[TheGiraffe3]

This is the equivalent of specifying -w with the original codespell.
Feedback welcome.

related: #68

[larsoner]

Seems reasonable to me, but a test and README update would be good

[TheGiraffe3]

Added a test. README update was already done

[larsoner]

Okay, thinking more about this and how actions like this (pre-commit, prek, etc.) normally operate, the action typically fails if it makes changes. I think we should probably do the same. So maybe in write mode we should do:

1. A normal dry run, unchanged. Fixed typos then show up as annotations, which is useful because they show what was changed.
2. A second run with --write-changes to apply the fixes, with its stdout sent to /dev/null. Otherwise every misspelling is reported twice and gets duplicate annotations. As a bonus, -w prints the fixed lines indented ( path:1: x ==> y), and the default matcher regex ^(.+):(\d+): would capture those leading spaces as part of the file path.

And longer-term, we update codespell to optionally allow returning non-zero exit code in -w mode, then things can be simplified here if desired.

[TheGiraffe3]

If we're deleting output of the second run, it doesn't print the fixed lines. Should we be sending that to stdout too?