please write .mypy_cache/.gitignore by default

[douglasbagnall]

If when mypy writes a cache, it also did something like:

   with open(os.path.join(cache_dir, '.gitignore', 'w')) as f:
         f.write('*')

i.e. echo '*' > .mypy_cache/.gitignore

then nobody would ever be annoyed by the sudden appearance of thousands of untracked files.

Note this differs from issue #3336 which added the ignore files to particular projects. This would work for every project, and not interfere with anybody's existing top-level .gitignores.

I have not found the right place in the code to make the patch.

[refi64]

IMO this shouldn't be the default. People will do weird stuff involving intentionally committing the cache to a repo for saving or reproducibility. It's definitely a bit odd, but this would make it work pretty far against the principle of least surprise.

Realistically, people will probably have other Python tooling-related stuff in their gitignore anyway.

[ethanhs]

On top of the principle of least surprise, I'd rather not do this for each vcs that exists, and doing this just for git seems... weird.

[takeda]

@refi64 @ethanhs looks like it was added anyway, sigh: #8193

This is an issue, if there is a tooling that parses .gitignore but doesn't traverse subdirectories. In those cases the .mypy_cache might still end up being processed, because of this change I forgot to add it to root folder's .gitignore.

It won't kill me, but this was totally unnecessary, and it is a gotcha if one is not aware of it.

[ethanhs]

This is an issue, if there is a tooling that parses .gitignore but doesn't traverse subdirectories.

That sounds like a bug in the tool you are using, not mypy. Furthermore, I do think most people probably are using mypy in a git vcs system where they don't want to check the cache into version control. Covering the 80% case seems like a good idea to me.