[cestercian]

Summary

• rtk ruff format --check compressed nothing on ruff 0.12+ because filter_ruff_format only matched pre-0.12 wording (Would reformat: / left unchanged).
• Current ruff (verified on 0.16.6) emits unformatted: per file and N files already formatted as the summary, plus a full diagnostic. Neither old branch fired, so the filter fell through to identity (0% savings).
• Write mode still says left unchanged and was already fine; only the --check path was dead.

Root cause: substring checks were too specific. "would reformat" does not match "would be reformatted", and "left unchanged" does not match "already formatted".

Fix: recognize both generations of wording. Parse unformatted: <path> by splitting after the first colon (Windows-safe). Also collect paths from the default full diagnostic and from --output-format concise. Keep the pre-0.12 fixtures green.

Fixes #3930

Test plan

• cargo fmt --all && cargo clippy --all-targets && cargo test --all
• Unit fixtures: pre-0.12 plus current check output (unformatted:, full diagnostic, concise, already formatted)

Notes

• Draft PR refactor(cmds): drop module-name stutter from filter fn names #3912 (cosmetic rename) was not stacked; rebase if it lands first.
• Authored by Cestercian. Prepared with Cursor/AI assistance; no AI co-author trailer.

All PRs must target develop (not master). See CONTRIBUTING.md.

[CLAassistant]

All committers have signed the CLA.

[rtk-wshm-sync-bot]

wshm · Automated triage by AI

📊 Automated PR Analysis

🐛 Type	bug-fix
🟢 Risk	low

Summary

Fixes rtk ruff format --check so it correctly compresses output on ruff 0.12+, which changed its wording from 'Would reformat:'/'left unchanged' to 'unformatted:'/'N files already formatted'. The fix updates filter_ruff_format to recognize both old and new wording generations, parsing paths from the 'unformatted:' line, full diagnostics, and concise output format.

Review Checklist

• Tests present
• Breaking change
• Docs updated

Linked issues: #3930

---

Analyzed automatically by wshm · This is an automated analysis, not a human review.

[Protocol-zero-0]

Tested the current head against its own parent (8e9aa04) with Ruff 0.17.0 on Linux: 7 scenarios / 21 raw-parent-head commands. The default full-format and concise controls improve as intended. One regression remains for explicitly selected structured format output: failing JSON/JSON-lines checks lose the diagnostic records and fix edits, while the parent preserves them. All tested exit codes remain unchanged; the clean JSON case still returns [].

[Protocol-zero-0]

[P2] Preserve explicitly requested JSON/JSON-lines format diagnostics

The new would be reformatted match also fires inside Ruff JSON records (message: "File would be reformatted"). With bad.py containing x= [1,2,3] and clean.py containing x = [1, 2, 3], run rtk ruff format --check --output-format=json bad.py clean.py: this head replaces the JSON array with Ruff format: files need formatting plus a hint, losing the filename, location and fix edits. Its own parent 8e9aa04 preserves the same parsed records as raw Ruff. The separated --output-format json, 40 dirty files and json-lines cases reproduce this too; exit 1 is preserved.

Please keep the structured format modes on a raw-output path rather than applying the text formatter to their contents, and add a regression check for the failing JSON/JSON-lines cases (a clean [] alone stays valid and would miss this). This follows CONTRIBUTING.md's explicit-format/correctness and transparency rules.

[cestercian]

Good catch, thanks for testing this so thoroughly. When you pick json / json-lines (or any other non-text --output-format), rtk ruff format now just passes Ruff's output through untouched. I also added a guard inside filter_ruff_format so it never summarizes anything that starts as JSON. There are new tests covering failing JSON and json-lines records (filename + fix edits survive), and both fail without the change. I ran your bad.py / clean.py repro on ruff 0.17.0 too: json and json-lines come back the same as raw Ruff, the default and concise output still get compressed, and the exit code stays 1.

[Protocol-zero-0]

Re-tested exact head 693fe83 with Ruff 0.17.0 on Linux: 7 scenarios / 21 raw-previous-head executions. My previous structured-output blocker is fixed: separated/equal JSON options, a clean result, 40 dirty files and JSON-lines preserve the parsed raw diagnostic records, including filenames and fix edits. The prior reviewed head 77fa73f still reproduces the lossy cases. Default full and concise controls keep their exit codes. 17 focused Ruff tests, fmt and clippy with warnings denied pass.

One integration requirement remains before I can approve: the new format_output_is_text helper hand-scans --output-format. The repository's .claude/rules/rust-patterns.md rule 6 requires new argument handling to use the shared arg_tokenizer; please classify this option through that API and retain both spellings and the new failing-output regressions. This is a repository requirement, not a claim that the repaired JSON cases still fail.

GitHub currently reports a conflict with develop. Please keep the rebase and the choice between this PR and #3965 coordinated with the maintainer; they overlap. I tested this exact head in isolation, not a combined merge or other platforms.

[Protocol-zero-0]

Use the shared arg_tokenizer for this new option classification, as required by Rust Patterns rule 6. This manual loop adds a second parser for split/equal --output-format values. Keep the successful JSON/JSON-lines regression tests while moving the classification to the shared grammar API. The behavioral blocker is fixed in my real-Ruff rerun; this remaining request is about the project’s required argument-handling interface.

[cestercian]

Thanks for retesting! Rebased onto develop and switched format_output_is_text over to arg_tokenizer::tokenize_grammar with --output-format as a value-taking flag, so both spellings go through the shared tokenizer now and anything after -- is ignored. Last occurrence wins, same as ruff. Kept the json / json-lines / sarif regressions and added a couple for repeated flags. Also tidied two collapsible ifs clippy flagged on the newer toolchain. Happy to go with whichever of this or #3965 the maintainers prefer.

[Protocol-zero-0]

Re-reviewed 91b04b5 with real Ruff 0.17.0 on Linux. This clears my shared-tokenizer blocker: format output selection uses tokenize_grammar, respects split/equal value forms and ignores tokens after --.

Eleven CLI scenarios / 33 raw-previous-head comparisons cover failing/clean/55-file JSON, JSON-lines, SARIF, text controls, a literal filename --output-format=json after the separator, and repeated output-format flags. Structured payloads match raw Ruff, including filenames and fix edits, and the separator filename follows the text-format path. All head exit codes match raw Ruff. Seventeen Ruff unit tests, fmt and clippy (all targets, warnings denied) pass.

One factual clarification to the description: real Ruff 0.17.0 rejects repeated --output-format flags with exit 2 rather than using the last occurrence; this head preserves that rejection and diagnostic. The helper's last-occurrence policy is therefore not a demonstrated CLI behavior for that Ruff version.

This approval covers the refreshed formatter/option parsing. I did not run the full repository suite, rustdoc or other platforms; choosing between this PR and #3965, and integrating the shared-file changes with #3968/#3933, remains the maintainer's decision.

[cestercian]

Thanks for the approve! Small correction to my earlier note: you're right that real Ruff 0.17 rejects repeated --output-format with exit 2, so "last one wins" was wrong on my part. The tokenizer change just means we pick the value the same way for the single-flag case, and a repeated flag ends up with ruff's own error either way.