Show HN: Rbcop, RuboCop rewritten in Rust with identical output

Drop-in replacement for RuboCop 1.92.2. Same config, same command line, same output, about 14x faster than rubocop --parallel in CI.

gem "rbcop", require: false, then bundle exec rbcop.

A reimplementation of RuboCop 1.92.2 in Rust, built on Prism (ruby-prism crate).

The target is byte-identical output: same offenses, ranges, messages, [Correctable] flags, autocorrections, formatter output and exit status as RuboCop 1.92.2 for the same files and configuration. All 613 cops in RuboCop's config/default.yml are implemented. rubocop-performance, rubocop-rails, rubocop-minitest, rubocop-rspec, rubocop-rspec_rails, rubocop-factory_bot and rubocop-capybara run natively (docs/plugins.md). Other plugins (rubocop-sorbet and so on) and custom cops do not; rbcop hands them to RuboCop when one is installed, and otherwise says which cops it skips (see Plugins and custom cops).

Gemfile: gem "rbcop", require: false, then bundle exec rbcop where you ran bundle exec rubocop. The gem carries a prebuilt binary (macOS arm64/x86_64, Linux x86_64/aarch64 glibc and musl); no Rust toolchain is needed.

GitHub Actions: uses: costajohnt/rbcop@v<version> downloads and caches the release binary.

Homebrew: brew install costajohnt/tap/rbcop. Release archives with each release.

docs/migration.md covers switching a project from RuboCop: Gemfile, binstubs, CI, editors and plugins. docs/releasing.md describes how releases are built.

cargo build --release # needs Rust 1.89 or newer

The binary is target/release/rbcop. The crate is not published to crates.io: it builds against a patched copy of ruby-prism in vendor/ruby-prism.

The command line is RuboCop's. Options, .rubocop.yml lookup and inheritance, RUBOCOP_OPTS, .rubocop argument files, inline # rubocop:disable / enable / todo comments, output and exit status follow RuboCop 1.91.0.

rbcop [options] [paths...] rbcop -a [paths...] # safe autocorrect rbcop -A [paths...] # all autocorrections rbcop -f json -o report.json

All builtin formatters are supported, as are the display options, --fail-level, --stdin (with -a/-A it prints the corrected source after the report), -L, --show-cops, -v and -V.

--auto-gen-config (with its options), --regenerate-todo, --disable-uncorrectable and --report-unused-todo-entries are supported. The server options are accepted and run in process (no server is started), and --lsp runs RuboCop's language server. Not supported: --changed and the MCP mode. Plugin and custom cops rbcop does not implement, -r/--plugin and custom formatter classes run with RuboCop when one is installed (see Plugins and custom cops). ERB in config files is rendered by running ruby, as RuboCop renders it, and remote inherit_from URLs are fetched as RuboCop fetches them. The result cache (--cache, --cache-root, AllCops: UseCache) works as RuboCop's, in its own rbcop_cache directory. docs/known-gaps.md lists these and every other known difference, including a few deliberate ones (for example, -o pointing at a file being inspected is refused). -a/-A do not write files that resolve (through symlinks) outside the working directory and the paths given on the command line; those are reported uncorrected with a warning.

rbcop reads plugin configuration (plugins: and require: gems contribute their config/default.yml), and runs natively the cops it implements: RuboCop's own and those of the plugins in docs/plugins.md (rubocop-performance, rubocop-rails, rubocop-minitest, rubocop-rspec, rubocop-rspec_rails, rubocop-factory_bot, rubocop-capybara). Cops it cannot run (other plugins such as rubocop-sorbet, custom cops from require: ./lib/cops/my_cop.rb) go to RuboCop: the fallback.

The fallback is on by default and needs a RuboCop. rbcop uses RBCOP_RUBOCOP when set (for example RBCOP_RUBOCOP="bin/rubocop"), else bundle exec rubocop when the project's Gemfile.lock has the rubocop gem, else rubocop on PATH. It runs RuboCop once more for exactly the cops rbcop cannot run (--except every cop rbcop implements) on the same files with the same configuration and options, and merges its offenses into its own report, so formatters, --fail-level and the exit status see one run. The saving is smaller, since RuboCop still parses every file, but core cops no longer run in Ruby. A project that needs no fallback never starts RuboCop. If RuboCop fails to run (say bundle exec outside an installed bundle), rbcop exits 2 with its error.

RuboCop loads the project's Ruby (its Gemfile, plugins, cop files), as it would if you ran it, so rbcop says so first on stderr: rbcop: running bundle exec rubocop for 4 cops it does not implement (ThreadSafety: 4); set RBCOP_FALLBACK=0 to never run Ruby (stdout is unchanged; no line under RBCOP_FALLBACK=1, which asked for it). Under --lsp, --mcp, the server options, --editor-mode or --stdin (an editor opening a project), wherever they come from (command line, .rubocop, RUBOCOP_OPTS), the fallback and the hand-off run only with RBCOP_FALLBACK=1. RBCOP_FALLBACK=0 turns it off: no Ruby is started for it at all. rbcop also runs ruby to render ERB in config files and to fetch remote inherit_from URLs, only when the config has them; docs/known-gaps.md ("Security: when rbcop runs Ruby") lists every case and how to avoid each.

With no RuboCop found, or with RBCOP_FALLBACK=0, rbcop skips those cops and says so on stderr instead of skipping them silently:

rbcop: not running 45 cops it does not implement (Sorbet: 41, ThreadSafety: 4); set RBCOP_FALLBACK=1 to run them with RuboCop rbcop: RuboCop runs them when one is found: add the rubocop gem to the bundle, put rubocop on PATH, or set RBCOP_RUBOCOP

(the second line only when none was found). Cops defined by Ruby code the config requires (require: ./lib/cops/my_cop.rb, or a plugin gem rbcop could not load) are named as "cops from Ruby code the config requires", since rbcop cannot know their names. A plugin that defines no cops but works through Ruby code (rubocop-erb's ERB extraction) is named as "the Ruby code of plugin rubocop-erb": rbcop runs no cops on the files its AllCops: Include brings in (the .erb views), and with the fallback RuboCop runs every cop on them. RBCOP_FALLBACK=1 runs the fallback with rubocop even when none was found.

-r/--require, --plugin and custom formatter classes (-r ./my_formatter.rb --format MyFormatter) load Ruby code that can change anything, so rbcop hands the whole run to the same RuboCop with the same arguments: the output is RuboCop's, with no speedup. With no RuboCop found, or RBCOP_FALLBACK=0, rbcop refuses them (exit 2).

scripts/oracle.sh records RuboCop's offenses on a corpus with every cop enabled (config/all-cops.yml), and scripts/check-cops.sh compares rbcop with those recordings. On nine corpora every implemented cop matches with zero differences:

scripts/check-cold.sh covers cops that rarely fire on those corpora, using RuboCop's own spec examples with their config variants: 1,443 offenses, 0 differences, and -A output identical to a live rubocop -A run. scripts/check-autocorrect.sh and scripts/check-autocorrect-all.sh compare -a/-A output with RuboCop on corpus files. Files skipped by the gates are listed in scripts/check-skip.txt, scripts/cold-skip.txt and scripts/autocorrect-skip.txt, with the reasons in docs/known-gaps.md.

CI-shaped runs (Linux container, 4 vCPUs, each project's own config, RuboCop with an empty result cache, median of 3): rbcop is 10x to 19x faster than rubocop --parallel, about 14x on average.

rbcop with a warm result cache (an unchanged tree, as RuboCop's cache serves it) takes about 3% of the CPU time of an uncached run: 0.9 s of CPU instead of 27 s on rails, 3.0 s instead of 82 s on discourse. Method, 2 vCPU numbers and raw data: docs/benchmarks.md.

cargo fmt --check cargo clippy --release --all-targets -- -D warnings cargo test --release RBCOP_CORPUS=<corpus dir> RBCOP_ORACLE=<oracle dir> scripts/verify.sh # the above plus check-cops.sh all

cargo test replays recorded RuboCop output: every cop's RuboCop spec examples with their config variants, offenses and -A results (tests/spec_corpus, recorded by scripts/record-spec-corpus.sh), config scenarios (tests/scenarios), formatter and CLI goldens (tests/formatters), --show-cops (tests/show_cops) and autocorrect cases (tests/autocorrect). The scripts that record them (scripts/record-spec-corpus.sh, scripts/scenarios.sh, scripts/formatters.sh, scripts/show-cops.sh, tests/autocorrect/record.sh) and the oracle, autocorrect and cold-corpus gates need RuboCop 1.92.2 as $RUBOCOP or on PATH. Each script's header documents its arguments and environment.

docs/porting-cops.md describes how a cop is ported and checked, including how the corpora and oracles are laid out. docs/rubocop-config-semantics.md describes the configuration rules src/config/ follows. docs/upgrading-rubocop.md is the process for moving to a new RuboCop release.

rbcop is released under the MIT license (see LICENSE).