Re: pre-v2.34.0-rc0 regressions: t7900-maintenance.sh broken due to 'systemd-analyze' (was: What's cooking in git.git (Oct 2021, #06; Mon, 25))

[Eric Sunshine <sunshine@sunshineco.com>]

On Tue, Oct 26, 2021 at 11:40 AM Ævar Arnfjörð Bjarmason
<avarab@gmail.com> wrote:
> [...]
> So "fixing" that is easy, just have the prereq test that thing in
> particular, and why does one thing have a /lib/ prefix, but not the
> other?
>
> But presumably this points to a bigger problem. I.e. we just did a "git
> maintenance start" a few lines earlier.
>
> If I could start something via systemd then presumably it's either up,
> or our "start" is buggy and we didn't actually start something, or are
> using the wrong (non-portable?) invocation to check the running status
> of the thing we just started?

The bit you're missing is that the test script only mocks up starting
and stopping the systemd unit:

    write_script print-args <<-\EOF &&
    printf "%s\n" "$*" >>args
    EOF

    GIT_TEST_MAINT_SCHEDULER="systemctl:./print-args" \
        git maintenance start --scheduler=systemd-timer &&

So, it's only running the `print-args` script, not actually invoking
`systemctl`.

On the other hand, the "analyze" check really is invoking a real
systemd command. (Why `systemd-analyze` needs to open /dev/tty0
explicitly rather than just emitting to stdout or stderr is a
different question...)

> Also aside from that shouldn't this be:
>
>     test_when_finished "systemd-something stop-it some-service" &&
>     systemd-something start-it some-service &&
>     [...]
>
> Or are runaway services handled somehow by systemd magic (tied to the
> PID of the test run?).

Answered above: we're just mocking starting/stopping the unit, not
actually launching any background services.