Zero-config autofixer

まずパッケージカタログとパッケージビルドサーバーの統合を確認し、次に既存の GitHub ワークフロー設定と、ここで説明されている info.rkt のオプトイン機構を調査します。パッケージの検出、pull request の認証、スケジューリング、制限、および base の特別な扱いを対象範囲として定義します。設定不要のワークフローについて合意された設計ができれば完了です。

The Autofixer is quite useful, but requires a lot of fiddly manual setup. Even with the custom action I've created, setting up weekly Autofixer pull requests requires creating a new GitHub workflow, figuring out how to properly install the project in that workflow, getting a private key for the Resyntax CI bot account, and using the jackfirth/create-resyntax-pull-request action. It's a lot of steps and it's easy to get something wrong.

Ideally, the Autofixer would operate on a zero configuration model by using the package catalog. Instead of each user setting up their own CI integration with the autofixer, a central server would periodically run Resyntax on all GitHub-based packages in the package catalog and create pull requests for each of them. Users could opt-in (or out, depending on the signup model) using the package's info.rkt file. Optionally users could specify limits on the number of fixes, frequency and size of pull requests, etc, but the default experience should only require opting in or out. The benefits of this approach include:

• Far simpler setup hopefully leading to much wider adoption, which in turn helps greatly with migrating the Racket ecosystem away from deprecated code.
• More motivation for third-party libraries to provide their own library-specific refactoring rules.
• Much easier debugging on my end, as I won't need to deal with permissions issues in dozens of different repositories.

Downsides:

• I'll have to either run my own infrastructure or integrate with the package build server.
• Users have a lot less control and are more likely to receive unwanted pull requests.
• Code not in published packages can't be refactored.
• Code in the base package can't be refactored, probably. At the very least, special handling of base is required.