poly-lint-and-format

<!-- AI-RULEZ :: GENERATED FILE — DO NOT EDIT Content-Hash: blake3:697a8913634eec72ba2340417d11f811af8d9fcd390f25ccdc09ed4689695209 Source-Hash: blake3:840c919261146454e2836095406355dd3ba8664fc068055c202682015400e43f Schema-Version: v1 --> # poly Lint and Format ## Commands - `poly lint [PATHS]…` — run the linters. poly applies no fixes without `--fix`; `--fix` applies autofixes (and the whole-project fix phase). `--no-workspace` restricts to the per-file tier — and is also what makes a run read-only: without it, plain `poly lint` still *executes* the configured whole-project tools against the live worktree, and their own side effects (a refreshed lock file, a populated build or type-checker cache) are not poly's to control. Naming paths that **narrow** the run skips that phase by default (`poly lint .` names the repo root, so it does *not* narrow and the phase still runs); `--workspace` opts back in, and the tools then cover the whole repository regardless of the named paths and of `[discovery] exclude`. - `poly fmt [PATHS]…` — **dry run by default**: reports what would change and writes nothing. `--fix` writes changes; `--check` is the explicit form of the default dry run and conflicts with `--fix`. `poly fmt` is a pure formatter — it never runs the whole-project lint phase. ## Flags - `--format pretty|json|toon` — `pretty` (the default) is the colored, human-oriented output; `json` and the compact `toon` variant are machine-readable. Under `--format json`/`toon`, `poly lint`'s whole-project section and every note (discovery, skips, errors) go to stderr so stdout stays a single valid document — a machine consumer must check the **exit code**, not just the payload. - `--exclude <glob>` — skip paths on top of `.gitignore` (repeatable; merged with `[discovery] exclude`). Gitignore-style: a glob without a leading `/` matches a directory of that name at **any** depth (`e2e/**` also prunes `src/test/java/io/xberg/e2e/`), while a leading `/` anchors it to the config directory (`/e2e/**`). `poly doctor` warns when a rule matches at more than one depth. - `--include-excluded` — check explicitly named files or directory roots even when they match the exclude set. Exclusions *below* an included directory stay active. Applying the exclude set to explicitly named paths is already the default, so `--force-exclude` is accepted only as a compatibility no-op — the flag is parsed and never read. - `--config <path>` — point at a specific `poly.toml`. - `--no-cache` — bypass the blake3 content-hash result cache. - `-j <N>` — parallelism; `--no-color` — plain output; `--verbose` — extra per-finding detail in `pretty` output; `--debug` — per-engine cache hit/miss and timing, plus debug-level logs on stderr. - `--fix-generated` — let `--fix` rewrite files marked `DO NOT EDIT` / `@generated`, which it otherwise reports on but leaves alone. - `--deny-skips` / `--max-skips <N>` — strict coverage. A **skipped** file is one nothing inspected: a path named on the command line that no engine covers (`App.csproj`), or a file every routed backend declined (Go-templated YAML, a hash-stamped generated file). A skip the caller instructed — `--only`/`--skip` narrowing an engine out, or `enabled = false` in config — is **not** charged against the budget, since it names itself in the report rather than losing coverage silently; only a poly limitation counts. Skips are always reported and named; these flags make a chargeable skip fail the run (exit `2`), naming every file it fired on. `--verbose` lists every skip in `pretty` output; `--format json`/`toon` carries the full set both as a top-level `skipped` array and as synthetic `results` entries — `summary.skipped` is the authoritative count, so read it first and use the set only to name the files. ## Exit codes - `0` — clean (no findings, no drift). - `1` — error-severity findings, formatting drift, or a failing whole-project tool. - `2` — an error (bad config, tool failure), or work the run could not verify: a missing path argument, a file an engine failed on — including a `poly fmt` file that could not reach a fixed point within its five-pass cap, now reported as an error rather than silently claimed as formatted — or a `--deny-skips`/`--max-skips` breach. `poly lint` exits non-zero only on **error-severity** findings; warnings do not fail CI — which is why the on-by-default `quality` and built-in ast-grep rules (all warnings) never turn a green pipeline red on their own. ## Inline suppression Silence one finding in place with a directive written in the host language's own comment syntax: ```rust let value = map.get(key).unwrap(); // poly: allow[unwrap-used] key is validated above ``` - `poly: allow[RULE] reason` — a comment-only line suppresses the next non-blank line; a trailing comment suppresses its own line. - `poly: allow-file[RULE] reason` — suppresses the whole file wherever it appears, and is the only form that can suppress a diagnostic with no span. - `[RULE]` accepts a comma-separated list, or `*` for every rule. - **A reason is mandatory.** A directive whose text after `]` has no alphanumeric character does **not** suppress, and reports `lazy-ignore` instead — an empty comment cannot buy a bypass. - Applied centrally in the runner, so it works for every engine (ruff, oxlint, quality, ast-grep, …). For a whole file or a class of files, use `[per-file-ignores]` in `poly.toml` instead. ## The loop 1. `poly fmt --check . --format json` and `poly lint . --format json` — capture drift and findings, checking the exit code. 2. Read the JSON to see exactly which files and rules are involved. 3. `poly fmt --fix .` then `poly lint --fix .` to apply what is auto-fixable. 4. Re-run the checks; hand-fix whatever remains (exit code back to 0).