Feedback: chezmoi init --one-shot is destructive and needs clearer warnings

[rcdailey]

I've been migrating from a standard git repo in my home directory to chezmoi for dotfile management. I appreciate chezmoi's security and templating features, but encountered an issue with the init --one-shot command that I'd like to provide feedback on.

While experimenting with the "Once" prompt types in my .chezmoi.toml.tmpl file, I needed to re-run these prompts for testing. I ran chezmoi init -h and noticed the --one-shot option with the description "Run in one-shot mode" - which wasn't descriptive enough to understand its function.

I mistakenly associated "one-shot" with "once" prompts and ran chezmoi init --one-shot while having unsaved changes in my working directory. The command failed with an error message, but more importantly, it wiped out my entire working copy. While I was able to restore the repository with chezmoi init <github-username>, I lost most of my local changes.

I take responsibility for not fully understanding the command before running it. However, I believe commands with destructive potential should have additional safeguards:

• Include clearer warnings in the CLI help that explicitly state this command can wipe out working directory files
• Add confirmation prompts or checks that prevent deletion when there are modifications or untracked files in the working directory

I later discovered the --prompt option, which was actually what I needed, but overlooked at the time.

I hope this feedback helps improve the user experience for others who might encounter similar situations.

[Lockszmith-GH]

Just sharing my experience, I too fail until I succeed (probably a trait of my ADHD).
Reading the docs, is not always my strong suite.

But after I got burnt once doing something stupid with chezmoi, I learned to read the docs, which are excellent.
Here is what --one-shot does:
https://www.chezmoi.io/reference/commands/init/#-one-shot

That being said, maybe some overcautious defaults which can be overridden later might be a good idea.

For the time being, you may be able to configure some rational defaults that work for you - and then you can suggest these to the community?

For example:
You can set interactive to true, and this will prompt you for confirmation of every action.

Or maybe add purge operations to the warnings mechanism, and allow disabling those for those who know what they are doing.

I'm sure one of the maintainers might join this conversation and offer some wisdom (or a solution)

[rcdailey]

Thanks so much for the kind and helpful response. You sound exactly like me, and I have ADHD too 😂