"trap ... INT TERM EXIT" really necessary?

Question

Many examples for trap use trap ... INT TERM EXIT for cleanup tasks. But is it really necessary to list all the three sigspecs?

The manual says:

If a SIGNAL_SPEC is EXIT (0) ARG is executed on exit from the shell.

which I believe applies whether the script finished normally or it finished because it received SIGINT or SIGTERM. An experiment also confirms my belief:

$ cat ./trap-exit
#!/bin/bash
trap 'echo TRAP' EXIT
sleep 3
$ ./trap-exit & sleep 1; kill -INT %1
[1] 759
TRAP
[1]+  Interrupt               ./trap-exit
$ ./trap-exit & sleep 1; kill -TERM %1
[1] 773
TRAP
[1]+  Terminated              ./trap-exit

Then why do so many examples list all of INT TERM EXIT? Or did I miss something and is there any case where a sole EXIT would miss?

Solution 1

The POSIX spec doesn't say much about the conditions resulting in executing the EXIT trap, only about what its environment must look like when it is executed.

In Busybox's ash shell, your trap-exit test does not echo 'TRAP' before exiting due to either SIGINT or SIGTERM. I would suspect there are other shells in existance that may not work that way as well.

# /tmp/test.sh & sleep 1; kill -INT %1
# 
[1]+  Interrupt                  /tmp/test.sh
# 
# 
# /tmp/test.sh & sleep 1; kill -TERM %1
# 
[1]+  Terminated                 /tmp/test.sh
# 

Solution 2

Yes, there is a difference.

This script will exit when you press Enter, or send it SIGINT or SIGTERM:

trap '' EXIT
echo ' --- press ENTER to close --- '
read response

This script will exit when you press Enter:

trap '' EXIT INT TERM
echo ' --- press ENTER to close --- '
read response

* Tested in sh, Bash, and Zsh.   (no longer works in sh when you add a command for trap to run)

---

There's also what @Shawn said: Ash and Dash don't trap signals with EXIT.

So, to handle signals robustly, it's best to avoid trapping EXIT altogether, and use something like this:

cleanup() {
    echo "Cleaning stuff up..."
    exit
}

trap cleanup INT TERM
echo ' --- press ENTER to close --- '
read var
cleanup

Solution 3

Refining the last answer, because it has issues:

# Our general exit handler
cleanup() {
    err=$?
    echo "Cleaning stuff up..."
    trap '' EXIT INT TERM
    exit $err 
}
sig_cleanup() {
    trap '' EXIT # some shells will call EXIT after the INT handler
    false # sets $?
    cleanup
}
trap cleanup EXIT
trap sig_cleanup INT QUIT TERM

Points above:

INT and TERM handlers don't quit for me when I test - they handle the error then the shell returns to exiting (and this is not too surprising). So I ensure that the cleanup exits afterwards, and in the case of the signals always uses an error code (and in the other case of a normal exit, preserves the error code).

With bash, it seems that exiting in the INT handler also calls the EXIT handler, hence I untrap the exit handler and call it myself (which will work in any shell regardless of behaviour).

I trap exit because shell scripts can exit before they reach the bottom - syntax errors, set -e and a nonzero return, simply calling exit. You can't rely on a shellscript getting to the bottom.

SIGQUIT is Ctrl-\ if you've never tried it. Gets you a bonus coredump. So I think it's also worth trapping, even if it's a little obscure.

Past experience says if you (like me) always press Ctrl-C several times, you'll sometimes catch it half way through the cleanup part of your shell script, so this works but not always as perfectly as you'd like.

Solution 4

This is how you can make the Bash script report its return code $?, while being able to catch the SIGINT and SIGTERM signals. I find this very useful for scripts running in a CI/CD pipeline:

notify() {
    [[ $1 = 0 ]] || echo ❌ EXIT $1
    # you can notify some external services here,
    # ie. Slack webhook, Github commit/PR etc.
}

trap '(exit 130)' INT
trap '(exit 143)' TERM
trap 'rc=$?; notify $rc; exit $rc' EXIT

View more solutions

Comments

[Admin]

Also keep in mind that with a spec like INT TERM EXIT the cleanup code is executed twice when SIGTERM or SIGINT is received.

[Bjoern Dahlgren]

The solution with cleanup does the right thing - very elegant! It has become an idiom for my bash scripts with mktemp calls.

[maxschlepzig]

dash also doesn't trap on just EXIT when it receives SIGINT/SIGTERM.

[maxschlepzig]

zsh as well - thus, perhaps bash is the only shell where EXIT also does match signals.

[ijw]

This doesn't work if you have shellscript errors in your code that cause it to exit prematurely.

[Chris Wolfe]

Can you clarify the purpose of trap '' EXIT INT TERM in the cleanup function? Is this to prevent accidental user interruption of cleanup that you mentioned in the last paragraph? Isn't the EXIT redundant?

[Zaz]

@ijw: In Bash and Ksh, you can trap ERR to handle that, but it is not portable.

[JoL]

@maxschlepzig zsh doesn't trap on EXIT when it receives INT, but it does when it receives TERM. EDIT: I just noticed how old this was...

[Nicholas Pipitone]

This solution isn't robust when another shell calls it. It doesn't handle wait on cooperative exit; you will want to trap - INT TERM; kill -2 $$ as the last line of cleanup, to tell the parent shell that it exited prematurely. If a parent shell foobar.sh calls your script (foo.sh), and then calls bar.sh, you don't want bar.sh to execute if INT/TERM is sent to your foo.sh. trap cleanup EXIT will handle this propagation automatically, so it is IMO the most robust. It also means you wouldn't have to call cleanup at the end of the script.

[Nicholas Pipitone]

Doing kill -2 $$ will propagate up the chain of non-interactive shells, killing each parent shell, and then stop at the first interactive shell, giving stdin back to the keyboard (Which is as excepted, ctrl+c should return to the interactive shell, not hang). Quite sad ash doesn't work with EXIT as bash makes it real easy (so nvm on EXIT being robust if you're talking compatibility), but the kill is important there.

[Kevin]

With set -o errexit, calling false in sig_cleanup() impedes cleanup() from being called. In this case, trap cleanup EXIT seems sufficient.

[mstorsjo]

@musiphil - err=$? in sig_cleanup only gives 0 in dash, so apparently the original signal code is lost when trapping signals there, and that's why something like calling false is required. But as @Enno says, this would need a set +e in sig_cleanup to avoid exiting before cleanup is called.

[jarno]

@NicholasPipitone Mere trap cleanup EXIT will not do, if you want to interupt the "ping" loop at mywiki.wooledge.org/…

[jarno]

@mstorsjo err=$? seems to work in dash 0.5.10.2-6; by which version did you try?

[jarno]

@musiphil why not use sig_cleanup() { err=$?; trap - EXIT; cleanup; exit $err; }?

[mstorsjo]

@jarno - I tried dash 0.5.8-2.10 (ubuntu 18.04), and I can confirm that I see that it behaves the way you describe in dash 0.5.10.2. I bisected it down to this change, git.kernel.org/pub/scm/utils/dash/dash.git/commit/…, which was included in dash 0.5.9.

[jarno]

@mstorsjo if the changed code is already in 0.5.8, how could the change be in 0.5.9? BTW In my experience mere exit works in dash, too, but it does not work in bash.

[mstorsjo]

@jarno - No, the change is in 0.5.9. With 0.5.8, $? is zero in sig_cleanup(), in 0.5.9 and onwards it has the actual signal return value.

[jarno]

@mstorsjo I tested it in Ubuntu 18.04 and it worked. $? is 130 after Ctrl-C. It has dash 0.5.8-2.10.

[mstorsjo]

@jarno With pastebin.com/d74jkFVU on Ubuntu 18.04 I'm getting $? in sig_cleanup, if I press ctrl+c during the sleep. With a newer dash I'm getting 130 there.

[jarno]

@mstorsjo oh, but that is only, if you have errexit flag set. I do not see why it should affect, though. And in never version it is fixed.

[jarno]

@mstorsjo see also my related answer.

[Noam Manos]

Redirecting INT and TERM to the EXIT trap is an elegant solution, which prevents dual calls to the trap command. Thanks!