Piping command output to tee but also save exit code of command [duplicate]

Question

I have a shell script in which I wrap a command (mvn clean install), to redirect the output to a logfile.

#!/bin/bash
...
mvn clean install $@ | tee $logfile
echo $? # Does not show the return code of mvn clean install

Now if mvn clean install fails with an error, I want my wrapper shell script also fail with that error. But since I'm piping all the output to tee, I cannot access the return code of mvn clean install, so when I access $? afterwards, it's always 0 (since tee successes).

I tried letting the command write the error output to a separate file and checking that afterwards, but the error output of mvn is always empty (seems like it only writes to stdout).

How can I preserve the return code of mvn clean install but still piping the output to a logfile?

Cailly

Answer 1

Since you're running bash, you can use its $PIPESTATUS variable instead of $?:

mvn clean install $@ | tee $logfile
echo ${PIPESTATUS[0]}

Almsgiver

As mentioned below if you have more than a single pipe then you'll need to check the status of each command to know where it failed. – Roanna

Also very important: the variable is ephemeral, so even "echo"-ing it out will lose you the value from running the pipeline. Assign it to another variable unless you will access it immediately, and only once. – Anta

Answer 2

You can set the pipefail shell option option on to get the behavior you want.

From the Bash Reference Manual:

The exit status of a pipeline is the exit status of the last command in the pipeline, unless the pipefail option is enabled (see The Set Builtin). If pipefail is enabled, the pipeline's return status is the value of the last (rightmost) command to exit with a non-zero status, or zero if all commands exit successfully.

Example:

$ false | tee /dev/null ; echo $?
0
$ set -o pipefail
$ false | tee /dev/null ; echo $?
1

To restore the original pipe setting:

$ set +o pipefail

Santalaceous

This seems a more elegant solution than the accepted solution – Metapsychology

Agreed. The accepted answer requires you to know in advance which command in your pipes will fail. If you are piping 5 different commands together, you will have to guess which one in the array will fail. – Maiamaiah

Or do a loop over PIPESTATUS – Roanna

to restore the original pipe setting: $ set +o pipefail – Uncertainty

Note, pipefail is also supported by dash on ubuntu – Burnout

Great answer, however I would point out, that set +o pipfail will turn OFF the pipefail (not return it to its original state) as the docs say Using ‘+’ rather than ‘-’ causes these options to be turned **off**. The options can also be used upon invocation of the shell. The current set of options may be found in $-. – Tropic

Also if you're going for a one-liner: bash -c "set -o pipefail && false | tee log.txt" will exit with an error code. Very useful for something like a build process that you're logging. – Theirs

if tee is killed, it doesn't return... – Goldplate

imho this should be the correct answer: (set -o pipefail && false | tee log.txt) execute the stuff in a sub-shell so that the original state of the pipefail flag will be restored – Observer

what is the scope of pipefail if not run in a subshell as @JiriKremser suggests? I.e. if the script fails before set +o pipefail is called, will pipefail still be enabled when other scripts are called? – Dyedinthewool

Answer 3

You could run the mvn command and cache the exit code... I use the "false" command for my example.

$ { false ; echo $? > /tmp/false.status ; } | tee $logfile
$ cat /tmp/false.status
1

That way you can use the status file content to make further decisions.

I'm curious now whether there is a more eloquent way to accomplish this.

Franconian

I hope the final answer isn't shell specific (ie: bash only). – Franconian

Perfect! I prefer this one to not rely only on bash :-) Thanks @Franconian – Stigmasterol

But I used ( ) instead of { } to capture output .... – Stigmasterol

Answer 4

Workaround (note: a perfer @Frederic's solution):

f=`mktemp`
(mvn clean install $@; echo $?>$f) | tee $logfile
e=`cat $f` #error in variable e
rm $f

Deloisedelong

Could we save a fork with read e < $f instead of the cat? – Choosey

sure. and you probably want to create a tempfile to avoid race conditions. but, as you can see there are better solutions... – Deloisedelong