Ask HN: JSON Output for CLI Tools?

3 points by zeisss 5 years ago | 4 comments

I am currently our public command-line tool some love and although we officially have an `--output=json` flag, it rarely works or does anything useful.

* While thinking about how to fix/implement this for more commands I was wondering if this kind of more structured output is actually of interest/use for developers/devops/SRE?

Some context: The tool in question is a client for our API and used for validation + updating resources and triggering long running loadtest jobs. Users are developers, although we also recommend it for integration into CI/CD pipelines etc.

Any feedback is welcome :)

mikece 5 years ago |

If the idea is to have output from CLI apps/scripts that can be easily parsed so that chaining scripts together without regexing yourself to death is the goal this sounds like what PowerShell Core (which runs on macOS and Linux as well) does: PS commands return objects -- with a default .ToString() output if you don't grab the result and use it in the next step/script -- allowing easy object.property accessors as well as foreach iteration if the command returns a list/array of objects.

zeisss 5 years ago | |

I understand the idea for fetching / listing resources, e.g. list me all ec2 machines.

My confusion is more about general logs ("preparing output in /tmp/randomid") and progress updates ("the job is 25% done", "deploying cluster", "docker image download: 90%"). Is that actually of interest as structured output or should that remain as human readable on stderr?

maerF0x0 5 years ago |

I find json output massively valuable because I combine it with `jq` instead of having to try and cut (or other command) slice up responses.

viraptor 5 years ago |

If your json-vs-text output drifts often, you could make the backing objects the same. AWS kind of does it in their cli tool. It's all objects and json inside, but you see nicely rendered tables. "--output=json" gives you the real/raw data.