The following example shows a command used to set an environment file, using a command of false to simulate a command error situation that returns a non-zero exit code.
Is there a simpler way to fail the action for this failure scenario? Or a better technique when combining a call to a command that could fail whose output is needed for a later step?
name: Failure in Command Test
on:
workflow_dispatch:
jobs:
test:
steps:
- name: Failure during env var set
run: |
echo "command_output=$(false)" >> "$GITHUB_ENV"
if [[ -z "$command_output" ]];then echo "::error ::Error calling command" && exit 1; fi
If the echo command fails to write to the file in $GITHUB_ENV, it returns an exit code of 1. That would cause the step to return failure and the parent job to also complete with a failure. The lines of code after that would not run.
In your example, the subprocess -- $(false) -- fails and returns an exit code. However, that exit code is swallowed by the echo; it's just capturing the output. The echo step then writes the empty response to $GITHUB_ENV:
> cat $GITHUB_ENV
command_output=
$GITHUB_ENV only sets the environment values for subsequent steps, so that variable is not available for the if.
To get the error code from the subprocess, you can assign the output to a variable. If that subprocess fails, the exit code will be returned. A failure would halt the script automatically, while success would allow the script to continue to the next line.
The revised code:
command_output=$(false)
echo "command_output=${command_output}" >> "$GITHUB_ENV"
In this case, you'll get an error from the $(false). That will halt the script (and make the step and job fail). If the subprocess had finished successfully, the output would have been captured into the variable command_output, which is then expanded in the echo. It is then made available as an environment variable for future steps on that runner.