[tk0miya]

Summary

• Evaluate the RuboCop plugin proposed in Ruby: Add RuboCop support for Ruby in ERB templates marcoroth/herb#2260 (gems/rubocop-herb at d0323df) and compare it with tk0miya/rubocop-herb 2cf8414 (latest main), both in implementation and in actual runs (research/herb-pr2260.md)
• Add the reproduction setup and steps (research/repro/herb-pr2260/, new "herb PR #2260" section in research/repro/README.md)

Key findings

• Both convert the whole template into one position-preserving Ruby source for RuboCop's ruby_extractors API, but the PR blanks out <%= and HTML via Herb.extract_ruby and has no autocorrect guard
• On 333 real views (rails/rails and kaigionrails/conference-app), rubocop -a changed the HTML of 63 files (59 with new Herb parse errors) with the PR, vs 0 with tk0miya/rubocop-herb. The main causes are Layout/ExtraSpacing (enabled by default) and Lint/Void
• The PR reports 2,455 offenses vs 992, mostly false positives from Layout/ExtraSpacing, Lint/Void on output tags, and Lint/EmptyBlock on HTML-only blocks
• tk0miya/rubocop-herb shows a few likely false positives that the PR excludes: Lint/EmptyFile, Naming/FileName, Layout/EmptyLineAfterMagicComment

Reproduction environment

• herb is the prebuilt 0.11.0 gem from rubygems; the ruby platform is removed from Gemfile.lock, and the steps use BUNDLE_FROZEN=true bundle install, so the herb C extension is never built
• The PR's gem is loaded from the Bundler git source (glob: gems/rubocop-herb/*.gemspec), pinned to the PR head
• All numbers in the memo were reproduced by following the documented steps

Notes

• report.md, switching-decision.md and the existing "rubocop-herb" section of research/repro/README.md still target 6601d3e; this memo targets 2cf8414. Updating them is left for a separate change

🤖 Generated with Claude Code