is implemented in terms of alarm(2) and
SIGALRM, on Solaris and most other platforms as well. On many SysV-ish
platforms, including Solaris, select() is implemented in terms of
poll(), which has only millisecond accuracy.
Overall, the best approach for programs like ping would be to use
select, with a loop to trap interruptions:
void safe_sleep(int seconds)
{
time_t when, now;
struct timeval tv;
when = time(0) + seconds;
for (;;) {
now = time(0);
if (now >= when) return;
tv.tv_sec = when - now;
tv.tv_usec = 0;
if (select(0, 0, 0, 0, &tv) >= 0 || errno != EINTR) return;
}
}
-- Jeffrey T. Hutzelman (N3NHS) <jhutz+@cmu.edu>
Systems Programmer
School of Computer Science - Research Computing Facility
Carnegie Mellon University - Pittsburgh, PA
This archive was generated by hypermail 2b30 : Fri Apr 13 2001 - 13:54:13 PDT