nanosleep() replacement Take 2

[Darren Tucker]

Tim Rice wrote:

Combining replies here,

[about my modified nanosleep]
> Something isn't quite right with this version.
> When I try this I get
> scp -l 512 openssh/sshd.o uw711:/tmp
> sshd.o             100%   84KB  84.1KB/s   00:01
> scp -l 256 openssh/sshd.o uw711:/tmp
> sshd.o             100%   84KB  42.0KB/s   00:02
> scp -l 128 openssh/sshd.o uw711:/tmp
> sshd.o             100%   84KB  42.0KB/s   00:02
> scp -l 64 openssh/sshd.o uw711:/tmp
> sshd.o             100%   84KB  42.0KB/s   00:02
> scp -l 8 openssh/sshd.o uw711:/tmp
> sshd.o             100%   84KB  42.0KB/s   00:02
> scp -l 1 openssh/sshd.o uw711:/tmp
> sshd               100%   84KB  42.0KB/s   00:02

Difference in select() behaviour between patforms?  Here's what I get
from the same function on Solaris 8:

$ for bw in 512 256 128 64 8 1; do scp -l $bw testfile localhost:/tmp/;
done
+ scp -l 512 testfile localhost:/tmp/
testfile            100%   84KB  80.6KB/s   00:01
+ scp -l 256 testfile localhost:/tmp/
testfile            100%   84KB  41.0KB/s   00:02
+ scp -l 128 testfile localhost:/tmp/
testfile            100%   84KB  20.7KB/s   00:04
+ scp -l 64 testfile localhost:/tmp/
testfile            100%   84KB   9.2KB/s   00:09
+ scp -l 8 testfile localhost:/tmp/
testfile            100%   84KB   1.1KB/s   01:16
+ scp -l 1 testfile localhost:/tmp/
testfile             57%   48KB   0.1KB/s - stalled

[snip]

About the original function:

> +int nanosleep(const struct timespec *req, struct timespec *rem)
> +{
> +       int rc, saverrno;
> +       extern int errno;
> +       struct timeval tsave, ttmp, time2wait;
> +
> +       TIMESPEC_TO_TIMEVAL(&time2wait, req)
> +
> +       gettimeofday(&tsave, NULL);
> +       rc = select(0, NULL, NULL, NULL, &time2wait);
> +       if (rc == -1) {
> +               saverrno = errno;
> +               gettimeofday (&ttmp, NULL);
> +               errno = saverrno;

The following can sometimes end up with a tv_usec above 1,000,000 (not
sure if that matters). Consider this example:

tsave: tv_sec=100, tv_usec=999999
ttmp: tv_sec=101, tv_usec=1
time2wait: tv_sec=2, tv_usec=999999

> +               ttmp.tv_sec -= tsave.tv_sec;
> +               ttmp.tv_usec -= tsave.tv_usec;
	-> ttmp.tv_usec = -999,998

> +               tsave.tv_sec = (time2wait.tv_sec - ttmp.tv_sec);
> +               tsave.tv_usec = (time2wait.tv_usec - ttmp.tv_usec);
	-> tsave.tv_usec = 999,999 - -999,998
			 = 1,999,997

> +               if(tsave.tv_sec < 0){
> +                       tsave.tv_sec = 0;
> +                       tsave.tv_usec += 1000000L;
> +               }

How about something like:
if (rc == -1) {
	tsave.tv_sec = ttmp.tv_sec - tsave.tv_sec - time2wait.tv_sec;
	tsave.tv_usec = ttmp.tv_usec - tsave.tv_usec - time2wait.tv_usec;
	tsave.tv_sec += tsave.tv_usec / 1000000L;
	tsave.tv_usec %= 1000000L;
} else {
	tsave.tv_sec = 0;
	tsave.tv_usec = 0;
}

> +       }
> +       else {

" } else {" ?

> +               tsave.tv_sec = 0;
> +               tsave.tv_usec = 0;
> +       }
> +       TIMEVAL_TO_TIMESPEC(&tsave, rem)
> +
> +       return(rc);
> +}

-- 
Darren Tucker (dtucker at zip.com.au)
GPG Fingerprint D9A3 86E9 7EEE AF4B B2D4  37C9 C982 80C7 8FF4 FA69
    Good judgement comes with experience. Unfortunately, the experience
usually comes from bad judgement.