Inspect only git tracked files when run under git repos

[ssbarnea]

Instead of linting all the folders under current directory or asking the user to maintain a long list of excludes, it would be far better to detect when the tool is run inside a git repository and perform linting only on tracked files: git ls-files

This would almost remove the need to maintain exclude lists and provide a very good user experience when running the tool for the first time on any repository.

I already used this approach and testing if inside a git repo is also another very simple command.

[peternewman]

This seems to be similar to, but perhaps not quite the same as #1887 . Would you agree?

Can't this be best handled by some external script suggestion, or a pre-commit hook, which just narrows down the list of files to run it against, without tying it entirely to git etc, there are still a few other VCS systems out there.

[ssbarnea]

The mentioned one is mostly the same, sorry I did not spot it.

There are other VSC out there but using a wrapper to achieve that does not fix the general UX for the tool itself. Think about calling codespell from other places than pre-commit, like from IDEs (vscode, eclipse). Does this means that every user that want to use codespell would need to spend time managing excludes in different places.

I do prefer to optimize for mos users while not breaking for others. Shortly if not a git repo or git is missing, fallback to old behavior. If someone wants to add support for other VSC, they are welcomed to propose a PR to improve it.

Considering that I already implemented a git-ls-files in two other python libraries, maybe we should write a very small library that does a smart-glob (uses git ls-files when appropriate but fallbacks to glob). At least it will be easier to reuse in other python projects. One example at https://github.com/ansible-community/ansible-lint/blob/878a90edbce2a985837d26883bff636ec9c7c3a7/src/ansiblelint/file_utils.py#L203-L233

[KapJI]

I agree that this should be codespell default behaviour (or enabled with single setting) and users don't need to think of how to configure these excluded paths and cover all possible cases by basically duplicating .gitignore.

I think the ideal UX is when you install it, run codespell and it just works.

[ilyagr]

Instead of linting all the folders under current directory or asking the user to maintain a long list of excludes, it would be far better to detect when the tool is run inside a git repository and perform linting only on tracked files: git ls-files

Thanks for mentioning git ls-files! This made me realize you can just do codespell $(git ls-files).

(In order to not duplicate the discussion, let's put replies to this after #1887 (comment) )

[mdeweerd]

I opened PR #3676 that adds the option --git-only.
You can install this (for test for instance) using

pip install --upgrade git+https://github.com/mdeweerd/codespell.git@feat/git_filter 

That replaces your current installation, unless you do that in a virtual environement (python virtualenv).

After that you can do the following to apply codespell only on files that are added to the local git repository.

 codespell --git-only

The update should be functionally ok (at least the tests are).

Regarding the test case(s) for this new function: I am open to receive PRs from other volunteers (or add access to my fork) to add those. I think tests can reuse the tests to check the file filter as they are and apply them to the command with the extra option --git-only, and add an extra test to validate that files that are not under git control are excluded. I did it.