Greg A. Woods wrote:

> It seems to me like a waste of time and bother to even try to check if
> the file was modified -- it should not matter if the daemon gets "poked"
> for no reason (there are many better ways to cause problems with cron
> than this!).

Aggreed; but if you want to do the check, the easy way without any kernel
work is: let crontab use utime to preserve the original timestamps of the
crontab file on the working copy.

> On the other hand, I wonder what would break and what the performance
> impact would be if the *kernel* were to force a process to pause on
> close (not on write) until the second ticks over and then to update the
> timestamps (i.e. iff it knows the file has been written to)....  (0.25 ;-)

Refining the granularity of FS timestamps would sure be a better idea (when
we move to 2038 safe timestamps...)


Martin