pathlint 0.0.16

Lint the PATH environment variable against declarative ordering rules.
Documentation
# pathlint

[![crates.io](https://img.shields.io/crates/v/pathlint.svg)](https://crates.io/crates/pathlint)
[![CI](https://github.com/ShortArrow/pathlint/actions/workflows/ci.yml/badge.svg)](https://github.com/ShortArrow/pathlint/actions/workflows/ci.yml)
[![License: MIT OR Apache-2.0](https://img.shields.io/crates/l/pathlint.svg)](#license)

> Verify that each command on `PATH` resolves from the installer you expect.

> **⚠ Pre-alpha (0.0.x).** Schema and CLI surface are still moving;
> until 0.1.0 lands, both minor and patch releases may break the
> TOML schema or the CLI. The current 0.0.x binary is functional —
> just don't bake it into anything load-bearing yet.

---

## Why

Most "PATH problems" come from one place: **the wrong copy of an
executable resolves first.** Examples:

- I `cargo install runex` on this machine, but the binary that runs
  is the older one from `winget` — same name, different file.
- `python` should come from `mise`, not from the Microsoft Store
  `windows_apps` stub.
- `node` should come from `volta`, not from the system `apt` install.
- macOS `gcc` should come from Homebrew, not from `/usr/bin/gcc`.

`which python` will tell you what wins, but won't tell you whether
that's what *should* win in a form you can commit to a dotfiles repo
and check on every machine.

`pathlint` makes that intent explicit: write down "**`runex` should
come from `cargo`, not from `winget`**" once, and the tool checks it
on every machine you own.

## How it works

Two TOML concepts:

1. **`[[expect]]`** — per-command expectations. "command X should be
   resolved from source S." This is what users actually write.
2. **`[source.<name>]`** — how to recognize an installer on disk
   ("`cargo` lives at `~/.cargo/bin`"). pathlint ships built-in
   defaults for `cargo`, `mise`, `volta`, `aqua`, `winget`, `choco`,
   `scoop`, `brew_arm`, `brew_intel`, `apt`, `pacman`, `dnf`, `pkg`,
   `flatpak`, `snap`, `windows_apps`, and more — users only override
   when their layout is non-standard.

For each `[[expect]]`, pathlint resolves the command against the real
PATH, looks at where the winning binary lives, and matches that
location to the source labels.

## Status

The 0.0.x line ships six subcommands: `check` (default), `doctor`,
`trace`, `sort`, `init`, and `catalog` (with `list` and
`relations`). `pathlint where` is kept as a visible alias of
`pathlint trace`, and `--rules` is kept as a visible alias of
`--config`. Both aliases are slated for removal in a future
release; the exact timing is undecided and will be announced
ahead of the breaking version. The TOML schema and CLI surface
are still moving, but the resolve / match / report pipeline is
in place and covered by tests. See [docs/PRD.md](docs/PRD.md)
for the full design.

## What pathlint *won't* tell you

`pathlint` is **path-prefix based**: it resolves the command, looks at
the resolved binary's full path, and asks "does any defined source's
per-OS path appear in it as a substring?". That makes it fast (no
package-manager calls, no network), but it leaves blind spots you
should know about:

- **AUR / Homebrew tap / `make install` / any custom prefix.** If a
  binary lands somewhere not listed in your `[source.<name>]` entries,
  `pathlint` reports `NG (unknown source)` even when the install is
  legitimate. Add a `[source.my_prefix]` for it, or accept that
  pathlint can't tell that case apart from a real misordering.
- **Symlinked system dirs.** On Arch / openSUSE TW / Solus,
  `/usr/sbin → /usr/bin`. `which ls` reports `/usr/sbin/ls`, so the
  built-in `apt` / `pacman` / `dnf` source (`/usr/bin`) doesn't match
  alone. Reference the built-in `os_baseline_linux_sbin` source
  alongside the package manager:

  ```toml
  [[expect]]
  command = "ls"
  prefer = ["pacman", "os_baseline_linux_sbin"]
  ```
- **Which package owns this binary.** `pathlint` does not call
  `dpkg -S` / `rpm -qf` / `pacman -Qo` / `brew which-formula`. That's
  intentional in 0.0.x for speed and offline correctness; revisiting
  is on the 0.2 list.

The full set of known limitations and future trade-offs lives in
[docs/PRD.md §14, §16](docs/PRD.md).

## Usage

```sh
# Check the current process PATH against ./pathlint.toml
pathlint                          # = pathlint check

# Check the User-only or Machine-only PATH (Windows registry)
pathlint --target user
pathlint --target machine

# Verbose: also show n/a expectations and the resolved PATH
pathlint --verbose

# Explain why each NG fired (resolved / matched / prefer / avoid /
# diagnosis / hint), 0.0.7+
pathlint check --explain

# Same data, machine-readable (CI), 0.0.7+
pathlint check --json

# Drop a starter pathlint.toml in the current directory
pathlint init
pathlint init --emit-defaults     # also embeds the full source catalog

# Inspect every known source (built-in + user-defined)
pathlint catalog list             # paths for the running OS
pathlint catalog list --all       # every per-OS field
pathlint catalog list --names-only
pathlint catalog relations        # 0.0.9+: declared source relations
pathlint catalog relations --json # 0.0.9+: same, machine-readable

# Find a command's provenance and uninstall hint
pathlint trace lazygit            # who installed this binary?
pathlint trace lazygit --json     # 0.0.6+: machine-readable output

# Filter doctor diagnostics for CI
pathlint doctor --exclude shortenable,missing
pathlint doctor --include duplicate,malformed
pathlint doctor --json            # 0.0.7+: machine-readable output

# Propose a PATH order satisfying every [[expect]] rule (read-only)
pathlint sort                     # 0.0.8+: before/after diff
pathlint sort --json              # 0.0.8+: SortPlan JSON
```

## `pathlint.toml` (minimal example)

```toml
[[expect]]
command = "runex"
prefer  = ["cargo"]
avoid   = ["winget"]

[[expect]]
command = "python"
prefer  = ["mise"]
avoid   = ["windows_apps", "choco"]

[[expect]]
command = "node"
prefer  = ["mise", "volta"]

[[expect]]
command = "gcc"
prefer  = ["mingw", "msys"]
avoid   = ["strawberry"]
os      = ["windows"]
```

Add `severity = "warn"` to a rule to keep its NG visible without
blocking CI (exit stays 0; the line is tagged `[warn]` instead of
`[NG]`):

```toml
[[expect]]
command  = "rg"
prefer   = ["cargo"]
severity = "warn"   # 0.0.7+ — nudge, not a hard fail
```

Add `kind = "executable"` to also verify the resolved path is an
actual executable file — catches the case where a directory of the
same name shadows the binary, or where the file the symlink points
at has gone missing:

```toml
[[expect]]
command = "rustc"
prefer  = ["cargo"]
kind    = "executable"
```

No `[source.*]` section is needed for any of the names above —
they're all in the built-in catalog. The whole file is the user's
intent.

To override a built-in (mise installed in a non-standard location):

```toml
[source.mise]
windows = "D:/tools/mise"
```

To add a new source:

```toml
[source.my_dotfiles_bin]
unix = "$HOME/dotfiles/bin"
```

`os = [...]` accepts `windows | macos | linux | termux | unix`.
Match is substring + case-insensitive, after env-var expansion (both
`%VAR%` and `$VAR` work everywhere) and slash normalization.

## Working with mise

mise serves binaries from two distinct places, and pathlint exposes
each as its own source so rules can be specific:

- **`mise_shims`**`$HOME/.local/share/mise/shims/<bin>` on Unix,
  `$LocalAppData/mise/shims/<bin>` on Windows. This is the layer
  shells front-load when you run `mise activate`. It's the
  recommended source to reference in `prefer` for most rules.
- **`mise_installs`**`$HOME/.local/share/mise/installs/<tool>/<ver>/bin/<bin>`.
  Hit when `mise activate` rewrites PATH directly (no shims), or
  when a plugin (`cargo-*`, `npm-*`, ...) ships its bin under
  `installs/<plugin>/<ver>/bin`.
- **`mise`** — catch-all that matches both layers. Useful when you
  don't care which mise mode is in use; rules written before 0.0.3
  keep working unchanged.

```toml
# Strict: only accept mise's shim layer.
[[expect]]
command = "python"
prefer  = ["mise_shims"]

# Looser: anything mise serves is fine.
[[expect]]
command = "node"
prefer  = ["mise"]
```

`pathlint trace <command>` is plugin-aware: when the resolved
binary lives under `mise/installs/<segment>/...` and `<segment>`
starts with `cargo-` / `npm-` / `pipx-` / `go-` / `aqua-`, the
output adds a `provenance:` line and a `mise uninstall ...` hint
so you don't have to remember which plugin you used:

```
$ pathlint trace lazygit
lazygit
  resolved: ~/.local/share/mise/installs/cargo-jesseduffield-lazygit/0.61/bin/lazygit
  sources:  mise_installs, mise
  provenance: cargo (via mise plugin `cargo-jesseduffield-lazygit`)
  hint:     mise uninstall cargo:'jesseduffield-lazygit'  (best-guess; verify with `mise plugins ls`)
```

The provenance is a path heuristic — it never causes
`prefer = ["cargo"]` to match a mise-served binary. Source
labels stay catalog-driven; provenance is purely a `where`
display.

If you set `MISE_DATA_DIR` or `XDG_DATA_HOME` to a non-standard
location, override the three sources in your `pathlint.toml`:

```toml
[source.mise]
unix = "/data/tools/mise"

[source.mise_shims]
unix = "/data/tools/mise/shims"

[source.mise_installs]
unix = "/data/tools/mise/installs"
```

## Editor support (JSON Schema)

`pathlint.toml` ships with a JSON Schema generated from the live
Rust types. Add this single line at the top of your config to get
autocomplete and inline validation in [Taplo] (the dominant TOML
LSP) and the [Even Better TOML][ebt] VS Code extension:

```toml
#:schema https://raw.githubusercontent.com/ShortArrow/pathlint/main/schemas/pathlint.schema.json
```

Pin to a specific release for reproducibility (replace `<TAG>`
with the version you want, e.g. `v0.0.14`):

```toml
#:schema https://github.com/ShortArrow/pathlint/releases/download/<TAG>/pathlint.schema.json
```

The schema is also attached as `pathlint.schema.json` to every
GitHub Release alongside the binaries. As of 0.0.15 each release
also ships `check.schema.json`, which describes the JSON shape of
`pathlint check --json` output for downstream consumers.

[Taplo]: https://taplo.tamasfe.dev/
[ebt]: https://marketplace.visualstudio.com/items?itemName=tamasfe.even-better-toml

## Pinning the catalog version

The built-in source catalog evolves: a new pathlint version may
change a source's per-OS path (because `winget` reshuffled its
layout, say). If you want a guarantee that your `pathlint.toml`
runs against a sufficiently fresh catalog, declare a minimum:

```toml
require_catalog = 1
```

When the running binary embeds an older catalog, pathlint exits
with code 2 and a message naming the gap, instead of silently
matching against stale rules. `pathlint catalog list` prints the
embedded version on its first line so you can pick a value.

The opposite direction is not enforced — running against a *newer*
catalog is always fine. Bumping `catalog_version` is reserved for
real path or semantics changes; adding a new source does not bump
it, so old rules don't break.

## Installation

```sh
# From crates.io
cargo install pathlint

# From source (latest main)
cargo install --git https://github.com/ShortArrow/pathlint

# Pre-built binaries
# https://github.com/ShortArrow/pathlint/releases
# Linux x86_64 / Windows x86_64 / macOS x86_64 / macOS aarch64
```

## Documentation

- [日本語 README]docs/README.jp.md
- [PRD (English)]docs/PRD.md — the full design, including the
  built-in source catalog
- [PRD (日本語)]docs/PRD.jp.md
- [Release process]docs/RELEASE.md — how to cut a new version
- [リリース手順 (日本語)]docs/RELEASE.jp.md
- [Releases]https://github.com/ShortArrow/pathlint/releases — version history with auto-generated notes

## License

Licensed under either of:

- Apache License, Version 2.0 ([LICENSE-APACHE]LICENSE-APACHE or <http://www.apache.org/licenses/LICENSE-2.0>)
- MIT license ([LICENSE-MIT]LICENSE-MIT or <http://opensource.org/licenses/MIT>)

at your option.

### Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.