changie still produces a changelog file on error in batch command

[Magic-Mayo]

Thanks for submitting a bug report, please fill in the following details.

Description
I recently had a misuse of a tempate function in my changie.yaml which led to an expected error when running changie batch auto. The unexpected part was that it still produced a partial changelog. The specific issue I had was only sending in one arg to the customs function instead of two. This led to changie stopping processing but not before it wrote to the next version's changelog "fragment" (not sure how you refer to it internally 😅 ) that lives under the .changes directory.

I understand how things are happening in code and I know this was user error but it might be good to save the user from their own mistakes sometimes. I could see this happening and being ignored or not seen before running changie merge and a release being cut before it's ever noticed...then there's larger issues to overcome. This is just one issue I ran into but if templates are being written and not cleaned up on error I can see it happening for different types of user errors.

Reproduction Steps
Steps to reproduce the behavior:

1. Set the changeFormat key in changie.yaml to intentionally misuse the customs template function so it errors: {{customs .Changes}}
2. Run changie new and create a new change for one of the projects
3. Run changie batch auto
4. See error

What happened
An error like:

Error: template: {{customs .Changes}}:1:2: executing "{{customs .Changes}}" at <customs>: wrong number of args for customs: want 2 got 1

as well as a versioned changelog file that looked like

## 1.5.0 - 2024-09-09
### Feature

Expected behavior
An error but nothing written to any files or files cleaned up before the process exits.

Additional context

[miniscruff]

Correct, changie writes directly to the file unless you specify --dry-run. It's possible we could clean up the incomplete file on failure but I have sort of taken the stance that this file may help diagnose the problem. Or you can use dry run for testing.

Also if you missed the error and then ran merge it would be bad with or without the cleanup as a version would be missing. One argument you could add is the latest and next commands are now inaccurate. But it's not like the program exited with a zero code.

I think the issue of cleaning up is more complex when you take into account projects and replacements. So far the strategy is simply fail leaving your files as they were, and let you diagnose, maybe reset and try again.

Is it perfect, probably not, but there is a bit of work involved here than it may seem.

[Magic-Mayo]

Makes sense. I figured it was intentional but in the case it wasn't I wanted to report it. I would definitely use the exit code to determine whether or not to proceed if I'm automating this in some kind of way. I wanted to report in the case that someone may be doing this manually and overlooks it. Right now I'm still in the process of POC'ing changie so everything I'm currently doing is manual just to get the hang of things and figure it all out.

I think the issue of cleaning up is more complex when you take into account projects and replacements. So far the strategy is simply fail leaving your files as they were, and let you diagnose, maybe reset and try again.

I think because this happens in batch and not merge it wouldn't be too bad. batch just writes the potentially incomplete file to .changes/. But like you said, if the merge command is run with that incomplete file it can get messy. Cleaning up after erroring on the batch command should be easy...

Again, I realize this would only happen from user error on top of a user also ignoring or otherwise not understanding the error output. Another argument for cleanup on failure is that if you're writing to stdout on --dry-run you'll have the output for debugging anyway.

But I'm fine with this being closed out as it was intentionally coded to potentially write an incomplete file. I could also send up a PR if you'd like.

[miniscruff]

Hmm on second thought I believe you are right in that cleaning up batch issues should be relatively easy. We only delete the files at the end and we can just as easily delete the new file on error. Or only write to the file at the end and instead use an intermediate writer during the process.

[Magic-Mayo]

This is what I did to get the functionality for cleanup

[miniscruff]

This is what I did to get the functionality for cleanup

You can make a pull request for it not I'm not sure that will work entirely. Will have to review it.