tail: add test for tail -F against newsyslog rotation

[allanjude]

Summary

Newsyslog hardlinks the original file to the new name, then renames a
new file into place, so that no log entries can be missed.

The broken tail -F on 13.0 does not fail the current follow_rename test
but does fail this new follow_rotate test.

Reported by: ScaleEngine Inc.
Sponsored by: Klara Inc.

[allanjude]

Fix commit message, broken tail -F is on 13.0 not 12.2

[emaste]

One possible optimization, but overall LGTM

usr.bin/tail/tests/tail_test.sh

408

What do you think about something like this, to avoid unnecessary dead time:

for i in `jot 20`; do
        if [ $(stat -f%z outfile) -ge 10 ]; then
                break
        fi
        sleep 0.1
done

[allanjude]

Don't sleep longer than required

[allanjude]

usr.bin/tail/tests/tail_test.sh

408

Fixed here, and where I copied it from (forward_rename test)

[nsultana_seas.upenn.edu]

usr.bin/tail/tests/tail_test.sh

351

LGTM
Would it be worth replacing the constant 10 with $(stat -f%z expectfile) (and -ge with -eq) to ensure that the check is coupled with what's expected, in case at some point in the future somebody modified the test without changing the constant?

[emaste]

usr.bin/tail/tests/tail_test.sh

351

If we're going to do that we'd probably want expectsize=$(stat -f%z expectfile) rather than execing the expectfile stat each time through the loop.

I think -ge is preferable though, if the outfile is larger then there is no value in waiting the full 2s to report the failure.

[oshogbo]

LGTM.

[allanjude]

Update