Handling Failures in Pipes
Using set -e
When there is no pipe but just a sequence of commands, we can use set -e:
#!/bin/bash
# This will ensure that the script exists
# if any command fails and return code is non-zero
set -e
echo This will happen
ls some_file
echo This will not happen
Failures in Pipes
The return code in a sequence of piped commands is the last one. So if one of the piped commands fails but the last command succeeds, we will erroneously get back a 0 return code and dependent scripts will continue execution.
For example, if we pipe a false (which returns a non-zero code) into true (which returns a zero code), we will get a 0 return code:
> false | true
> echo $?
0
Bash has an array variable called PIPESTATUS which captures all of the return codes from each program in the pipe chain:
> false | true | false | true
> echo "${PIPESTATUS[0]} ${PIPESTATUS[1]} ${PIPESTATUS[2]} ${PIPESTATUS[3]}"
1 0 1 0
PIPESTATUS only holds the return codes until the next program runs, and trying to determine which return code goes with which program can get very messy very quickly.
This is where set -o (options) and pipefail come in.
set -eo pipefail
echo This will happen first
cat script-99.sh | wc -l
echo This will not be printed
This will return a 1 and will not execute the second echo.
Catching Uninitialized Variables
Uninitialized variables can be difficult to spot in a real-world script. If we try to echo the value of an uninitialized variable, echo simply prints a blank line. It doesn’t raise an error message. The rest of the script will continue to execute.
We can trap this type of error using the set -u (unset) option.
#!/bin/bash
set -eou pipefail
echo "$notset"
echo "Another echo command"
# notset: unbound variable
# echo $?
# 1
If we want to initialize a value:
#!/bin/bash
set -euo pipefail
if [ -z "${New_Var:-}" ]; then
echo "New_Var has no value assigned to it."
fi