This is part 27 of a series on how to approach bash programming in a way that’s safer and more structured than your basic script.

See part 1 if you want to catch the series from the start.

Last time, we discussed returning values from functions. This time, let’s discuss bash’s trap facility.

It’s a Trap!

You know I had to.

Traps are signal handlers, where a signal is an operating-system defined message which can be sent to a running process. In our case, the process is a script.

The operating system will interrupt whatever the script is doing and the script will execute whatever handler it has defined for the particular signal.

If the handler doesn’t cause the script to exit, for example by calling exit, then the script may continue running where it left off. In the case of bash scripts, if the command which was interrupted was an external command, it is likely (but not guaranteed, depending on the command) that the return code from the command will be an error.

This isn’t due to bash, but rather because it will probably have also received the signal and reacted to it independently.

Signals are actually numbers, but most programs also understand them via names.

Some programs always refer to them with a “SIG” prefix, such as SIGTERM. Others understand the signal names without the prefix, such as TERM. Ones which understand the non-prefixed versions typically understand both forms, as does bash.

The typical termination signals which are seen by programs include:

  • SIGTERM - 15 - terminate, the default signal sent by kill - it is a request for program termination

  • SIGINT - 2 - interrupt, the signal sent by Ctrl-C - it tells the program that it is being interrupted and should end itself, but should also communicate to the parent that it was interrupted and that’s the reason why it exited

  • SIGQUIT - 3 - quit - like interrupt, but also produce a core dump

  • SIGHUP - 1 - hangup - sent when the controlling (pseudo-)terminal is disconnected or when the controlling process on a terminal exits

SIGKILL is another well-known termination signal, used by kill -9, but it never makes it to the process. The operating system instead responds directly by killing the process with extreme prejudice, i.e. no opportunity for the process to clean itself up.

There are other signals as well. Run kill -l to see them all.

When a bash script runs a foreground command, any signal sent to bash will not be handled until the command ends.

Bash also defines some pseudo-signals which can be handled with traps, but do not always directly correspond to the defined signals, or differ in their behavior from other shells. They are more like events than signals. They include:

  • EXIT - when the process is exiting, pretty much for any reason

  • ERR - when the process exits because the errexit option was set and an error was encountered

  • DEBUG - used for debugging and variable tracing

I’ll discuss the first two more later, but you’ll have to do your own research on DEBUG and the other signals I haven’t mentioned.

The Trappist

If you want to make use of traps in your scripts, there are a number of typical use cases. Some are described above, but here are other typical ones:

  • cleanup - one of the most important use cases can be closing/deleting in-use resources when the script is going to exit

  • linearizability - preventing interruption when the script is doing something which is critical to complete

  • user communication - the user-defined signals can be sent by the user to notify the script of some event’s occurrence

  • traceback on unplanned termination - my personal favorite for debugging purposes…what was going on when the program stopped?

Etc., etc…

The way you set a trap handler is with the trap command:

trap 'command(s)' SIGNAL [SIGNAL]...

Where command(s) is either a simple command/function call or a set of commands separated by semicolons.

To reset a trap which has been set up by a prior trap command, either call it with no command argument, or “-“ for the command argument. This will reinstate the default handler for the signal, if there is one.

trap - SIGNAL

And to ignore a signal using a trap, feed it a null argument (empty quotes):

trap '' SIGNAL

Some guides mention using “:” (true) for the command argument to ignore a signal, but it is not fully equivalent to a null argument.

If your trap should be available for the duration of your script, you should set it prior to calling main, or near the beginning of your script.

Here’s an example of how to call a cleanup function when your script gets the INT (e.g. Ctrl-C) signal:

cleanup () {
  rm $file
  trap - INT  # reinstate default handler
  kill -INT $$

trap cleanup INT

[...some other work happens, then the user hits Ctrl-C]

Here we’re calling a function which we created expressly to handle the signal. First it removes the temp file we created. This would be where you might do other things, such as close a database connection, etc.

Since the signal can be sent multiple times, this function will be called each time, perhaps interrupting itself. If you are writing a longer cleanup function, be aware it may be entered more than once and may have partially completed beforehand.

For this reason, you may also want to revert or disable the handler if you have commands which should not be invoked again once they are successful, such as long-running ones.

Once the file is removed, we need to kill the script with the INT signal as handled by the default handler, since INT needs special treatment for the operating system’s benefit. See this article on wait and cooperative exit for details.

In order to get the default handler back, we unset the trap we defined. This is not something you’ll usually need to do for other signals.

Finally, the script sends itself INT, which now triggers the default handler.

If this were another trap which we wanted to cause the exit of the script, we would have to call exit explicitly within the handler, otherwise execution of the script would continue (ERR and EXIT are exceptions to the rule).

Stage Left

Rather than handling INT or the other termination signals explicitly, it can be easier to use bash’s EXIT “signal”.

EXIT is triggered for pretty much any reason that the script is exiting, so it is easier to define than remembering to list all of the signals which might terminate your script. EXIT also triggers on errexit, which other signals don’t (except ERR). It is the most useful of the termination-related signals.

Rather than trap on INT as above, trapping on EXIT is better:

trap 'rm $file' EXIT

Here we don’t have to worry about disabling the trap or sending INT, nor calling exit. It’s all been handled for us.

The only downside to EXIT is that you can’t stop the exiting process and resume script execution, if that’s what you want. It also does no good to try to disable EXIT with a null argument.

Is Human

ERR is another special “signal” which triggers when errexit is tripped by a command failing. Like EXIT, ERR can’t be disabled with a null argument, nor can it return to normal execution.

If ERR and EXIT are both set, ERR happens first, then EXIT. Since you can’t short-circuit them, if both are defined, then both will run.

Continue with part 28 - tracebacks