trap EXIT INT does not seem to work at all with CTRL-C

[netvl]

This very simple script does not print anything when I run it like fish ./test-trap and then press CTRL-C:

trap "echo After" EXIT INT
sleep 100s

If I run exactly the same script with bash and press CTRL-C, it prints After two times.

It can be reproduced with a manual function whatever --on-event fish_exit --on-signal INT as well.

Fish version: 3.0.2
Linux void 5.4.11-arch1-1 SMP PREEMPT Sun, 12 Jan 2020 12:15:27 +0000 x86_64 GNU/Linux

[ridiculousfish]

I can reproduce, the bash script needs to be run as a script, not directly from the command line or else you engage job control. This should work in fish, I will figure out why it doesn't.

[ridiculousfish]

In non-interactive mode, fish responds to a SIGINT-exited process by SIGINT itself. The function is enqueued but there's no place to execute it, fish is already exiting. I'll have to think about how this could work.

[scooter-dangle]

I don't know if it's relevant, but I had noticed a change in behavior of the the --on-signal INT flag when upgrading to 3.1 (I think from 3.0, but I can at least verify the old, expected behavior on an installation of 2.7.1).

Previously, executing the following and then hitting ctrl-c would cause hola to be printed:

function stuffs --on-signal INT
    echo hola
end

As of 3.1.0, it doesn't.

[ridiculousfish]

Heh, yes, we enqueue the event but refuse to execute it, because we're cancelling.

[ridiculousfish]

Second issue bisects to f8ba0ac.

[ridiculousfish]

It appears that non-interactive INT handlers have never worked in fish. The interactive case is a 3.1.0 regression and is fixed in bc702cc

@zanchey this would be a good addition for 3.1.1, this comment is what is fixed.

[ridiculousfish]

cherry-picked to 3.1.1 as f776c4e

[zanchey]

Note from the original post and #6744 this appears to have been broken in 3.0.0 as well.

[zanchey]

Moving this out of the 3.1.1 milestone as the non-interactive case is still not fixed.

[Korkman]

Note that trapping SIGINT or SIGTERM should prevent the shell from performing the default action (i.e. exit). At least there should be a way to tell fish not to exit (trap function return status?)

[bakhtiyarneyman]

I would really like to see this fixed, so I posted a bounty on this bug.

[suzaku]

When I use kill -s INT, I can see the output. But I see nothing with CTRL-C, so it seems like CTRL-C's not triggering SIGINT in this case.

[floam]

Huh, this actually works as expected if I attach a debugger and step through it.

[floam]

Er, you need to do a pro hand -p true -s false SIGINT in lldb to get the signal passed like normal.

[ridiculousfish]

Fixed in cf2ca56

[ridiculousfish]

@bakhtiyarneyman I appreciate your offering a bounty but it is not necessary, please keep your money or donate it to a worthy cause. Thank you!

[ysooqe]

On 3.5.1, the fix does not seem to work for custom fish-functions (defined in ~/.config/fish/functions/).

Example:
test_sigint.fish

function test_sigint
    function on_sigint --on-signal SIGINT
        echo "received SIGINT"
    end
    sleep 20
end

Calling test_sigint from the command line and doing a CTRL-C will not print out the expected "received SIGINT" line.

It does work though if I define a simple script with the exact same test_sigint definition inside and then calling it immediately. Doing a CTRL-C will then result in an expected "received SIGINT".

Is it expected that this does not work for custom fish-functions? If so, what would be the proper workaround to use SIGINT with those?