rust/doc/adding_lints.md

Adding a new lint

You are probably here because you want to add a new lint to Clippy. If this is the first time you're contributing to Clippy, this document guides you through creating an example lint from scratch.

To get started, we will create a lint that detects functions called foo, because that's clearly a non-descriptive name.

• Adding a new lint
  • Setup
  • Getting Started
  • Testing
  • Rustfix tests
  • Edition 2018 tests
  • Testing manually
  • Lint declaration
  • Lint passes
  • Emitting a lint
  • Adding the lint logic
  • Author lint
  • Documentation
  • Running rustfmt
  • Debugging
  • PR Checklist
  • Cheatsheet

Setup

When working on Clippy, you will need the current git master version of rustc, which can change rapidly. Make sure you're working near rust-clippy's master, and use the setup-toolchain.sh script to configure the appropriate toolchain for the Clippy directory.

Getting Started

There is a bit of boilerplate code that needs to be set up when creating a new lint. Fortunately, you can use the clippy dev tools to handle this for you. We are naming our new lint foo_functions (lints are generally written in snake case), and we don't need type information so it will have an early pass type (more on this later on). To get started on this lint you can run cargo dev new_lint --name=foo_functions --pass=early --category=pedantic (category will default to nursery if not provided). This command will create two files: tests/ui/foo_functions.rs and clippy_lints/src/foo_functions.rs, as well as run cargo dev update_lints to register the new lint. Next, we'll open up these files and add our lint!

Testing

Let's write some tests first that we can execute while we iterate on our lint.

Clippy uses UI tests for testing. UI tests check that the output of Clippy is exactly as expected. Each test is just a plain Rust file that contains the code we want to check. The output of Clippy is compared against a .stderr file. Note that you don't have to create this file yourself, we'll get to generating the .stderr files further down.

We start by opening the test file created at tests/ui/foo_functions.rs.

Update the file with some examples to get started:

#![warn(clippy::foo_functions)]

// Impl methods
struct A;
impl A {
    pub fn fo(&self) {}
    pub fn foo(&self) {}
    pub fn food(&self) {}
}

// Default trait methods
trait B {
    fn fo(&self) {}
    fn foo(&self) {}
    fn food(&self) {}
}

// Plain functions
fn fo() {}
fn foo() {}
fn food() {}

fn main() {
    // We also don't want to lint method calls
    foo();
    let a = A;
    a.foo();
}

Now we can run the test with TESTNAME=foo_functions cargo uitest, currently this test is meaningless though.

While we are working on implementing our lint, we can keep running the UI test. That allows us to check if the output is turning into what we want.

Once we are satisfied with the output, we need to run tests/ui/update-all-references.sh to update the .stderr file for our lint. Please note that, we should run TESTNAME=foo_functions cargo uitest every time before running tests/ui/update-all-references.sh. Running TESTNAME=foo_functions cargo uitest should pass then. When we commit our lint, we need to commit the generated .stderr files, too.

Rustfix tests

If the lint you are working on is making use of structured suggestions, the test file should include a // run-rustfix comment at the top. This will additionally run rustfix for that test. Rustfix will apply the suggestions from the lint to the code of the test file and compare that to the contents of a .fixed file.

Use tests/ui/update-all-references.sh to automatically generate the .fixed file after running the tests.

Edition 2018 tests

Some features require the 2018 edition to work (e.g. async_await), but compile-test tests run on the 2015 edition by default. To change this behavior add // edition:2018 at the top of the test file (note that it's space-sensitive).

Testing manually

Manually testing against an example file can be useful if you have added some println!s and the test suite output becomes unreadable. To try Clippy with your local modifications, run env CLIPPY_TESTS=true cargo run --bin clippy-driver -- -L ./target/debug input.rs from the working copy root.

With tests in place, let's have a look at implementing our lint now.

Lint declaration

Let's start by opening the new file created in the clippy_lints crate at clippy_lints/src/foo_functions.rs. That's the crate where all the lint code is. This file has already imported some initial things we will need:

use rustc_lint::{EarlyLintPass, EarlyContext};
use rustc_session::{declare_lint_pass, declare_tool_lint};
use rustc_ast::ast::*;

The next step is to update the lint declaration. Lints are declared using the declare_clippy_lint! macro, and we just need to update the auto-generated lint declaration to have a real description, something like this:

declare_clippy_lint! {
    /// **What it does:**
    ///
    /// **Why is this bad?**
    ///
    /// **Known problems:** None.
    ///
    /// **Example:**
    ///
    /// ```rust
    /// // example code
    /// ```
    pub FOO_FUNCTIONS,
    pedantic,
    "function named `foo`, which is not a descriptive name"
}

• The section of lines prefixed with /// constitutes the lint documentation section. This is the default documentation style and will be displayed [like this][example_lint_page].
• FOO_FUNCTIONS is the name of our lint. Be sure to follow the lint naming guidelines here when naming your lint. In short, the name should state the thing that is being checked for and read well when used with allow/warn/deny.
• pedantic sets the lint level to Allow. The exact mapping can be found here
• The last part should be a text that explains what exactly is wrong with the code

The rest of this file contains an empty implementation for our lint pass, which in this case is EarlyLintPass and should look like this:

// clippy_lints/src/foo_functions.rs

// .. imports and lint declaration ..

declare_lint_pass!(FooFunctions => [FOO_FUNCTIONS]);

impl EarlyLintPass for FooFunctions {}

Normally after declaring the lint, we have to run cargo dev update_lints, which updates some files, so Clippy knows about the new lint. Since we used cargo dev new_lint ... to generate the lint declaration, this was done automatically. While update_lints automates most of the things, it doesn't automate everything. We will have to register our lint pass manually in the register_plugins function in clippy_lints/src/lib.rs:

store.register_early_pass(|| box foo_functions::FooFunctions);

Lint passes

Writing a lint that only checks for the name of a function means that we only have to deal with the AST and don't have to deal with the type system at all. This is good, because it makes writing this particular lint less complicated.

We have to make this decision with every new Clippy lint. It boils down to using either EarlyLintPass or LateLintPass.

In short, the LateLintPass has access to type information while the EarlyLintPass doesn't. If you don't need access to type information, use the EarlyLintPass. The EarlyLintPass is also faster. However linting speed hasn't really been a concern with Clippy so far.

Since we don't need type information for checking the function name, we used --pass=early when running the new lint automation and all the imports were added accordingly.

Emitting a lint

With UI tests and the lint declaration in place, we can start working on the implementation of the lint logic.

Let's start by implementing the EarlyLintPass for our FooFunctions:

impl EarlyLintPass for FooFunctions {
    fn check_fn(&mut self, cx: &EarlyContext<'_>, fn_kind: FnKind<'_>, span: Span, _: NodeId) {
        // TODO: Emit lint here
    }
}

We implement the check_fn method from the EarlyLintPass trait. This gives us access to various information about the function that is currently being checked. More on that in the next section. Let's worry about the details later and emit our lint for every function definition first.

Depending on how complex we want our lint message to be, we can choose from a variety of lint emission functions. They can all be found in clippy_lints/src/utils/diagnostics.rs.

span_lint_and_help seems most appropriate in this case. It allows us to provide an extra help message and we can't really suggest a better name automatically. This is how it looks:

impl EarlyLintPass for FooFunctions {
    fn check_fn(&mut self, cx: &EarlyContext<'_>, fn_kind: FnKind<'_>, span: Span, _: NodeId) {
        span_lint_and_help(
            cx,
            FOO_FUNCTIONS,
            span,
            "function named `foo`",
            "consider using a more meaningful name"
        );
    }
}

Running our UI test should now produce output that contains the lint message.

Adding the lint logic

Writing the logic for your lint will most likely be different from our example, so this section is kept rather short.

Using the check_fn method gives us access to FnKind that has the FnKind::Fn variant. It provides access to the name of the function/method via an Ident.

With that we can expand our check_fn method to:

impl EarlyLintPass for FooFunctions {
    fn check_fn(&mut self, cx: &EarlyContext<'_>, fn_kind: FnKind<'_>, span: Span, _: NodeId) {
        if is_foo_fn(fn_kind) {
            span_lint_and_help(
                cx,
                FOO_FUNCTIONS,
                span,
                "function named `foo`",
                "consider using a more meaningful name"
            );
        }
    }
}

We separate the lint conditional from the lint emissions because it makes the code a bit easier to read. In some cases this separation would also allow to write some unit tests (as opposed to only UI tests) for the separate function.

In our example, is_foo_fn looks like:

// use statements, impl EarlyLintPass, check_fn, ..

fn is_foo_fn(fn_kind: FnKind<'_>) -> bool {
    match fn_kind {
        FnKind::Fn(_, ident, ..) => {
            // check if `fn` name is `foo`
            ident.name.as_str() == "foo"
        }
        // ignore closures
        FnKind::Closure(..) => false
    }
}

Now we should also run the full test suite with cargo test. At this point running cargo test should produce the expected output. Remember to run tests/ui/update-all-references.sh to update the .stderr file.

cargo test (as opposed to cargo uitest) will also ensure that our lint implementation is not violating any Clippy lints itself.

That should be it for the lint implementation. Running cargo test should now pass.

Author lint

If you have trouble implementing your lint, there is also the internal author lint to generate Clippy code that detects the offending pattern. It does not work for all of the Rust syntax, but can give a good starting point.

The quickest way to use it, is the Rust playground: play.rust-lang.org. Put the code you want to lint into the editor and add the #[clippy::author] attribute above the item. Then run Clippy via Tools -> Clippy and you should see the generated code in the output below.

Here is an example on the playground.

If the command was executed successfully, you can copy the code over to where you are implementing your lint.

Documentation

The final thing before submitting our PR is to add some documentation to our lint declaration.

Please document your lint with a doc comment akin to the following:

declare_clippy_lint! {
    /// **What it does:** Checks for ... (describe what the lint matches).
    ///
    /// **Why is this bad?** Supply the reason for linting the code.
    ///
    /// **Known problems:** None. (Or describe where it could go wrong.)
    ///
    /// **Example:**
    ///
    /// ```rust,ignore
    /// // Bad
    /// Insert a short example of code that triggers the lint
    ///
    /// // Good
    /// Insert a short example of improved code that doesn't trigger the lint
    /// ```
    pub FOO_FUNCTIONS,
    pedantic,
    "function named `foo`, which is not a descriptive name"
}

Once your lint is merged, this documentation will show up in the lint list.

Running rustfmt

Rustfmt is a tool for formatting Rust code according to style guidelines. Your code has to be formatted by rustfmt before a PR can be merged. Clippy uses nightly rustfmt in the CI.

It can be installed via rustup:

rustup component add rustfmt --toolchain=nightly

Use cargo dev fmt to format the whole codebase. Make sure that rustfmt is installed for the nightly toolchain.

Debugging

If you want to debug parts of your lint implementation, you can use the [dbg!] macro anywhere in your code. Running the tests should then include the debug output in the stdout part.

PR Checklist

Before submitting your PR make sure you followed all of the basic requirements:

• Followed lint naming conventions
• Added passing UI tests (including committed .stderr file)
• cargo test passes locally
• Executed cargo dev update_lints
• Added lint documentation
• Run cargo dev fmt

Cheatsheet

Here are some pointers to things you are likely going to need for every lint:

• Clippy utils - Various helper functions. Maybe the function you need is already in here (implements_trait, match_path, snippet, etc)
• Clippy diagnostics
• The if_chain macro
• from_expansion and in_external_macro
• Span
• Applicability
• The rustc-dev-guide explains a lot of internal compiler concepts
• The nightly rustc docs which has been linked to throughout this guide

For EarlyLintPass lints:

• EarlyLintPass
• rustc_ast::ast

For LateLintPass lints:

• LateLintPass
• Ty::TyKind

While most of Clippy's lint utils are documented, most of rustc's internals lack documentation currently. This is unfortunate, but in most cases you can probably get away with copying things from existing similar lints. If you are stuck, don't hesitate to ask on Discord or in the issue/PR.