codespell was a fun experiment. However, it has many issues, when used in this project:
- The number of false-positives and true-positives are in the same ballpark. There are also many false-negatives, so the overall net-benefit is questionable.
- There is often confusion around spelling errors leading to a failing CI (they do not, which was intended).
- LLMs released this year are capable to detect typos with less false-positives and less false-negatives, so the
codespellintegration is a bit redundant in that sense.
Fix all issues by removing it.
Going forward, anyone is free to continue to use codespell, or any LLM, or any other tool, locally. Also, DrahtBot has the LLM typo linter integrated in the summary comment. I think the options are plenty, and are more than sufficient for now.