Semantic linter for Nixpkgs using tree-sitter 🌳 + ❄️

This is a semantic linter for Nixpkgs that uses tree-sitter. Currently we have the following detections:

• cmake, makeWrapper, pkg-config in buildInputs
• redundant packages from stdenv in nativeBuildInputs

Features

• Fast: lints all of Nixpkgs in under 3 seconds
• Semantic linting: forget about hacking up regexes, we run queries directly on parse trees created by tree-sitter
• Syntax-aware: nixpkgs-lint can easily handle multi-line expressions, eliminates false-positives from strings and comments and gives exact spans for matches
• Robust: lint Nix files even in the presence of syntax errors
• Hackable: create your own lints by writing queries or Rust code

Usage

To use without installing, run nix run github:nix-community/nixpkgs-lint.

The tool will recurse through every .nix file in the provided path(,s).

$ nix build 
$ ./result/bin/nixpkgs-lint <files or directories>

Motivation

Why another linter? My motivation for this was spawned after doing a series of treewide PRs such as moving cmake to buildInputs. The strategy was similar each time; write some shell one-liner to go through every file (27,000+ of them) in Nixpkgs and find some anti-pattern to fix. However, this is quickly problematic for multiple reasons:

• it is hard to account for multi-line expressions
• it is hard to filter out false positives
• it is hard to query for more complex features

In general discussions on IRC and Matrix, a more AST-aware approach to linting was viewed favorably but not many people took it on, despite the availability of Nix parsers in various languages. I have some subjective reasons myself:

• need to learn the AST representation in the respective library
• need to traverse the AST with a query
• need to locate this information back to a source location

Often one or more of these would be pretty involved. Furthermore, it locks you into a specific parser (which may or may not have provided source information, parse all things correctly, etc.).

Enter tree-sitter. The Nix grammar for tree-sitter has been well-tested and tree-sitter having bindings in several languages gives you options in how to work with the resulting AST. You also get things like a location-annotated AST and error recovery for free.

License

This repository is licensed under the MIT license.