Oh huh. I think I can use that to get rid of MTIME entirely: when we're looking for logfile entries, just do something like:

atf_check -r 5 -o match:"${msg}" tail -n 1 "${SYSLOGD_LOGFILE}"

I'll give that a try.

This almost works, but there's a small issue: if the command times out, as it will in some syslogd tests which are marked expected-to-fail, the test result is "broken" instead of "expected failure".

I think I'll just modify those test cases to use syslogd_check_log_nopoll() instead, as the -r trick makes this patch much simpler.

Yeah, I tend to agree. I think if the global test timeout elapses, then the test should be marked broken regardless of whether it's expected to fail, but individual atf_check timeouts should be handled the same as if atf_check itself had failed.

More info on the "bug": If you use any timeout less than 5, it works as expected:

atf_check -r 4 -o match:"${msg}" tail -n 1 "${SYSLOGD_LOGFILE}"

Ahh, good find, we have this set_common_atf_metadata() function which sets the timeout. syslogd/tests/Makefile already sets a timeout of 20s for each test, which I think is fine... what do you think of simply removing this 5s timeout?

Ah, yes, I agree. Also, you should be able to remove syslogd_check_log_nopoll(), as syslogd_check_log() will fail as expected after 10 seconds (and not broken).