The toml formatter sorts non-toml files (such as poetry.lock)!

[thclark]

It's possible for files to appear as valid TOML without being TOML. For example the poetry.lock file in many python projects is an internal file format to poetry but is based on TOML.

It's important not to format files that aren't TOML, but this pre-commit hook applies itself to every single file in your commit. If any file accidentally happens to be valid toml then it will get reworked.

In the case of poetry.lock files this creates an incredibly nasty hell because it's already challenging to resolve merge conflicts in these files. Doing so during a rebase requires regeneration of the file; which of course won't be to format. Then you end up with 500 merge conflicts against the next commit that touches it.

I can imagine there would be an unbelievable number of side effects and bugs that could result from this arbitrarily rearranging contents of potentially any file!

Proposed solution

The toml file formatter should only operate on .toml files.

[macisamuele]

I'm effectively surprised that non toml files are considered in here.

PRs are welcome, but as intermediate solution you could mark the file to be ignored or marching the desired extensions via files or exclude arguments (https://pre-commit.com/)

[danielhollas]

We started to run into the same issue with the uv's uv.lock file, which is in TOML format as well. It took me while to figure out why suddenly pre-commit started to identify it as toml!

It's important not to format files that aren't TOML, but this pre-commit hook applies itself to every single file in your commit. If any file accidentally happens to be valid toml then it will get reworked.

This is not true. The pretty-format-hook does correctly specify that it should only be applied to toml files, see:

language-formatters-pre-commit-hooks/.pre-commit-hooks.yaml

Line 47 in f002661

types: [toml]

So how come that pre-commit is suddenly so clever that it recognizes uv.lock and poetry.lock as toml, even though it doesn't have .toml extension? It uses the identify library, and it turns out that both of these lock files are hardcoded there as being toml. The uv.lock change has been merged only three weeks ago, pre-commit/identify#569, while the Poetry's lockfile is there since 2022.

So indeed, I think the correct solution here is to exclude these files from pretty formatting if you don't desire it. I personally think that since both uv.lock and poetry.lock are both autogenerated files (and formmatted already by uv/poetry) they shouldn't be touched by this pre-commit hook by default. So perhaps they should be added to the default list of excluded files here? But that would obviously be a breaking change at this point.