Re: [PATCH] trace2: intercept all common signals

From: Jeff King <peff@peff.net>
Date: May 10, 2024, 19:46 UTC
Message-ID: <20240510194630.GB1954863@coredump.intra.peff.net>
In-Reply-To: <CAJoAoZmvzZaLN6cQkH4XeD9-=OwWFjT1adRA1oFHaUVyVWwLXQ@mail.gmail.com>

On Fri, May 10, 2024 at 12:34:07PM -0700, Emily Shaffer wrote:

Show 8 quoted lines

> Yes, it is because test_must_fail expects "natural" death. You can
> tell test_must_fail which signal you'd expect to receive, in theory,
> but I didn't get it to work (and it will be tricky to provide the
> correct signal in shell - I had originally hardcoded signal ints in
> sh, but then moved the signal enum->int resolution into
> helper/test-trace2.c because the alternative is doing some nasty
> grepping on other shell utility outputs, since the signal codes aren't
> platform/arch consistent).

We have test_match_signal(). Unfortunately it's not integrated with test_expect_code(), so you have to do:

  { thing_which_fails; OUT=$?; } &&
  test_match_signal 15 "$OUT"

See 5263e22cba (t7006: simplify exit-code checks for sigpipe tests, 2021-11-21) for an example.

Show 5 quoted lines

> I also wonder - do we want to capture SIGKILL as well? Some people may
> have muscle memory for `kill -9` (I do, for better or worse) instead
> of gentler `kill`. My intent was to notice any indication of user
> frustration resulting in manual termination, which would include `kill
> -9` too...

You can't catch SIGKILL; its whole purpose is to be un-catchable.

-Peff