Skip to content

137_fix-dry-run.md

Latest commit

 

History

History
166 lines (135 loc) · 4.68 KB

137_fix-dry-run.md

File metadata and controls

166 lines (135 loc) · 4.68 KB
Raw
id 137
title `mdsmith fix --dry-run`
status 🔲
model sonnet
summary Add `--dry-run` to `mdsmith fix` so agents and CI scripts can preview the file changes before writing. Emits the same diagnostics as `check` plus a per-file count of fixes that would apply.

mdsmith fix --dry-run

Goal

Let a user run mdsmith fix --dry-run and see exactly which files would change, which rules would fire, and how many violations each file carries. No bytes are written. The exit code matches what fix would have returned, so a CI script can gate on dry-run output.

Background

Plan C-6 in the mdbase research calls this out as a quick ergonomic win. The trigger is an agent loop or CI run that wants a safety net before writing.

Today the only preview is mdsmith check, which reports diagnostics but does not say which would be auto-fixed. A user who wants to know "what will fix change?" has to diff the worktree afterward.

The fix engine already produces a fixed-content buffer per file before writing. Gating the write behind a flag is small.

Non-Goals

  • Diff output. The flag prints which rules would fire and how many; a textual diff per file is out of scope. Users who want a diff can run mdsmith fix against a copy.
  • Partial application. The flag is all-or-nothing per run. No "fix MDS001 but preview MDS006".
  • A separate subcommand. --dry-run on fix matches the convention other CLIs use.

Design

Flag behavior

mdsmith fix --dry-run docs/

Output, one line per file with at least one fix:

docs/api.md: would fix 3 violations (MDS001 ×2, MDS006)
docs/index.md: would regenerate <?catalog?> body

Files with no fixes do not appear. A trailing summary matches the existing fix summary line:

stats: checked=12 fixed=0 failures=0 unfixed=4 would-fix=8

The dry-run summary keeps the existing checked= / fixed= / failures= / unfixed= fields for machine-parsability. fixed=0 is always literal-zero on a dry run, since nothing was written. would-fix=N is additive — the count of violations a real run would have auto-fixed. A consumer that only watches fixed= therefore sees 0 (no surprise apply); a consumer that watches would-fix= sees the preview count.

JSON output

--format json emits one record per file:

{"path":"docs/api.md","would_fix":3,
 "rules":["MDS001","MDS006"],"diagnostics":[]}

The diagnostics array carries the same shape check --format json returns today.

Exit code

The exit code is the same fix would have returned without --dry-run:

  • 0 — every diagnostic is fixable; a real run would leave the worktree clean.
  • non-zero — at least one unfixable diagnostic remains.

This lets a CI step run dry-run as a gate on "fix can clean this PR".

Compatibility

--dry-run is additive. Default behavior is unchanged. The new line in the summary (would-fix=N) is suppressed on a real run.

Tasks

  1. Add --dry-run to the fix subcommand flag set in runFix in cmd/mdsmith/main.go.
  2. Gate the write step in the fix pipeline: build the fixed buffer as today, but on --dry-run skip the write and record the would-fix count per rule.
  3. Update the per-file output formatter to emit "would fix N violations" lines and the trailing would-fix=N summary stat.
  4. Extend the JSON output struct with would_fix and rules fields when --dry-run is set.
  5. Document the flag in docs/reference/cli/fix.md with one worked example.
  6. Tests:
  • dry-run reports the same fix count a real run would apply (regression compares both),
  • dry-run leaves the worktree byte-identical,
  • dry-run exit code matches the real-run exit code on the same input,
  • JSON output exposes would_fix and rules.

Acceptance Criteria

  • mdsmith fix --dry-run writes nothing to disk (regression test asserts every candidate file is byte-identical after the run).
  • The per-file output names each rule that would fire and how many times.
  • The summary line includes would-fix=N. The existing checked= / fixed= / failures= / unfixed= fields are present on the same line; fixed= reads 0 since nothing was written.
  • --format json exposes would_fix and rules per file.
  • Exit code matches the real-run exit code on identical input.
  • docs/reference/cli/fix.md documents the flag with an example.
  • All tests pass: go test ./...
  • go tool golangci-lint run reports no issues.