I'm seeing some weird behavior with bash and trapping EXIT inside subshells. I'd expect the four following lines to all output the same thing ("hi trapped"):
a=$(trap 'echo trapped' EXIT ; echo hi); echo $a
a=$(trap 'echo trapped' EXIT && echo hi); echo $a
a=$(trap 'echo trapped' EXIT ; /bin/echo hi); echo $a
a=$(trap 'echo trapped' EXIT && /bin/echo hi); echo $a
The first three do print "hi trapped", but not the last one. It just outputs "hi". The trap is not being called. You can verify this with set -x:
set -x; a=$(trap 'echo trapped' EXIT ; echo hi); set +x; echo $a
set -x; a=$(trap 'echo trapped' EXIT && echo hi); set +x; echo $a
set -x; a=$(trap 'echo trapped' EXIT ; /bin/echo hi); set +x; echo $a
set -x; a=$(trap 'echo trapped' EXIT && /bin/echo hi); set +x; echo $a
Through some trial and error I've found that the EXIT trap is not called under the following conditions:
The entirety of the subshell program is a list of commands chained together with &&.
If you use ;, or even || at any point, the trap will execute.
All commands in the chain must execute.
If any one of the commands (except the last) exits with a non-zero exit status such that the last command never executes, the trap will execute.
The final command must be a program on the system, not a shell builtin and not a function.
Non-final commands can be builtins or functions and the trap will not run as long as the final command is a program
Is this intentional? Is it documented?
For reference, I came across this because rvm overwrites cd with its own function that ends up adding a trap on EXIT which does (among other things) echo -n 'Saving session...'. I was running a shell script that uses this bash idiom:
some_dir=$( cd "$( dirname "${BASH_SOURCE[0]}" )" > /dev/null && pwd )
So some_dir was getting 'Saving session...' appended to it. It was hard to debug, because subshells weren't always running the EXIT trap rvm was adding.
I used strace -e clone,execve -f -p $$& to see what the current shell is doing when running echo version and /bin/echo version. I put a & so that it will continue to read commands.
In the /bin/echo version, I believe bash did an shortcut and execve-ed the () subshell for /bin/echo, so the trap is not there anymore (traps do not survive execve, I guess).
In the bare echo version, it's a shell builtin, so there's no need to execve, so the current () subshell exit as a shell, and trap is called.
Now, another weird thing is, if I do this: bash -c 'a=$(trap "echo trapped" EXIT && /bin/echo hi); echo $a', you will see that it is trapped!
I guess this is because bash does shortcut only in interactive mode. Another example difference between batch mode and interactive mode is for x in $(seq 1 30); sleep 1; done. If you input it in the terminal, and press C-z immediately, and use fg to bring it back, you will see that it will exit immediatly -- the remaining sleeps are skipped. If you put it in a script, and C-z, fg, it will continue to sleep for the remaining loops.
Related
I have a Bash shell script that invokes a number of commands.
I would like to have the shell script automatically exit with a return value of 1 if any of the commands return a non-zero value.
Is this possible without explicitly checking the result of each command?
For example,
dosomething1
if [[ $? -ne 0 ]]; then
exit 1
fi
dosomething2
if [[ $? -ne 0 ]]; then
exit 1
fi
Add this to the beginning of the script:
set -e
This will cause the shell to exit immediately if a simple command exits with a nonzero exit value. A simple command is any command not part of an if, while, or until test, or part of an && or || list.
See the bash manual on the "set" internal command for more details.
It's really annoying to have a script stubbornly continue when something fails in the middle and breaks assumptions for the rest of the script. I personally start almost all portable shell scripts with set -e.
If I'm working with bash specifically, I'll start with
set -Eeuo pipefail
This covers more error handling in a similar fashion. I consider these as sane defaults for new bash programs. Refer to the bash manual for more information on what these options do.
To add to the accepted answer:
Bear in mind that set -e sometimes is not enough, specially if you have pipes.
For example, suppose you have this script
#!/bin/bash
set -e
./configure > configure.log
make
... which works as expected: an error in configure aborts the execution.
Tomorrow you make a seemingly trivial change:
#!/bin/bash
set -e
./configure | tee configure.log
make
... and now it does not work. This is explained here, and a workaround (Bash only) is provided:
#!/bin/bash
set -e
set -o pipefail
./configure | tee configure.log
make
The if statements in your example are unnecessary. Just do it like this:
dosomething1 || exit 1
If you take Ville Laurikari's advice and use set -e then for some commands you may need to use this:
dosomething || true
The || true will make the command pipeline have a true return value even if the command fails so the the -e option will not kill the script.
If you have cleanup you need to do on exit, you can also use 'trap' with the pseudo-signal ERR. This works the same way as trapping INT or any other signal; bash throws ERR if any command exits with a nonzero value:
# Create the trap with
# trap COMMAND SIGNAME [SIGNAME2 SIGNAME3...]
trap "rm -f /tmp/$MYTMPFILE; exit 1" ERR INT TERM
command1
command2
command3
# Partially turn off the trap.
trap - ERR
# Now a control-C will still cause cleanup, but
# a nonzero exit code won't:
ps aux | grep blahblahblah
Or, especially if you're using "set -e", you could trap EXIT; your trap will then be executed when the script exits for any reason, including a normal end, interrupts, an exit caused by the -e option, etc.
The $? variable is rarely needed. The pseudo-idiom command; if [ $? -eq 0 ]; then X; fi should always be written as if command; then X; fi.
The cases where $? is required is when it needs to be checked against multiple values:
command
case $? in
(0) X;;
(1) Y;;
(2) Z;;
esac
or when $? needs to be reused or otherwise manipulated:
if command; then
echo "command successful" >&2
else
ret=$?
echo "command failed with exit code $ret" >&2
exit $ret
fi
Run it with -e or set -e at the top.
Also look at set -u.
On error, the below script will print a RED error message and exit.
Put this at the top of your bash script:
# BASH error handling:
# exit on command failure
set -e
# keep track of the last executed command
trap 'LAST_COMMAND=$CURRENT_COMMAND; CURRENT_COMMAND=$BASH_COMMAND' DEBUG
# on error: print the failed command
trap 'ERROR_CODE=$?; FAILED_COMMAND=$LAST_COMMAND; tput setaf 1; echo "ERROR: command \"$FAILED_COMMAND\" failed with exit code $ERROR_CODE"; put sgr0;' ERR INT TERM
An expression like
dosomething1 && dosomething2 && dosomething3
will stop processing when one of the commands returns with a non-zero value. For example, the following command will never print "done":
cat nosuchfile && echo "done"
echo $?
1
#!/bin/bash -e
should suffice.
I am just throwing in another one for reference since there was an additional question to Mark Edgars input and here is an additional example and touches on the topic overall:
[[ `cmd` ]] && echo success_else_silence
Which is the same as cmd || exit errcode as someone showed.
For example, I want to make sure a partition is unmounted if mounted:
[[ `mount | grep /dev/sda1` ]] && umount /dev/sda1
I have a Bash shell script that invokes a number of commands.
I would like to have the shell script automatically exit with a return value of 1 if any of the commands return a non-zero value.
Is this possible without explicitly checking the result of each command?
For example,
dosomething1
if [[ $? -ne 0 ]]; then
exit 1
fi
dosomething2
if [[ $? -ne 0 ]]; then
exit 1
fi
Add this to the beginning of the script:
set -e
This will cause the shell to exit immediately if a simple command exits with a nonzero exit value. A simple command is any command not part of an if, while, or until test, or part of an && or || list.
See the bash manual on the "set" internal command for more details.
It's really annoying to have a script stubbornly continue when something fails in the middle and breaks assumptions for the rest of the script. I personally start almost all portable shell scripts with set -e.
If I'm working with bash specifically, I'll start with
set -Eeuo pipefail
This covers more error handling in a similar fashion. I consider these as sane defaults for new bash programs. Refer to the bash manual for more information on what these options do.
To add to the accepted answer:
Bear in mind that set -e sometimes is not enough, specially if you have pipes.
For example, suppose you have this script
#!/bin/bash
set -e
./configure > configure.log
make
... which works as expected: an error in configure aborts the execution.
Tomorrow you make a seemingly trivial change:
#!/bin/bash
set -e
./configure | tee configure.log
make
... and now it does not work. This is explained here, and a workaround (Bash only) is provided:
#!/bin/bash
set -e
set -o pipefail
./configure | tee configure.log
make
The if statements in your example are unnecessary. Just do it like this:
dosomething1 || exit 1
If you take Ville Laurikari's advice and use set -e then for some commands you may need to use this:
dosomething || true
The || true will make the command pipeline have a true return value even if the command fails so the the -e option will not kill the script.
If you have cleanup you need to do on exit, you can also use 'trap' with the pseudo-signal ERR. This works the same way as trapping INT or any other signal; bash throws ERR if any command exits with a nonzero value:
# Create the trap with
# trap COMMAND SIGNAME [SIGNAME2 SIGNAME3...]
trap "rm -f /tmp/$MYTMPFILE; exit 1" ERR INT TERM
command1
command2
command3
# Partially turn off the trap.
trap - ERR
# Now a control-C will still cause cleanup, but
# a nonzero exit code won't:
ps aux | grep blahblahblah
Or, especially if you're using "set -e", you could trap EXIT; your trap will then be executed when the script exits for any reason, including a normal end, interrupts, an exit caused by the -e option, etc.
The $? variable is rarely needed. The pseudo-idiom command; if [ $? -eq 0 ]; then X; fi should always be written as if command; then X; fi.
The cases where $? is required is when it needs to be checked against multiple values:
command
case $? in
(0) X;;
(1) Y;;
(2) Z;;
esac
or when $? needs to be reused or otherwise manipulated:
if command; then
echo "command successful" >&2
else
ret=$?
echo "command failed with exit code $ret" >&2
exit $ret
fi
Run it with -e or set -e at the top.
Also look at set -u.
On error, the below script will print a RED error message and exit.
Put this at the top of your bash script:
# BASH error handling:
# exit on command failure
set -e
# keep track of the last executed command
trap 'LAST_COMMAND=$CURRENT_COMMAND; CURRENT_COMMAND=$BASH_COMMAND' DEBUG
# on error: print the failed command
trap 'ERROR_CODE=$?; FAILED_COMMAND=$LAST_COMMAND; tput setaf 1; echo "ERROR: command \"$FAILED_COMMAND\" failed with exit code $ERROR_CODE"; put sgr0;' ERR INT TERM
An expression like
dosomething1 && dosomething2 && dosomething3
will stop processing when one of the commands returns with a non-zero value. For example, the following command will never print "done":
cat nosuchfile && echo "done"
echo $?
1
#!/bin/bash -e
should suffice.
I am just throwing in another one for reference since there was an additional question to Mark Edgars input and here is an additional example and touches on the topic overall:
[[ `cmd` ]] && echo success_else_silence
Which is the same as cmd || exit errcode as someone showed.
For example, I want to make sure a partition is unmounted if mounted:
[[ `mount | grep /dev/sda1` ]] && umount /dev/sda1
I am writing a script which will run an external program (arecord) and do some cleanup if it's interrupted by either a POSIX signal or input on a named pipe. Here's the draft in full
#!/bin/bash
X=`date '+%Y-%m-%d_%H.%M.%S'`
F=/tmp/$X.wav
P=/tmp/$X.$$.fifo
mkfifo $P
trap "echo interrupted && (rm $P || echo 'couldnt delete $P') && echo 'removed fifo' && exit" INT
# this forked process will wait for input on the fifo
(echo 'waiting for fifo' && cat $P >/dev/null && echo 'fifo hit' && kill -s SIGINT $$)&
while true
do
echo waiting...
sleep 1
done
#arecord $F
This works perfectly as it is: the script ends when a signal arrives and a signal is generated if the fifo is written-to.
But instead of the while true loop I want the now-commented-out arecord command, but if I run that program instead of the loop the SIGINT doesn't get caught in the trap and arecord keeps running.
What should I do?
It sounds like you really need this to work more like an init script. So, start arecord in the background and put the pid in a file. Then use the trap to kill the arecord process based on the pidfile.
#!/bin/bash
PIDFILE=/var/run/arecord-runner.pid #Just somewhere to store the pid
LOGFILE=/var/log/arecord-runner.log
#Just one option for how to format your trap call
#Note that this does not use &&, so one failed function will not
# prevent other items in the trap from running
trapFunc() {
echo interrupted
(rm $P || echo 'couldnt delete $P')
echo 'removed fifo'
kill $(cat $PIDFILE)
exit 0
}
X=`date '+%Y-%m-%d_%H.%M.%S'`
F=/tmp/$X.wav
P=/tmp/$X.$$.fifo
mkfifo $P
trap "trapFunc" INT
# this forked process will wait for input on the fifo
(echo 'waiting for fifo' && cat $P >/dev/null && echo 'fifo hit' && kill -s SIGINT $$)&
arecord $F 1>$LOGFILE 2>&1 & #Run in the background, sending logs to file
echo $! > $PIDFILE #Save pid of the last background process to file
while true
do
echo waiting...
sleep 1
done
Also... you may have your trap written with '&&' clauses for a reason, but as an alternative, you can give a function name as I did above, or a sort of anonymous function like this:
trap "{ command1; command2 args; command3; exit 0; }"
Just make sure that each command is followed by a semicolon and there are spaces between the braces and the commands. The risk of using && in the trap is that your script will continue to run past the interrupt if one of the commands before the exit fails to execute (but maybe you want that?).
Is it possible to propagate an exit code to the caller in case of a syntax error in a Bash script with an EXIT trap? For example, if I have:
#! /bin/bash
set -eu
trap "echo dying!!" EXIT
echo yeah
echo $UNBOUND_VARIABLE
echo boo
Then, running it gives an exit code 0 even if the script did not really end successfully:
$ bash test.sh
yeah
test.sh: line 8: UNBOUND_VARIABLE: unbound variable
dying!!
$ echo $?
0
But if I comment out the exit trap, the script returns 1. Alternatively, if I replace the line with the unbound variable with a command that returns nonzero (e.g. /bin/false), that exit value is propagated as I would like it to.
The shell exits with the result of the last executed command. In your trap case, that's echo, which usually returns with success.
To propagate your value, simply exit with it.
#!/bin/bash
set -eu
die() {
echo "Dying!!"
exit "$1"
}
trap 'die $?' EXIT
echo yeah
echo $unbound
echo boo
Also note that set -e is considered harmful -- it makes you think the script will exit if a command fails, which it won't always do.
This behavior is related to different Bash versions. The original script works as expected on Bash 4.2 but not on 3.2. Having the error-prone code in a separate script file and running it in a subshell works around problems in earlier Bash versions:
#!/bin/bash
$BASH sub.sh
RETVAL=$?
if [[ "$RETVAL" != "0" ]]; then
echo "Dying!! Exit code: $RETVAL"
fi
sub.sh:
set -eu
echo yeah
echo $UNBOUND_VARIABLE
echo boo
In my bash script, I'm running an external command that's stored in $cmd variable. (It could be anything, even some simple bash oneliner.)
If ctrl+C is pressed while running the script, I want it to kill the currently running $cmd but it should still continue running the main script. However, I would like to preserve the option to kill the main script with ctrl+C when the main script is running.
#!/bin/bash
cmd='read -p "Ooook?" something; echo $something; sleep 4 '
while true; do
echo "running cmd.."
eval "$cmd" # ctrl-C now should terminate the eval and print "done cmd"
echo "done cmd"
sleep 5 # ctrl-C now should terminate the main script
done
Any idea how to do it some nice bash way?
Changes applied based on answers:
#! /bin/bash
cmd='read -p "Ooook1?" something; read -p "Oook2?" ; echo $something; sleep 4 '
while true; do
echo "running cmd.."
trap "echo Interrupted" INT
eval "($cmd)" # ctrl-C now should terminate the eval and print "done cmd"
trap - INT
echo "done cmd"
sleep 5 # ctrl-C now should terminate the main script
done
Now, pressing ctrl+C while "Ooook1?" read will break the eval only after that read is done. (it will interrupt just before "Oook2") However it will interrupt "sleep 4" instantly.
In both cases it will do the right thing - it will just interrupt the eval subshell, so we're almost there - just that weird read behaviour..
If you can afford having the eval part run in a subshell, "all" you need to do is trap SIGINT.
#! /bin/bash
cmd='read -p "Ooook1?" something; read -p "Oook2?" ; echo $something; sleep 4 '
while true; do
echo "running cmd.."
trap "echo Interrupted" INT
eval "($cmd)" # ctrl-C now should terminate the eval and print "done cmd"
trap - INT
echo "done cmd"
sleep 5 # ctrl-C now should terminate the main script
done
Don't know if that will fit your specific need though.
$ ./t.sh
running cmd..
Ooook1?^CInterrupted
done cmd
^C
$ ./t.sh
running cmd..
Ooook1?qsdqs^CInterrupted
done cmd
^C
$ ./t.sh
running cmd..
Ooook1?qsd
Oook2?^CInterrupted
done cmd
^C
$
GNU bash, version 4.1.9(2)-release (x86_64-pc-linux-gnu)
You can determine whether the sleep command exited abnormally by examining the last exit status echo $?. A non-zero status probably indicates Ctrl-C.
No, read is not an external command, it is internal builtin bash command being executed in the same process as the other instructions. So at Ctrl-C all the process will be killed.
P.S.
Yes. you can execute command in subshell. Something like this
#!/bin/bash
cmd='trap - INT; echo $$; read -p "Ooook?" something; echo $something; sleep 4 '
echo $$
while true; do
echo "$cmd" > tmpfile
echo "running cmd.."
trap "" INT
bash tmpfile
rm tmpfile
trap - INT
echo "done cmd"
sleep 5 # ctrl-C now should terminate the main script
done