|
| 1 | +# Lint passes |
| 2 | + |
| 3 | +Before working on the logic of a new lint, there is an important decision |
| 4 | +that every Clippy developers must make: to use |
| 5 | +[`EarlyLintPass`][early_lint_pass] or [`LateLintPass`][late_lint_pass]. |
| 6 | + |
| 7 | +In short, the `LateLintPass` has access to type and symbol information while the |
| 8 | +`EarlyLintPass` doesn't. If you don't need access to type information, use the |
| 9 | +`EarlyLintPass`. |
| 10 | + |
| 11 | +Let us expand on these two traits more below. |
| 12 | + |
| 13 | +## `EarlyLintPass` |
| 14 | + |
| 15 | +If you examine the documentation on [`EarlyLintPass`][early_lint_pass] closely, |
| 16 | +you'll see that every method defined for this trait utilizes a |
| 17 | +[`EarlyContext`][early_context]. In `EarlyContext`'s documentation, it states: |
| 18 | + |
| 19 | +> Context for lint checking of the AST, after expansion, before lowering to HIR. |
| 20 | +
|
| 21 | +Voilà. `EarlyLintPass` works only on the Abstract Syntax Tree (AST) level. |
| 22 | +And AST is generated during the [lexing and parsing][lexing_and_parsing] phase |
| 23 | +of code compilation. Therefore, it doesn't know what a symbol means or information about types, and it should |
| 24 | +be our trait choice for a new lint if the lint only deals with syntax-related issues. |
| 25 | + |
| 26 | +While linting speed has not been a concern for Clippy, |
| 27 | +the `EarlyLintPass` is faster, and it should be your choice |
| 28 | +if you know for sure a lint does not need type information. |
| 29 | + |
| 30 | +As a reminder, run the following command to generate boilerplate for lints |
| 31 | +that use `EarlyLintPass`: |
| 32 | + |
| 33 | +```sh |
| 34 | +$ cargo dev new_lint --name=<your_new_lint> --pass=early --category=<your_category_choice> |
| 35 | +``` |
| 36 | + |
| 37 | +### Example for `EarlyLintPass` |
| 38 | + |
| 39 | +Take a look at the following code: |
| 40 | + |
| 41 | +```rust |
| 42 | +let x = OurUndefinedType; |
| 43 | +x.non_existing_method(); |
| 44 | +``` |
| 45 | + |
| 46 | +From the AST perspective, both lines are "grammatically" correct. |
| 47 | +The assignment uses a `let` and ends with a semicolon. The invocation |
| 48 | +of a method looks fine, too. As programmers, we might raise a few |
| 49 | +questions already, but the parser is okay with it. This is what we |
| 50 | +mean when we say `EarlyLintPass` deals with only syntax on the AST level. |
| 51 | + |
| 52 | +Alternatively, think of the `foo_functions` lint we mentioned in |
| 53 | +[define new lints](define_lints.md#name-the-lint) chapter. |
| 54 | + |
| 55 | +We want the `foo_functions` lint to detect functions with `foo` as their name. |
| 56 | +Writing a lint that only checks for the name of a function means that we only |
| 57 | +work with the AST and don't have to access the type system at all (the type system is where |
| 58 | +`LateLintPass` comes into the picture). |
| 59 | + |
| 60 | +## `LateLintPass` |
| 61 | + |
| 62 | +In contrast to `EarlyLintPass`, `LateLintPass` contains type information. |
| 63 | + |
| 64 | +If you examine the documentation on [`LateLintPass`][late_lint_pass] closely, |
| 65 | +you see that every method defined in this trait utilizes a |
| 66 | +[`LateContext`][late_context]. |
| 67 | + |
| 68 | +In `LateContext`'s documentation we will find methods that |
| 69 | +deal with type-checking, which do not exist in `EarlyContext`, such as: |
| 70 | + |
| 71 | +- [`maybe_typeck_results`](https://doc.rust-lang.org/nightly/nightly-rustc/rustc_lint/context/struct.LateContext.html#method.maybe_typeck_results) |
| 72 | +- [`typeck_results`](https://doc.rust-lang.org/nightly/nightly-rustc/rustc_lint/context/struct.LateContext.html#method.typeck_results) |
| 73 | + |
| 74 | +### Example for `LateLintPass` |
| 75 | + |
| 76 | +Let us take a look with the following example: |
| 77 | + |
| 78 | +```rust |
| 79 | +let x = OurUndefinedType; |
| 80 | +x.non_existing_method(); |
| 81 | +``` |
| 82 | + |
| 83 | +These two lines of code are syntactically correct code from the perspective |
| 84 | +of the AST. We have an assignment and invoke a method on the variable that |
| 85 | +is of a type. Grammatically, everything is in order for the parser. |
| 86 | + |
| 87 | +However, going down a level and looking at the type information, |
| 88 | +the compiler will notice that both `OurUndefinedType` and `non_existing_method()` |
| 89 | +**are undefined**. |
| 90 | + |
| 91 | +As Clippy developers, to access such type information, we must implement |
| 92 | +`LateLintPass` on our lint. |
| 93 | +When you browse through Clippy's lints, you will notice that almost every lint |
| 94 | +is implemented in a `LateLintPass`, specifically because we often need to check |
| 95 | +not only for syntactic issues but also type information. |
| 96 | + |
| 97 | +Another limitation of the `EarlyLintPass` is that the nodes are only identified |
| 98 | +by their position in the AST. This means that you can't just get an `id` and |
| 99 | +request a certain node. For most lints that is fine, but we have some lints |
| 100 | +that require the inspection of other nodes, which is easier at the HIR level. |
| 101 | +In these cases, `LateLintPass` is the better choice. |
| 102 | + |
| 103 | +As a reminder, run the following command to generate boilerplate for lints |
| 104 | +that use `LateLintPass`: |
| 105 | + |
| 106 | +```sh |
| 107 | +$ cargo dev new_lint --name=<your_new_lint> --pass=late --category=<your_category_choice> |
| 108 | +``` |
| 109 | + |
| 110 | +## Additional Readings for Beginners |
| 111 | + |
| 112 | +If a dear reader of this documentation has never taken a class on compilers |
| 113 | +and interpreters, it might be confusing as to why AST level deals with only |
| 114 | +the language's syntax. And some readers might not even understand what lexing, |
| 115 | +parsing, and AST mean. |
| 116 | + |
| 117 | +This documentation serves by no means as a crash course on compilers or language design. |
| 118 | +And for details specifically related to Rust, the [Rustc Development Guide][rustc_dev_guide] |
| 119 | +is a far better choice to peruse. |
| 120 | + |
| 121 | +The [Syntax and AST][ast] chapter and the [High-Level IR][hir] chapter are |
| 122 | +great introduction to the concepts mentioned in this chapter. |
| 123 | + |
| 124 | +Some readers might also find the [introductory chapter][map_of_territory] of |
| 125 | +Robert Nystrom's _Crafting Interpreters_ a helpful overview of compiled and |
| 126 | +interpreted languages before jumping back to the Rustc guide. |
| 127 | + |
| 128 | +[ast]: https://rustc-dev-guide.rust-lang.org/syntax-intro.html |
| 129 | +[early_context]: https://doc.rust-lang.org/nightly/nightly-rustc/rustc_lint/context/struct.EarlyContext.html |
| 130 | +[early_lint_pass]: https://doc.rust-lang.org/nightly/nightly-rustc/rustc_lint/trait.EarlyLintPass.html |
| 131 | +[hir]: https://rustc-dev-guide.rust-lang.org/hir.html |
| 132 | +[late_context]: https://doc.rust-lang.org/nightly/nightly-rustc/rustc_lint/context/struct.LateContext.html |
| 133 | +[late_lint_pass]: https://doc.rust-lang.org/nightly/nightly-rustc/rustc_lint/trait.LateLintPass.html |
| 134 | +[lexing_and_parsing]: https://rustc-dev-guide.rust-lang.org/overview.html#lexing-and-parsing |
| 135 | +[rustc_dev_guide]: https://rustc-dev-guide.rust-lang.org/ |
| 136 | +[map_of_territory]: https://craftinginterpreters.com/a-map-of-the-territory.html |
0 commit comments