[WIP] Don't run hooks on submodule roots

[chriskuehl]

git tracks submodules like files. In particular, the roots of submodules appear in the output of git ls-files. This means we get weird errors like:

ckuehl@anthrax:~/tmp$ git init derp && cd derp
ckuehl@anthrax:~/tmp/derp$ git submodule add git@github.com:chriskuehl/dotfiles herp
ckuehl@anthrax:~/tmp/derp$ cat > .pre-commit-config.yaml
-   repo: https://github.com/pre-commit/pre-commit-hooks.git
    sha: 5dd2605fbe26937c721a8ecc0e53949d3eea4f12
    hooks:
    -   id: detect-private-key
ckuehl@anthrax:~/tmp/derp$ pre-commit run --all-files
Detect Private Key.............................................................Failed
Traceback (most recent call last):
  File "/home/c/ck/ckuehl/.pre-commit/repokqqh2jfj/py_env-default/bin/detect-private-key", line 9, in <module>
    load_entry_point('pre-commit-hooks==0.4.2', 'console_scripts', 'detect-private-key')()
  File "/home/c/ck/ckuehl/.pre-commit/repokqqh2jfj/py_env-default/local/lib/python2.7/site-packages/pre_commit_hooks/detect_private_key.py", line 15, in detect_private_key
    content = open(filename, 'rb').read()
IOError: [Errno 21] Is a directory: 'herp'

The only way to avoid this right now is to explicitly exclude submodule paths in you .pre-commit-config.yaml (or to modify every hook to check if an input file is actually a file, which seems bad).

We can exclude these with a slightly more complicated ls-files command.

I didn't test this extensively or even check if the tests still pass, mostly just looking to see if this looks like a reasonable solution. If it does I'll go ahead and write real tests (and test it on some older git versions to make sure it still works).

[chriskuehl]

Could also exclude symlinks here if desired. I didn't because I can imagine some useful hook (e.g. checking for symlinks pointing outside of the repo), but the current behavior could also potentially be dangerous (e.g. we could end up running a hook on a symlink pointing outside of the git repo, transparently to the hook, which then "fixes" a file outside of the git repo).

[chriskuehl]

There are also maybe useful hooks to run against submodules, but imo those should just read .gitmodules instead. Random hooks shouldn't have to worry that they'll be passed in a directory.

[asottile]

This seems similar to the approach that I wanted to take. One thing that I wanted though is a way for hooks to opt into checking submodules / symlinks in hooks.yaml

[chriskuehl]

What would you think about adding a per-hook option (defined in hooks.yaml or overriden in .pre-commit-config.yaml) like:

types: ['files', 'symlinks', 'submodules']

where the default (if omitted) is just ['files']? Could also separate executable files but not sure that's really helpful.

[asottile]

Yeah that sounds good, also makes it easy to implement #220

[chriskuehl]

Abandoning this PR, I'll make another that takes care of #191, #220, and #257.