Maybe simple way to improve accuracy of "cron_sleep" function ("seconds_to_wait" variable)

[steffenweber]

I've noticed that cronjobs always start 1.00 to 1.99 seconds too late. Not a big issue, of course, but it has caught my attention.

The "cron_sleep" function (https://github.com/cronie-crond/cronie/blob/v4.3/src/cron.c#L372) computes a variable seconds_to_wait. The left part of the expression (everything to the left of the + 1) computes a value that looks as accurate as possible for full-seconds sleeps to me (it's up to 0.00 and 0.99 too big but afaik that cannot be avoided if you use full-second integer sleep calls instead of something like usleep). However, for some reason 1 second is then added to the value which makes it 1.00 to 1.99 too big).

I've tried to find a rationale for this + 1 addition but couldn't. It was introduced as part of "changes #12-17 from tcmiller (for *BSD alignment)" in 2003: vixie/cron@1b363f1

From my limited understanding, the addition of 1 seems unecessary. A patched cronie that removes the addition of 1 seems to be working fine. But maybe I'm missing something?

[t8m]

I think it is there to avoid calling sleep(0) in some case which perhaps had some unwanted behavior on some old systems.

So we could just make it min(1, seconds_to_wait) instead of adding 1.

[steffenweber]

So something like this:

diff --git a/src/cron.c b/src/cron.c
--- a/src/cron.c
+++ b/src/cron.c
@@ -628,7 +628,8 @@ static void cron_sleep(int target, cron_db * db) {
 	int seconds_to_wait;
 
 	t1 = time(NULL) + GMToff;
-	seconds_to_wait = (int) (target * SECONDS_PER_MINUTE - t1) + 1;
+	seconds_to_wait = (int) (target * SECONDS_PER_MINUTE - t1);
+	if (seconds_to_wait < 1) seconds_to_wait = 1;
 	Debug(DSCH, ("[%ld] Target time=%ld, sec-to-wait=%d\n",
 			(long) getpid(), (long) target * SECONDS_PER_MINUTE,
 			seconds_to_wait));

[t8m]

I've done something similar on master branch so this should be now fixed.

[t8m]

78e6349