Clippy
A collection of lints to catch common mistakes and improve your Rust code.
There are over 650 lints included in this crate!
Lints are divided into categories, each with a default lint
level. You can choose how
much Clippy is supposed to annoy help you by changing the lint level by
category.
Category | Description | Default level |
---|---|---|
clippy::all | all lints that are on by default (correctness, suspicious, style, complexity, perf) | warn/deny |
clippy::correctness | code that is outright wrong or useless | deny |
clippy::suspicious | code that is most likely wrong or useless | warn |
clippy::complexity | code that does something simple but in a complex way | warn |
clippy::perf | code that can be written to run faster | warn |
clippy::style | code that should be written in a more idiomatic way | warn |
clippy::pedantic | lints which are rather strict or might have false positives | allow |
clippy::nursery | new lints that are still under development | allow |
clippy::cargo | lints for the cargo manifest | allow |
More to come, please file an issue if you have ideas!
The lint list also contains "restriction lints", which are for things which are usually not considered "bad", but may be useful to turn on in specific cases. These should be used very selectively, if at all.
Installation
If you're using rustup
to install and manage your Rust toolchains, Clippy is
usually already installed. In that case you can skip this chapter and go to
the Usage chapter.
Note: If you used the
minimal
profile when installing a Rust toolchain, Clippy is not automatically installed.
Using Rustup
If Clippy was not installed for a toolchain, it can be installed with
$ rustup component add clippy [--toolchain=<name>]
From Source
Take a look at the Basics chapter in the Clippy developer guide to find step-by-step instructions on how to build and install Clippy from source.
Usage
This chapter describes how to use Clippy to get the most out of it. Clippy can
be used as a cargo
subcommand or, like rustc
, directly with the
clippy-driver
binary.
Note: This chapter assumes that you have Clippy installed already. If you're not sure, take a look at the Installation chapter.
Cargo subcommand
The easiest and most common way to run Clippy is through cargo
. To do that,
just run
cargo clippy
Lint configuration
The above command will run the default set of lints, which are included in the
lint group clippy::all
. You might want to use even more lints, or you may not
agree with every Clippy lint, and for that there are ways to configure lint
levels.
Note: Clippy is meant to be used with a generous sprinkling of
#[allow(..)]
s through your code. So if you disagree with a lint, don't feel bad disabling them for parts of your code or the whole project.
Command line
You can configure lint levels on the command line by adding
-A/W/D clippy::lint_name
like this:
cargo clippy -- -Aclippy::style -Wclippy::double_neg -Dclippy::perf
For CI all warnings can be elevated to errors which will inturn fail
the build and cause Clippy to exit with a code other than 0
.
cargo clippy -- -Dwarnings
Note: Adding
-D warnings
will cause your build to fail if any warnings are found in your code. That includes warnings found by rustc (e.g.dead_code
, etc.).
For more information on configuring lint levels, see the rustc documentation.
Even more lints
Clippy has lint groups which are allow-by-default. This means, that you will have to enable the lints in those groups manually.
For a full list of all lints with their description and examples, please refer to Clippy's lint list. The two most important allow-by-default groups are described below:
clippy::pedantic
The first group is the pedantic
group. This group contains really opinionated
lints, that may have some intentional false positives in order to prevent false
negatives. So while this group is ready to be used in production, you can expect
to sprinkle multiple #[allow(..)]
s in your code. If you find any false
positives, you're still welcome to report them to us for future improvements.
FYI: Clippy uses the whole group to lint itself.
clippy::restriction
The second group is the restriction
group. This group contains lints that
"restrict" the language in some way. For example the clippy::unwrap
lint from
this group won't allow you to use .unwrap()
in your code. You may want to look
through the lints in this group and enable the ones that fit your need.
Note: You shouldn't enable the whole lint group, but cherry-pick lints from this group. Some lints in this group will even contradict other Clippy lints!
Too many lints
The most opinionated warn-by-default group of Clippy is the clippy::style
group. Some people prefer to disable this group completely and then cherry-pick
some lints they like from this group. The same is of course possible with every
other of Clippy's lint groups.
Note: We try to keep the warn-by-default groups free from false positives (FP). If you find that a lint wrongly triggers, please report it in an issue (if there isn't an issue for that FP already)
Source Code
You can configure lint levels in source code the same way you can configure
rustc
lints:
#![allow(clippy::style)]
#[warn(clippy::double_neg)]
fn main() {
let x = 1;
let y = --x;
// ^^ warning: double negation
}
Automatically applying Clippy suggestions
Clippy can automatically apply some lint suggestions, just like the compiler. Note that --fix
implies
--all-targets
, so it can fix as much code as it can.
cargo clippy --fix
Workspaces
All the usual workspace options should work with Clippy. For example the
following command will run Clippy on the example
crate in your workspace:
cargo clippy -p example
As with cargo check
, this includes dependencies that are members of the
workspace, like path dependencies. If you want to run Clippy only on the
given crate, use the --no-deps
option like this:
cargo clippy -p example -- --no-deps
Using Clippy without cargo
: clippy-driver
Clippy can also be used in projects that do not use cargo. To do so, run
clippy-driver
with the same arguments you use for rustc
. For example:
clippy-driver --edition 2018 -Cpanic=abort foo.rs
Note:
clippy-driver
is designed for running Clippy and should not be used as a general replacement forrustc
.clippy-driver
may produce artifacts that are not optimized as expected, for example.
Configuring Clippy
Note: The configuration file is unstable and may be deprecated in the future.
Some lints can be configured in a TOML file named clippy.toml
or .clippy.toml
, which is searched for in:
- The directory specified by the
CLIPPY_CONF_DIR
environment variable, or - The directory specified by the CARGO_MANIFEST_DIR environment variable, or
- The current directory.
It contains a basic variable = value
mapping e.g.
avoid-breaking-exported-api = false
disallowed-names = ["toto", "tata", "titi"]
The table of configurations contains all config values, their default, and a list of lints they affect. Each configurable lint , also contains information about these values.
For configurations that are a list type with default values such as
disallowed-names,
you can use the unique value ".."
to extend the default values instead of replacing them.
# default of disallowed-names is ["foo", "baz", "quux"]
disallowed-names = ["bar", ".."] # -> ["bar", "foo", "baz", "quux"]
To deactivate the "for further information visit lint-link" message you can define the CLIPPY_DISABLE_DOCS_LINKS
environment variable.
Allowing/denying lints
You can add options to your code to allow
/warn
/deny
Clippy lints:
-
the whole set of
Warn
lints using theclippy
lint group (#![deny(clippy::all)]
) -
all lints using both the
clippy
andclippy::pedantic
lint groups (#![deny(clippy::all)]
,#![deny(clippy::pedantic)]
). Note thatclippy::pedantic
contains some very aggressive lints prone to false positives. -
only some lints (
#![deny(clippy::single_match, clippy::box_vec)]
, etc.) -
allow
/warn
/deny
can be limited to a single function or module using#[allow(...)]
, etc.
Note: allow
means to suppress the lint for your code. With warn
the lint will only emit a warning, while with deny
the lint will emit an error, when triggering for your code. An error causes clippy to exit with an error code, so is
useful in scripts like CI/CD.
If you do not want to include your lint levels in your code, you can globally enable/disable lints by passing extra flags to Clippy during the run:
To allow lint_name
, run
cargo clippy -- -A clippy::lint_name
And to warn on lint_name
, run
cargo clippy -- -W clippy::lint_name
This also works with lint groups. For example, you can run Clippy with warnings for all lints enabled:
cargo clippy -- -W clippy::pedantic
If you care only about a single lint, you can allow all others and then explicitly warn on the lint(s) you are interested in:
cargo clippy -- -A clippy::all -W clippy::useless_format -W clippy::...
Specifying the minimum supported Rust version
Projects that intend to support old versions of Rust can disable lints pertaining to newer features by specifying the minimum supported Rust version (MSRV) in the clippy configuration file.
msrv = "1.30.0"
The MSRV can also be specified as an attribute, like below.
#![feature(custom_inner_attributes)]
#![clippy::msrv = "1.30.0"]
fn main() {
...
}
You can also omit the patch version when specifying the MSRV, so msrv = 1.30
is equivalent to msrv = 1.30.0
.
Note: custom_inner_attributes
is an unstable feature, so it has to be enabled explicitly.
Lints that recognize this configuration option can be found here
Disabling evaluation of certain code
Note: This should only be used in cases where other solutions, like
#[allow(clippy::all)]
, are not sufficient.
Very rarely, you may wish to prevent Clippy from evaluating certain sections of code entirely. You can do this with
conditional compilation by checking that the
cargo-clippy
feature is not set. You may need to provide a stub so that the code compiles:
#![allow(unused)] fn main() { #[cfg(not(feature = "cargo-clippy"))] include!(concat!(env!("OUT_DIR"), "/my_big_function-generated.rs")); #[cfg(feature = "cargo-clippy")] fn my_big_function(_input: &str) -> Option<MyStruct> { None } }
This feature is not actually part of your crate, so specifying --all-features
to other tools, e.g. cargo test --all-features
, will not disable it.
Lint Configuration Options
The following list shows each configuration option, along with a description, its default value, an example and lints affected.
arithmetic-side-effects-allowed
Suppress checking of the passed type names in all types of operations.
If a specific operation is desired, consider using arithmetic_side_effects_allowed_binary
or arithmetic_side_effects_allowed_unary
instead.
Example
arithmetic-side-effects-allowed = ["SomeType", "AnotherType"]
Noteworthy
A type, say SomeType
, listed in this configuration has the same behavior of
["SomeType" , "*"], ["*", "SomeType"]
in arithmetic_side_effects_allowed_binary
.
Default Value: []
Affected lints:
arithmetic-side-effects-allowed-binary
Suppress checking of the passed type pair names in binary operations like addition or multiplication.
Supports the "*" wildcard to indicate that a certain type won't trigger the lint regardless
of the involved counterpart. For example, ["SomeType", "*"]
or ["*", "AnotherType"]
.
Pairs are asymmetric, which means that ["SomeType", "AnotherType"]
is not the same as
["AnotherType", "SomeType"]
.
Example
arithmetic-side-effects-allowed-binary = [["SomeType" , "f32"], ["AnotherType", "*"]]
Default Value: []
Affected lints:
arithmetic-side-effects-allowed-unary
Suppress checking of the passed type names in unary operations like "negation" (-
).
Example
arithmetic-side-effects-allowed-unary = ["SomeType", "AnotherType"]
Default Value: []
Affected lints:
avoid-breaking-exported-api
Suppress lints whenever the suggested change would cause breakage for other crates.
Default Value: true
Affected lints:
enum_variant_names
large_types_passed_by_value
trivially_copy_pass_by_ref
unnecessary_wraps
unused_self
upper_case_acronyms
wrong_self_convention
box_collection
redundant_allocation
rc_buffer
vec_box
option_option
linkedlist
rc_mutex
unnecessary_box_returns
single_call_fn
msrv
The minimum rust version that the project supports. Defaults to the rust-version
field in Cargo.toml
Affected lints:
manual_split_once
manual_str_repeat
cloned_instead_of_copied
redundant_field_names
option_map_unwrap_or
redundant_static_lifetimes
filter_map_next
checked_conversions
manual_range_contains
use_self
mem_replace_with_default
manual_non_exhaustive
option_as_ref_deref
map_unwrap_or
match_like_matches_macro
manual_strip
missing_const_for_fn
unnested_or_patterns
from_over_into
ptr_as_ptr
if_then_some_else_none
approx_constant
deprecated_cfg_attr
index_refutable_slice
map_clone
borrow_as_ptr
manual_bits
err_expect
cast_abs_to_unsigned
uninlined_format_args
manual_clamp
manual_let_else
unchecked_duration_subtraction
collapsible_str_replace
seek_from_current
seek_rewind
unnecessary_lazy_evaluations
transmute_ptr_to_ref
almost_complete_range
needless_borrow
derivable_impls
manual_is_ascii_check
manual_rem_euclid
manual_retain
type_repetition_in_bounds
tuple_array_conversions
manual_try_fold
manual_hash_one
cognitive-complexity-threshold
The maximum cognitive complexity a function can have
Default Value: 25
Affected lints:
excessive-nesting-threshold
The maximum amount of nesting a block can reside in
Default Value: 0
Affected lints:
disallowed-names
The list of disallowed names to lint about. NB: bar
is not here since it has legitimate uses. The value
".."
can be used as part of the list to indicate that the configured values should be appended to the
default configuration of Clippy. By default, any configuration will replace the default value.
Default Value: ["foo", "baz", "quux"]
Affected lints:
semicolon-inside-block-ignore-singleline
Whether to lint only if it's multiline.
Default Value: false
Affected lints:
semicolon-outside-block-ignore-multiline
Whether to lint only if it's singleline.
Default Value: false
Affected lints:
doc-valid-idents
The list of words this lint should not consider as identifiers needing ticks. The value
".."
can be used as part of the list to indicate, that the configured values should be appended to the
default configuration of Clippy. By default, any configuration will replace the default value. For example:
doc-valid-idents = ["ClipPy"]
would replace the default list with["ClipPy"]
.doc-valid-idents = ["ClipPy", ".."]
would appendClipPy
to the default list.
Default Value: ["KiB", "MiB", "GiB", "TiB", "PiB", "EiB", "DirectX", "ECMAScript", "GPLv2", "GPLv3", "GitHub", "GitLab", "IPv4", "IPv6", "ClojureScript", "CoffeeScript", "JavaScript", "PureScript", "TypeScript", "WebAssembly", "NaN", "NaNs", "OAuth", "GraphQL", "OCaml", "OpenGL", "OpenMP", "OpenSSH", "OpenSSL", "OpenStreetMap", "OpenDNS", "WebGL", "TensorFlow", "TrueType", "iOS", "macOS", "FreeBSD", "TeX", "LaTeX", "BibTeX", "BibLaTeX", "MinGW", "CamelCase"]
Affected lints:
too-many-arguments-threshold
The maximum number of argument a function or method can have
Default Value: 7
Affected lints:
type-complexity-threshold
The maximum complexity a type can have
Default Value: 250
Affected lints:
single-char-binding-names-threshold
The maximum number of single char bindings a scope may have
Default Value: 4
Affected lints:
too-large-for-stack
The maximum size of objects (in bytes) that will be linted. Larger objects are ok on the heap
Default Value: 200
Affected lints:
enum-variant-name-threshold
The minimum number of enum variants for the lints about variant names to trigger
Default Value: 3
Affected lints:
struct-field-name-threshold
The minimum number of struct fields for the lints about field names to trigger
Default Value: 3
Affected lints:
enum-variant-size-threshold
The maximum size of an enum's variant to avoid box suggestion
Default Value: 200
Affected lints:
verbose-bit-mask-threshold
The maximum allowed size of a bit mask before suggesting to use 'trailing_zeros'
Default Value: 1
Affected lints:
literal-representation-threshold
The lower bound for linting decimal literals
Default Value: 16384
Affected lints:
trivial-copy-size-limit
The maximum size (in bytes) to consider a Copy
type for passing by value instead of by
reference. By default there is no limit
Affected lints:
pass-by-value-size-limit
The minimum size (in bytes) to consider a type for passing by reference instead of by value.
Default Value: 256
Affected lints:
too-many-lines-threshold
The maximum number of lines a function or method can have
Default Value: 100
Affected lints:
array-size-threshold
The maximum allowed size for arrays on the stack
Default Value: 512000
Affected lints:
stack-size-threshold
The maximum allowed stack size for functions in bytes
Default Value: 512000
Affected lints:
vec-box-size-threshold
The size of the boxed type in bytes, where boxing in a Vec
is allowed
Default Value: 4096
Affected lints:
max-trait-bounds
The maximum number of bounds a trait can have to be linted
Default Value: 3
Affected lints:
max-struct-bools
The maximum number of bool fields a struct can have
Default Value: 3
Affected lints:
max-fn-params-bools
The maximum number of bool parameters a function can have
Default Value: 3
Affected lints:
warn-on-all-wildcard-imports
Whether to allow certain wildcard imports (prelude, super in tests).
Default Value: false
Affected lints:
disallowed-macros
The list of disallowed macros, written as fully qualified paths.
Default Value: []
Affected lints:
disallowed-methods
The list of disallowed methods, written as fully qualified paths.
Default Value: []
Affected lints:
disallowed-types
The list of disallowed types, written as fully qualified paths.
Default Value: []
Affected lints:
unreadable-literal-lint-fractions
Should the fraction of a decimal be linted to include separators.
Default Value: true
Affected lints:
upper-case-acronyms-aggressive
Enables verbose mode. Triggers if there is more than one uppercase char next to each other
Default Value: false
Affected lints:
matches-for-let-else
Whether the matches should be considered by the lint, and whether there should be filtering for common types.
Default Value: "WellKnownTypes"
Affected lints:
cargo-ignore-publish
For internal testing only, ignores the current publish
settings in the Cargo manifest.
Default Value: false
Affected lints:
standard-macro-braces
Enforce the named macros always use the braces specified.
A MacroMatcher
can be added like so { name = "macro_name", brace = "(" }
. If the macro
could be used with a full path two MacroMatcher
s have to be added one with the full path
crate_name::macro_name
and one with just the macro name.
Default Value: []
Affected lints:
enforced-import-renames
The list of imports to always rename, a fully qualified path followed by the rename.
Default Value: []
Affected lints:
allowed-scripts
The list of unicode scripts allowed to be used in the scope.
Default Value: ["Latin"]
Affected lints:
enable-raw-pointer-heuristic-for-send
Whether to apply the raw pointer heuristic to determine if a type is Send
.
Default Value: true
Affected lints:
max-suggested-slice-pattern-length
When Clippy suggests using a slice pattern, this is the maximum number of elements allowed in
the slice pattern that is suggested. If more elements are necessary, the lint is suppressed.
For example, [_, _, _, e, ..]
is a slice pattern with 4 elements.
Default Value: 3
Affected lints:
await-holding-invalid-types
Default Value: []
Affected lints:
max-include-file-size
The maximum size of a file included via include_bytes!()
or include_str!()
, in bytes
Default Value: 1000000
Affected lints:
allow-expect-in-tests
Whether expect
should be allowed in test functions or #[cfg(test)]
Default Value: false
Affected lints:
allow-unwrap-in-tests
Whether unwrap
should be allowed in test functions or #[cfg(test)]
Default Value: false
Affected lints:
allow-dbg-in-tests
Whether dbg!
should be allowed in test functions or #[cfg(test)]
Default Value: false
Affected lints:
allow-print-in-tests
Whether print macros (ex. println!
) should be allowed in test functions or #[cfg(test)]
Default Value: false
Affected lints:
large-error-threshold
The maximum size of the Err
-variant in a Result
returned from a function
Default Value: 128
Affected lints:
ignore-interior-mutability
A list of paths to types that should be treated like Arc
, i.e. ignored but
for the generic parameters for determining interior mutability
Default Value: ["bytes::Bytes"]
Affected lints:
allow-mixed-uninlined-format-args
Whether to allow mixed uninlined format args, e.g. format!("{} {}", a, foo.bar)
Default Value: true
Affected lints:
suppress-restriction-lint-in-const
Whether to suppress a restriction lint in constant code. In same cases the restructured operation might not be unavoidable, as the suggested counterparts are unavailable in constant code. This configuration will cause restriction lints to trigger even if no suggestion can be made.
Default Value: false
Affected lints:
missing-docs-in-crate-items
Whether to only check for missing documentation in items visible within the current
crate. For example, pub(crate)
items.
Default Value: false
Affected lints:
future-size-threshold
The maximum byte size a Future
can have, before it triggers the clippy::large_futures
lint
Default Value: 16384
Affected lints:
unnecessary-box-size
The byte size a T
in Box<T>
can have, below which it triggers the clippy::unnecessary_box
lint
Default Value: 128
Affected lints:
allow-private-module-inception
Whether to allow module inception if it's not public.
Default Value: false
Affected lints:
allowed-idents-below-min-chars
Allowed names below the minimum allowed characters. The value ".."
can be used as part of
the list to indicate, that the configured values should be appended to the default
configuration of Clippy. By default, any configuration will replace the default value.
Default Value: ["j", "z", "i", "y", "n", "x", "w"]
Affected lints:
min-ident-chars-threshold
Minimum chars an ident can have, anything below or equal to this will be linted.
Default Value: 1
Affected lints:
accept-comment-above-statement
Whether to accept a safety comment to be placed above the statement containing the unsafe
block
Default Value: true
Affected lints:
accept-comment-above-attributes
Whether to accept a safety comment to be placed above the attributes for the unsafe
block
Default Value: true
Affected lints:
allow-one-hash-in-raw-strings
Whether to allow r#""#
when r""
can be used
Default Value: false
Affected lints:
absolute-paths-max-segments
The maximum number of segments a path can have before being linted, anything above this will be linted.
Default Value: 2
Affected lints:
absolute-paths-allowed-crates
Which crates to allow absolute paths from
Default Value: []
Affected lints:
allowed-dotfiles
Additional dotfiles (files or directories starting with a dot) to allow
Default Value: []
Affected lints:
enforce-iter-loop-reborrow
Example
let mut vec = vec![1, 2, 3];
let rmvec = &mut vec;
for _ in rmvec.iter() {}
for _ in rmvec.iter_mut() {}
Use instead:
let mut vec = vec![1, 2, 3];
let rmvec = &mut vec;
for _ in &*rmvec {}
for _ in &mut *rmvec {}
Default Value: false
Affected lints:
Clippy's Lints
Clippy offers a bunch of additional lints, to help its users write more correct and idiomatic Rust code. A full list of all lints, that can be filtered by category, lint level or keywords, can be found in the Clippy lint documentation.
This chapter will give an overview of the different lint categories, which kind of lints they offer and recommended actions when you should see a lint out of that category. For examples, see the Clippy lint documentation and filter by category.
The different lint groups were defined in the Clippy 1.0 RFC.
Correctness
The clippy::correctness
group is the only lint group in Clippy which lints are
deny-by-default and abort the compilation when triggered. This is for good
reason: If you see a correctness
lint, it means that your code is outright
wrong or useless, and you should try to fix it.
Lints in this category are carefully picked and should be free of false
positives. So just #[allow]
ing those lints is not recommended.
Suspicious
The clippy::suspicious
group is similar to the correctness lints in that it
contains lints that trigger on code that is really sus and should be fixed. As
opposed to correctness lints, it might be possible that the linted code is
intentionally written like it is.
It is still recommended to fix code that is linted by lints out of this group
instead of #[allow]
ing the lint. In case you intentionally have written code
that offends the lint you should specifically and locally #[allow]
the lint
and add give a reason why the code is correct as written.
Complexity
The clippy::complexity
group offers lints that give you suggestions on how to
simplify your code. It mostly focuses on code that can be written in a shorter
and more readable way, while preserving the semantics.
If you should see a complexity lint, it usually means that you can remove or replace some code, and it is recommended to do so. However, if you need the more complex code for some expressiveness reason, it is recommended to allow complexity lints on a case-by-case basis.
Perf
The clippy::perf
group gives you suggestions on how you can increase the
performance of your code. Those lints are mostly about code that the compiler
can't trivially optimize, but has to be written in a slightly different way to
make the optimizer job easier.
Perf lints are usually easy to apply, and it is recommended to do so.
Style
The clippy::style
group is mostly about writing idiomatic code. Because style
is subjective, this lint group is the most opinionated warn-by-default group in
Clippy.
If you see a style lint, applying the suggestion usually makes your code more
readable and idiomatic. But because we know that this is opinionated, feel free
to sprinkle #[allow]
s for style lints in your code or #![allow]
a style lint
on your whole crate if you disagree with the suggested style completely.
Pedantic
The clippy::pedantic
group makes Clippy even more pedantic. You can enable
the whole group with #![warn(clippy::pedantic)]
in the lib.rs
/main.rs
of
your crate. This lint group is for Clippy power users that want an in depth
check of their code.
Note: Instead of enabling the whole group (like Clippy itself does), you may want to cherry-pick lints out of the pedantic group.
If you enable this group, expect to also use #[allow]
attributes generously
throughout your code. Lints in this group are designed to be pedantic and false
positives sometimes are intentional in order to prevent false negatives.
Restriction
The clippy::restriction
group contains lints that will restrict you from
using certain parts of the Rust language. It is not recommended to enable
the whole group, but rather cherry-pick lints that are useful for your code base
and your use case.
Note: Clippy will produce a warning if it finds a
#![warn(clippy::restriction)]
attribute in your code!
Lints from this group will restrict you in some way. If you enable a restriction
lint for your crate it is recommended to also fix code that this lint triggers
on. However, those lints are really strict by design, and you might want to
#[allow]
them in some special cases, with a comment justifying that.
Cargo
The clippy::cargo
group gives you suggestions on how to improve your
Cargo.toml
file. This might be especially interesting if you want to publish
your crate and are not sure if you have all useful information in your
Cargo.toml
.
Continuous Integration
It is recommended to run Clippy on CI with -Dwarnings
, so that Clippy lints
prevent CI from passing. To enforce errors on warnings on all cargo
commands
not just cargo clippy
, you can set the env var RUSTFLAGS="-Dwarnings"
.
We recommend to use Clippy from the same toolchain, that you use for compiling
your crate for maximum compatibility. E.g. if your crate is compiled with the
stable
toolchain, you should also use stable
Clippy.
Note: New Clippy lints are first added to the
nightly
toolchain. If you want to help with improving Clippy and have CI resources left, please consider adding anightly
Clippy check to your CI and report problems like false positives back to us. With that we can fix bugs early, before they can get to stable.
This chapter will give an overview on how to use Clippy on different popular CI providers.
GitHub Actions
GitHub hosted runners using the latest stable version of Rust have Clippy pre-installed.
It is as simple as running cargo clippy
to run lints against the codebase.
on: push
name: Clippy check
# Make sure CI fails on all warnings, including Clippy lints
env:
RUSTFLAGS: "-Dwarnings"
jobs:
clippy_check:
runs-on: ubuntu-latest
steps:
- uses: actions/checkout@v3
- name: Run Clippy
run: cargo clippy --all-targets --all-features
Travis CI
You can add Clippy to Travis CI in the same way you use it locally:
language: rust
rust:
- stable
- beta
before_script:
- rustup component add clippy
script:
- cargo clippy
# if you want the build job to fail when encountering warnings, use
- cargo clippy -- -D warnings
# in order to also check tests and non-default crate features, use
- cargo clippy --all-targets --all-features -- -D warnings
- cargo test
# etc.
Clippy Development
Hello fellow Rustacean! If you made it here, you're probably interested in making Clippy better by contributing to it. In that case, welcome to the project!
Note: If you're just interested in using Clippy, there's nothing to see from this point onward, and you should return to one of the earlier chapters.
Getting started
If this is your first time contributing to Clippy, you should first read the Basics docs. This will explain the basics on how to get the source code and how to compile and test the code.
Additional Readings for Beginners
If a dear reader of this documentation has never taken a class on compilers and interpreters, it might be confusing as to why AST level deals with only the language's syntax. And some readers might not even understand what lexing, parsing, and AST mean.
This documentation serves by no means as a crash course on compilers or language design. And for details specifically related to Rust, the Rustc Development Guide is a far better choice to peruse.
The Syntax and AST chapter and the High-Level IR chapter are great introduction to the concepts mentioned in this chapter.
Some readers might also find the introductory chapter of Robert Nystrom's Crafting Interpreters a helpful overview of compiled and interpreted languages before jumping back to the Rustc guide.
Writing code
If you have done the basic setup, it's time to start hacking.
The Adding lints chapter is a walk through on how to add a new lint to Clippy. This is also interesting if you just want to fix a lint, because it also covers how to test lints and gives an overview of the bigger picture.
If you want to add a new lint or change existing ones apart from bugfixing, it's also a good idea to give the stability guarantees and lint categories sections of the Clippy 1.0 RFC a quick read. The lint categories are also described earlier in this book.
Note: Some higher level things about contributing to Clippy are still covered in the
CONTRIBUTING.md
document. Some of those will be moved to the book over time, like:
- Finding something to fix
- IDE setup
- High level overview on how Clippy works
- Triage procedure
- Bors and Homu
Basics for hacking on Clippy
This document explains the basics for hacking on Clippy. Besides others, this includes how to build and test Clippy. For a more in depth description on the codebase take a look at Adding Lints or Common Tools.
Get the Code
First, make sure you have checked out the latest version of Clippy. If this is your first time working on Clippy, create a fork of the repository and clone it afterwards with the following command:
git clone git@github.com:<your-username>/rust-clippy
If you've already cloned Clippy in the past, update it to the latest version:
# If the upstream remote has not been added yet
git remote add upstream https://github.com/rust-lang/rust-clippy
# upstream has to be the remote of the rust-lang/rust-clippy repo
git fetch upstream
# make sure that you are on the master branch
git checkout master
# rebase your master branch on the upstream master
git rebase upstream/master
# push to the master branch of your fork
git push
Building and Testing
You can build and test Clippy like every other Rust project:
cargo build # builds Clippy
cargo test # tests Clippy
Since Clippy's test suite is pretty big, there are some commands that only run a subset of Clippy's tests:
# only run UI tests
cargo uitest
# only run UI tests starting with `test_`
TESTNAME="test_" cargo uitest
# only run dogfood tests
cargo dev dogfood
If the output of a UI test differs from the expected output, you can update the reference file with:
cargo bless
For example, this is necessary if you fix a typo in an error message of a lint, or if you modify a test file to add a test case.
Note: This command may update more files than you intended. In that case only commit the files you wanted to update.
cargo dev
Clippy has some dev tools to make working on Clippy more convenient. These tools
can be accessed through the cargo dev
command. Available tools are listed
below. To get more information about these commands, just call them with
--help
.
# formats the whole Clippy codebase and all tests
cargo dev fmt
# register or update lint names/groups/...
cargo dev update_lints
# create a new lint and register it
cargo dev new_lint
# deprecate a lint and attempt to remove code relating to it
cargo dev deprecate
# automatically formatting all code before each commit
cargo dev setup git-hook
# (experimental) Setup Clippy to work with IntelliJ-Rust
cargo dev setup intellij
# runs the `dogfood` tests
cargo dev dogfood
More about intellij command usage and reasons.
lintcheck
cargo lintcheck
will build and run clippy on a fixed set of crates and
generate a log of the results. You can git diff
the updated log against its
previous version and see what impact your lint made on a small set of crates.
If you add a new lint, please audit the resulting warnings and make sure there
are no false positives and that the suggestions are valid.
Refer to the tools README for more details.
PR
We follow a rustc no merge-commit policy. See https://rustc-dev-guide.rust-lang.org/contributing.html#opening-a-pr.
Common Abbreviations
Abbreviation | Meaning |
---|---|
UB | Undefined Behavior |
FP | False Positive |
FN | False Negative |
ICE | Internal Compiler Error |
AST | Abstract Syntax Tree |
MIR | Mid-Level Intermediate Representation |
HIR | High-Level Intermediate Representation |
TCX | Type context |
This is a concise list of abbreviations that can come up during Clippy development. An extensive general list can be found in the rustc-dev-guide glossary. Always feel free to ask if an abbreviation or meaning is unclear to you.
Install from source
If you are hacking on Clippy and want to install it from source, do the following:
First, take note of the toolchain
override in
/rust-toolchain
. We will use this override to install Clippy into the right
toolchain.
Tip: You can view the active toolchain for the current directory with
rustup show active-toolchain
.
From the Clippy project root, run the following command to build the Clippy binaries and copy them into the toolchain directory. This will override the currently installed Clippy component.
cargo build --release --bin cargo-clippy --bin clippy-driver -Zunstable-options --out-dir "$(rustc --print=sysroot)/bin"
Now you may run cargo clippy
in any project, using the toolchain where you
just installed Clippy.
cd my-project
cargo +nightly-2021-07-01 clippy
...or clippy-driver
clippy-driver +nightly-2021-07-01 <filename>
If you need to restore the default Clippy installation, run the following (from the Clippy project root).
rustup component remove clippy
rustup component add clippy
DO NOT install using
cargo install --path . --force
since this will overwrite rustup proxies. That is,~/.cargo/bin/cargo-clippy
and~/.cargo/bin/clippy-driver
should be hard or soft links to~/.cargo/bin/rustup
. You can repair these by runningrustup update
.
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
- Testing manually
- Running directly
- Lint declaration
- Lint registration
- Lint passes
- Emitting a lint
- Adding the lint logic
- Specifying the lint's minimum supported Rust version (MSRV)
- Author lint
- Print HIR lint
- Documentation
- Running rustfmt
- Debugging
- Conflicting lints
- PR Checklist
- Adding configuration to a lint
- Cheat Sheet
Setup
See the Basics documentation.
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). If you're unsure if the name you chose fits the lint,
take a look at our lint naming guidelines.
Defining Our Lint
To get started, there are two ways to define our lint.
Standalone
Command: cargo dev new_lint --name=foo_functions --pass=early --category=pedantic
(category will default to nursery if not provided)
This command will create a new file: clippy_lints/src/foo_functions.rs
, as well
as register the lint.
Specific Type
Command: cargo dev new_lint --name=foo_functions --type=functions --category=pedantic
This command will create a new file: clippy_lints/src/{type}/foo_functions.rs
.
Notice how this command has a --type
flag instead of --pass
. Unlike a standalone
definition, this lint won't be registered in the traditional sense. Instead, you will
call your lint from within the type's lint pass, found in clippy_lints/src/{type}/mod.rs
.
A "type" is just the name of a directory in clippy_lints/src
, like functions
in
the example command. These are groupings of lints with common behaviors, so if your
lint falls into one, it would be best to add it to that type.
Tests Location
Both commands will create a file: tests/ui/foo_functions.rs
. For cargo lints,
two project hierarchies (fail/pass) will be created by default under tests/ui-cargo
.
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:
#![allow(unused)] #![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 uibless
, 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 by checking the
.stderr
file that gets updated on every test run.
Running TESTNAME=foo_functions cargo uitest
should pass on its own. When we
commit our lint, we need to commit the generated .stderr
files, too. In
general, you should only commit files changed by cargo bless
for the
specific lint you are creating/editing.
Note: you can run multiple test files by specifying a comma separated list:
TESTNAME=foo_functions,test2,test3
.
Cargo lints
For cargo lints, the process of testing differs in that we are interested in the
Cargo.toml
manifest file. We also need a minimal crate associated with that
manifest.
If our new lint is named e.g. foo_categories
, after running cargo dev new_lint --name=foo_categories --type=cargo --category=cargo
we will find by
default two new crates, each with its manifest file:
tests/ui-cargo/foo_categories/fail/Cargo.toml
: this file should cause the new lint to raise an error.tests/ui-cargo/foo_categories/pass/Cargo.toml
: this file should not trigger the lint.
If you need more cases, you can copy one of those crates (under
foo_categories
) and rename it.
The process of generating the .stderr
file is the same, and prepending the
TESTNAME
variable to cargo uitest
works too.
Rustfix tests
If the lint you are working on is making use of structured suggestions, the test
will create a .fixed
file by running 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 cargo bless
to automatically generate the .fixed
file while running
the tests.
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
cargo dev lint input.rs
from the working copy root. With tests in place, let's have a look at implementing our lint now.
Running directly
While it's easier to just use cargo dev lint
, it might be desirable to get
target/release/cargo-clippy
and target/release/clippy-driver
to work as well in some cases.
By default, they don't work because clippy dynamically links rustc. To help them find rustc,
add the path printed byrustc --print target-libdir
(ran inside this workspace so that the rustc version matches)
to your library search path.
On linux, this can be done by setting the LD_LIBRARY_PATH
environment variable to that path.
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:
#![allow(unused)] fn main() { 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:
#![allow(unused)] fn main() { declare_clippy_lint! { /// ### What it does /// /// ### Why is this bad? /// /// ### Example /// ```rust /// // example code /// ``` #[clippy::version = "1.29.0"] 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. To render and open this documentation locally in a browser, runcargo dev serve
. - The
#[clippy::version]
attribute will be rendered as part of the lint documentation. The value should be set to the current Rust version that the lint is developed in, it can be retrieved by runningrustc -vV
in the rust-clippy directory. The version is listed under release. (Use the version without the-nightly
) suffix. 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 withallow
/warn
/deny
.pedantic
sets the lint level toAllow
. 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:
#![allow(unused)] fn main() { // clippy_lints/src/foo_functions.rs // .. imports and lint declaration .. declare_lint_pass!(FooFunctions => [FOO_FUNCTIONS]); impl EarlyLintPass for FooFunctions {} }
Lint registration
When using cargo dev new_lint
, the lint is automatically registered and
nothing more has to be done.
When declaring a new lint by hand and cargo dev update_lints
is used, the lint
pass may have to be registered manually in the register_lints
function in
clippy_lints/src/lib.rs
:
store.register_early_pass(|| Box::new(foo_functions::FooFunctions));
As one may expect, there is a corresponding register_late_pass
method
available as well. Without a call to one of register_early_pass
or
register_late_pass
, the lint pass in question will not be run.
One reason that cargo dev update_lints
does not automate this step is that
multiple lints can use the same lint pass, so registering the lint pass may
already be done when adding a new lint. Another reason that this step is not
automated is that the order that the passes are registered determines the order
the passes actually run, which in turn affects the order that any emitted lints
are output in.
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_utils/src/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`",
None,
"consider using a more meaningful name"
);
}
}
Running our UI test should now produce output that contains the lint message.
According to the rustc-dev-guide, the text should be matter of fact and avoid capitalization and periods, unless multiple sentences are needed. When code or an identifier must appear in a message or label, it should be surrounded with single grave accents `.
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:
#![allow(unused)] fn main() { 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`", None, "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:
#![allow(unused)] fn main() { // 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 cargo bless
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.
Specifying the lint's minimum supported Rust version (MSRV)
Sometimes a lint makes suggestions that require a certain version of Rust. For
example, the manual_strip
lint suggests using str::strip_prefix
and
str::strip_suffix
which is only available after Rust 1.45. In such cases, you
need to ensure that the MSRV configured for the project is >= the MSRV of the
required Rust feature. If multiple features are required, just use the one with
a lower MSRV.
First, add an MSRV alias for the required feature in clippy_config::msrvs
.
This can be accessed later as msrvs::STR_STRIP_PREFIX
, for example.
#![allow(unused)] fn main() { msrv_aliases! { .. 1,45,0 { STR_STRIP_PREFIX } } }
In order to access the project-configured MSRV, you need to have an msrv
field
in the LintPass struct, and a constructor to initialize the field. The msrv
value is passed to the constructor in clippy_lints/lib.rs
.
#![allow(unused)] fn main() { pub struct ManualStrip { msrv: Msrv, } impl ManualStrip { #[must_use] pub fn new(msrv: Msrv) -> Self { Self { msrv } } } }
The project's MSRV can then be matched against the feature MSRV in the LintPass
using the Msrv::meets
method.
#![allow(unused)] fn main() { if !self.msrv.meets(msrvs::STR_STRIP_PREFIX) { return; } }
The project's MSRV can also be specified as an attribute, which overrides
the value from clippy.toml
. This can be accounted for using the
extract_msrv_attr!(LintContext)
macro and passing
LateContext
/EarlyContext
.
impl<'tcx> LateLintPass<'tcx> for ManualStrip {
fn check_expr(&mut self, cx: &LateContext<'tcx>, expr: &'tcx Expr<'_>) {
...
}
extract_msrv_attr!(LateContext);
}
Once the msrv
is added to the lint, a relevant test case should be added to
the lint's test file, tests/ui/manual_strip.rs
in this example. It should
have a case for the version below the MSRV and one with the same contents but
for the MSRV version itself.
...
#[clippy::msrv = "1.44"]
fn msrv_1_44() {
/* something that would trigger the lint */
}
#[clippy::msrv = "1.45"]
fn msrv_1_45() {
/* something that would trigger the lint */
}
As a last step, the lint should be added to the lint documentation. This is done
in clippy_config/src/conf.rs
:
#![allow(unused)] fn main() { define_Conf! { /// Lint: LIST, OF, LINTS, <THE_NEWLY_ADDED_LINT>. The minimum rust version that the project supports (msrv: Option<String> = None), ... } }
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 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.
Print HIR lint
To implement a lint, it's helpful to first understand the internal
representation that rustc uses. Clippy has the #[clippy::dump]
attribute that
prints the High-Level Intermediate Representation (HIR) of the item,
statement, or expression that the attribute is attached to. To attach the
attribute to expressions you often need to enable
#![feature(stmt_expr_attributes)]
.
Here you can find an example, just select Tools and run Clippy.
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:
#![allow(unused)] fn main() { declare_clippy_lint! { /// ### What it does /// Checks for ... (describe what the lint matches). /// /// ### Why is this bad? /// Supply the reason for linting the code. /// /// ### Example /// /// ```rust,ignore /// // A short example of code that triggers the lint /// ``` /// /// Use instead: /// ```rust,ignore /// // A short example of improved code that doesn't trigger the lint /// ``` #[clippy::version = "1.29.0"] 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.
Conflicting lints
There are several lints that deal with the same pattern but suggest different approaches. In other words, some lints may suggest modifications that go in the opposite direction to what some other lints already propose for the same code, creating conflicting diagnostics.
When you are creating a lint that ends up in this scenario, the following tips should be encouraged to guide classification:
- The only case where they should be in the same category is if that category is
restriction
. For example,semicolon_inside_block
andsemicolon_outside_block
. - For all the other cases, they should be in different categories with different levels of allowance. For example,
implicit_return
(restriction, allow) andneedless_return
(style, warn).
For lints that are in different categories, it is also recommended that at least one of them should be in the
restriction
category. The reason for this is that the restriction
group is the only group where we don't
recommend to enable the entire set, but cherry pick lints out of.
PR Checklist
Before submitting your PR make sure you followed all 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
Adding configuration to a lint
Clippy supports the configuration of lints values using a clippy.toml
file which is searched for in:
- The directory specified by the
CLIPPY_CONF_DIR
environment variable, or - The directory specified by the CARGO_MANIFEST_DIR environment variable, or
- The current directory.
Adding a configuration to a lint can be useful for thresholds or to constrain some behavior that can be seen as a false positive for some users. Adding a configuration is done in the following steps:
-
Adding a new configuration entry to
clippy_config::conf
like this:/// Lint: LINT_NAME. /// /// <The configuration field doc comment> (configuration_ident: Type = DefaultValue),
The doc comment is automatically added to the documentation of the listed lints. The default value will be formatted using the
Debug
implementation of the type. -
Adding the configuration value to the lint impl struct:
-
This first requires the definition of a lint impl struct. Lint impl structs are usually generated with the
declare_lint_pass!
macro. This struct needs to be defined manually to add some kind of metadata to it:#![allow(unused)] fn main() { // Generated struct definition declare_lint_pass!(StructName => [ LINT_NAME ]); // New manual definition struct #[derive(Copy, Clone)] pub struct StructName {} impl_lint_pass!(StructName => [ LINT_NAME ]); }
-
Next add the configuration value and a corresponding creation method like this:
#![allow(unused)] fn main() { #[derive(Copy, Clone)] pub struct StructName { configuration_ident: Type, } // ... impl StructName { pub fn new(configuration_ident: Type) -> Self { Self { configuration_ident, } } } }
-
-
Passing the configuration value to the lint impl struct:
First find the struct construction in the
clippy_lints
lib file. The configuration value is now cloned or copied into a local value that is then passed to the impl struct like this:// Default generated registration: store.register_*_pass(|| box module::StructName); // New registration with configuration value let configuration_ident = conf.configuration_ident.clone(); store.register_*_pass(move || box module::StructName::new(configuration_ident));
Congratulations the work is almost done. The configuration value can now be accessed in the linting code via
self.configuration_ident
. -
Adding tests:
- The default configured value can be tested like any normal lint in
tests/ui
. - The configuration itself will be tested separately in
tests/ui-toml
. Simply add a new subfolder with a fitting name. This folder contains aclippy.toml
file with the configuration value and a rust file that should be linted by Clippy. The test can otherwise be written as usual.
- The default configured value can be tested like any normal lint in
-
Update Lint Configuration
Run
cargo collect-metadata
to generate documentation changes for the book.
Cheat Sheet
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 (
is_type_diagnostic_item
,implements_trait
,snippet
, etc) - Clippy diagnostics
- Let chains
from_expansion
andin_external_macro
Span
Applicability
- Common tools for writing lints helps with common operations
- 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:
For LateLintPass
lints:
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 Zulip or in the issue/PR.
Define New Lints
The first step in the journey of a new lint is the definition and registration of the lint in Clippy's codebase. We can use the Clippy dev tools to handle this step since setting up the lint involves some boilerplate code.
Lint types
A lint type is the category of items and expressions in which your lint focuses on.
As of the writing of this documentation update, there are 12 types of lints
besides the numerous standalone lints living under clippy_lints/src/
:
cargo
casts
functions
loops
matches
methods
misc_early
operators
transmute
types
unit_types
utils / internal
(Clippy internal lints)
These types group together lints that share some common behaviors. For instance,
functions
groups together lints that deal with some aspects of functions in
Rust, like definitions, signatures and attributes.
For more information, feel free to compare the lint files under any category with All Clippy lints or ask one of the maintainers.
Lint name
A good lint name is important, make sure to check the lint naming guidelines. Don't worry, if the lint name doesn't fit, a Clippy team member will alert you in the PR process.
We'll name our example lint that detects functions named "foo" foo_functions
.
Check the lint naming guidelines to see why this name makes
sense.
Add and Register the Lint
Now that a name is chosen, we shall register foo_functions
as a lint to the
codebase. There are two ways to register a lint.
Standalone
If you believe that this new lint is a standalone lint (that doesn't belong to
any specific type like functions
or loops
), you can run the
following command in your Clippy project:
$ cargo dev new_lint --name=lint_name --pass=late --category=pedantic
There are two things to note here:
--pass
: We set--pass=late
in this command to do a late lint pass. The alternative is anearly
lint pass. We will discuss this difference in a later chapter.--category
: If not provided, thecategory
of this new lint will default tonursery
.
The cargo dev new_lint
command will create a new file:
clippy_lints/src/foo_functions.rs
as well as register the
lint.
Overall, you should notice that the following files are modified or created:
$ git status
On branch foo_functions
Changes not staged for commit:
(use "git add <file>..." to update what will be committed)
(use "git restore <file>..." to discard changes in working directory)
modified: CHANGELOG.md
modified: clippy_lints/src/lib.register_lints.rs
modified: clippy_lints/src/lib.register_pedantic.rs
modified: clippy_lints/src/lib.rs
Untracked files:
(use "git add <file>..." to include in what will be committed)
clippy_lints/src/foo_functions.rs
tests/ui/foo_functions.rs
Specific Type
Note: Lint types are listed in the "Lint types" section
If you believe that this new lint belongs to a specific type of lints,
you can run cargo dev new_lint
with a --type
option.
Since our foo_functions
lint is related to function calls, one could
argue that we should put it into a group of lints that detect some behaviors
of functions, we can put it in the functions
group.
Let's run the following command in your Clippy project:
$ cargo dev new_lint --name=foo_functions --type=functions --category=pedantic
This command will create, among other things, a new file:
clippy_lints/src/{type}/foo_functions.rs
.
In our case, the path will be clippy_lints/src/functions/foo_functions.rs
.
Notice how this command has a --type
flag instead of --pass
. Unlike a standalone
definition, this lint won't be registered in the traditional sense. Instead, you will
call your lint from within the type's lint pass, found in clippy_lints/src/{type}/mod.rs
.
A type is just the name of a directory in clippy_lints/src
, like functions
in
the example command. Clippy groups together some lints that share common behaviors,
so if your lint falls into one, it would be best to add it to that type.
Overall, you should notice that the following files are modified or created:
$ git status
On branch foo_functions
Changes not staged for commit:
(use "git add <file>..." to update what will be committed)
(use "git restore <file>..." to discard changes in working directory)
modified: CHANGELOG.md
modified: clippy_lints/src/declared_lints.rs
modified: clippy_lints/src/functions/mod.rs
Untracked files:
(use "git add <file>..." to include in what will be committed)
clippy_lints/src/functions/foo_functions.rs
tests/ui/foo_functions.rs
The define_clippy_lints
macro
After cargo dev new_lint
, you should see a macro with the name
define_clippy_lints
. It will be in the same file if you defined a standalone
lint, and it will be in mod.rs
if you defined a type-specific lint.
The macro looks something like this:
#![allow(unused)] fn main() { declare_clippy_lint! { /// ### What it does /// /// // Describe here what does the lint do. /// /// Triggers when detects... /// /// ### Why is this bad? /// /// // Describe why this pattern would be bad /// /// It can lead to... /// /// ### Example /// ```rust /// // example code where clippy issues a warning /// ``` /// Use instead: /// ```rust /// // example code which does not raise clippy warning /// ``` #[clippy::version = "1.70.0"] // <- In which version was this implemented, keep it up to date! pub LINT_NAME, // <- The lint name IN_ALL_CAPS pedantic, // <- The lint group "default lint description" // <- A lint description, e.g. "A function has an unit return type." } }
Lint registration
If we run the cargo dev new_lint
command for a new lint, the lint will be
automatically registered and there is nothing more to do.
However, sometimes we might want to declare a new lint by hand. In this case,
we'd use cargo dev update_lints
command afterwards.
When a lint is manually declared, we might need to register the lint pass
manually in the register_lints
function in clippy_lints/src/lib.rs
:
#![allow(unused)] fn main() { store.register_late_pass(|_| Box::new(foo_functions::FooFunctions)); }
As you might have guessed, where there's something late, there is something
early: in Clippy there is a register_early_pass
method as well. More on early
vs. late passes in a later chapter.
Without a call to one of register_early_pass
or register_late_pass
, the lint
pass in question will not be run.
Testing
Developing lints for Clippy is a Test-Driven Development (TDD) process because our first task before implementing any logic for a new lint is to write some test cases.
Develop Lints with Tests
When we develop Clippy, we enter a complex and chaotic realm full of programmatic issues, stylistic errors, illogical code and non-adherence to convention. Tests are the first layer of order we can leverage to define when and where we want a new lint to trigger or not.
Moreover, writing tests first help Clippy developers to find a balance for the first iteration of and further enhancements for a lint. With test cases on our side, we will not have to worry about over-engineering a lint on its first version nor missing out some obvious edge cases of the lint. This approach empowers us to iteratively enhance each lint.
Clippy UI Tests
We use UI tests for testing in Clippy. These UI tests check that the output of Clippy is exactly as we expect it to be. 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
with the command cargo bless
(seen later on).
Write Test Cases
Let us now think about some tests for our imaginary foo_functions
lint. We
start by opening the test file tests/ui/foo_functions.rs
that was created by
cargo dev new_lint
.
Update the file with some examples to get started:
#![warn(clippy::foo_functions)] // < Add this, so the lint is guaranteed to be enabled in this file // Impl methods struct A; impl A { pub fn fo(&self) {} pub fn foo(&self) {} //~ ERROR: function called "foo" pub fn food(&self) {} } // Default trait methods trait B { fn fo(&self) {} fn foo(&self) {} //~ ERROR: function called "foo" fn food(&self) {} } // Plain functions fn fo() {} fn foo() {} //~ ERROR: function called "foo" fn food() {} fn main() { // We also don't want to lint method calls foo(); let a = A; a.foo(); }
Without actual lint logic to emit the lint when we see a foo
function name,
this test will just pass, because no lint will be emitted. However, we can now
run the test with the following command:
$ TESTNAME=foo_functions cargo uitest
Clippy will compile and it will conclude with an ok
for the tests:
...Clippy warnings and test outputs...
test compile_test ... ok
test result: ok. 3 passed; 0 failed; 0 ignored; 0 measured; 0 filtered out; finished in 0.48s
This is normal. After all, we wrote a bunch of Rust code but we haven't really
implemented any logic for Clippy to detect foo
functions and emit a lint.
As we gradually implement our lint logic, we will keep running this UI test command. Clippy will begin outputting information that allows us to check if the output is turning into what we want it to be.
Example output
As our foo_functions
lint is tested, the output would look something like this:
failures:
---- compile_test stdout ----
normalized stderr:
error: function called "foo"
--> $DIR/foo_functions.rs:6:12
|
LL | pub fn foo(&self) {}
| ^^^
|
= note: `-D clippy::foo-functions` implied by `-D warnings`
error: function called "foo"
--> $DIR/foo_functions.rs:13:8
|
LL | fn foo(&self) {}
| ^^^
error: function called "foo"
--> $DIR/foo_functions.rs:19:4
|
LL | fn foo() {}
| ^^^
error: aborting due to 3 previous errors
Note the failures label at the top of the fragment, we'll get rid of it (saving this output) in the next section.
Note: You can run multiple test files by specifying a comma separated list:
TESTNAME=foo_functions,bar_methods,baz_structs
.
cargo bless
Once we are satisfied with the output, we need to run this command to
generate or update the .stderr
file for our lint:
$ TESTNAME=foo_functions cargo uibless
This writes the emitted lint suggestions and fixes to the .stderr
file, with
the reason for the lint, suggested fixes, and line numbers, etc.
Running TESTNAME=foo_functions cargo uitest
should pass then. When we commit
our lint, we need to commit the generated .stderr
files, too.
In general, you should only commit files changed by cargo bless
for the
specific lint you are creating/editing.
Note: If the generated
.stderr
, and.fixed
files are empty, they should be removed.
toml
Tests
Some lints can be configured through a clippy.toml
file. Those configuration
values are tested in tests/ui-toml
.
To add a new test there, create a new directory and add the files:
clippy.toml
: Put here the configuration value you want to test.lint_name.rs
: A test file where you put the testing code, that should see a different lint behavior according to the configuration set in theclippy.toml
file.
The potential .stderr
and .fixed
files can again be generated with cargo bless
.
Cargo Lints
The process of testing is different for Cargo lints in that now we are
interested in the Cargo.toml
manifest file. In this case, we also need a
minimal crate associated with that manifest. Those tests are generated in
tests/ui-cargo
.
Imagine we have a new example lint that is named foo_categories
, we can run:
$ cargo dev new_lint --name=foo_categories --pass=late --category=cargo
After running cargo dev new_lint
we will find by default two new crates,
each with its manifest file:
tests/ui-cargo/foo_categories/fail/Cargo.toml
: this file should cause the new lint to raise an error.tests/ui-cargo/foo_categories/pass/Cargo.toml
: this file should not trigger the lint.
If you need more cases, you can copy one of those crates (under
foo_categories
) and rename it.
The process of generating the .stderr
file is the same as for other lints
and prepending the TESTNAME
variable to cargo uitest
works for Cargo lints too.
Rustfix Tests
If the lint you are working on is making use of structured suggestions,
rustfix
will apply the suggestions from the lint to the test file code and
compare that to the contents of a .fixed
file.
Structured suggestions tell a user how to fix or re-write certain code that has
been linted with span_lint_and_sugg
.
Should span_lint_and_sugg
be used to generate a suggestion, but not all
suggestions lead to valid code, you can use the //@no-rustfix
comment on top
of the test file, to not run rustfix
on that file.
We'll talk about suggestions more in depth in a later chapter.
Use cargo bless
to automatically generate the .fixed
file after running
the tests.
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 from the working copy root.
$ cargo dev lint input.rs
Lint passes
Before working on the logic of a new lint, there is an important decision
that every Clippy developer must make: to use
EarlyLintPass
or LateLintPass
.
In short, the LateLintPass
has access to type and symbol information while the
EarlyLintPass
doesn't. If you don't need access to type information, use the
EarlyLintPass
.
Let us expand on these two traits more below.
EarlyLintPass
If you examine the documentation on EarlyLintPass
closely,
you'll see that every method defined for this trait utilizes a
EarlyContext
. In EarlyContext
's documentation, it states:
Context for lint checking of the AST, after expansion, before lowering to HIR.
Voilà . EarlyLintPass
works only on the Abstract Syntax Tree (AST) level.
And AST is generated during the lexing and parsing phase
of code compilation. Therefore, it doesn't know what a symbol means or information about types, and it should
be our trait choice for a new lint if the lint only deals with syntax-related issues.
While linting speed has not been a concern for Clippy,
the EarlyLintPass
is faster, and it should be your choice
if you know for sure a lint does not need type information.
As a reminder, run the following command to generate boilerplate for lints
that use EarlyLintPass
:
$ cargo dev new_lint --name=<your_new_lint> --pass=early --category=<your_category_choice>
Example for EarlyLintPass
Take a look at the following code:
#![allow(unused)] fn main() { let x = OurUndefinedType; x.non_existing_method(); }
From the AST perspective, both lines are "grammatically" correct.
The assignment uses a let
and ends with a semicolon. The invocation
of a method looks fine, too. As programmers, we might raise a few
questions already, but the parser is okay with it. This is what we
mean when we say EarlyLintPass
deals with only syntax on the AST level.
Alternatively, think of the foo_functions
lint we mentioned in
define new lints chapter.
We want the foo_functions
lint to detect functions with foo
as their name.
Writing a lint that only checks for the name of a function means that we only
work with the AST and don't have to access the type system at all (the type system is where
LateLintPass
comes into the picture).
LateLintPass
In contrast to EarlyLintPass
, LateLintPass
contains type information.
If you examine the documentation on LateLintPass
closely,
you see that every method defined in this trait utilizes a
LateContext
.
In LateContext
's documentation we will find methods that
deal with type-checking, which do not exist in EarlyContext
, such as:
Example for LateLintPass
Let us take a look with the following example:
#![allow(unused)] fn main() { let x = OurUndefinedType; x.non_existing_method(); }
These two lines of code are syntactically correct code from the perspective of the AST. We have an assignment and invoke a method on the variable that is of a type. Grammatically, everything is in order for the parser.
However, going down a level and looking at the type information,
the compiler will notice that both OurUndefinedType
and non_existing_method()
are undefined.
As Clippy developers, to access such type information, we must implement
LateLintPass
on our lint.
When you browse through Clippy's lints, you will notice that almost every lint
is implemented in a LateLintPass
, specifically because we often need to check
not only for syntactic issues but also type information.
Another limitation of the EarlyLintPass
is that the nodes are only identified
by their position in the AST. This means that you can't just get an id
and
request a certain node. For most lints that is fine, but we have some lints
that require the inspection of other nodes, which is easier at the HIR level.
In these cases, LateLintPass
is the better choice.
As a reminder, run the following command to generate boilerplate for lints
that use LateLintPass
:
$ cargo dev new_lint --name=<your_new_lint> --pass=late --category=<your_category_choice>
Emitting a lint
Once we have defined a lint, written UI tests and chosen the lint pass for the lint, we can begin the implementation of the lint logic so that we can emit it and gradually work towards a lint that behaves as expected.
Note that we will not go into concrete implementation of a lint logic in this chapter. We will go into details in later chapters as well as in two examples of real Clippy lints.
To emit a lint, we must implement a pass (see Lint Passes) for the lint that we have declared. In this example we'll implement a "late" lint, so take a look at the LateLintPass documentation, which provides an abundance of methods that we can implement for our lint.
#![allow(unused)] fn main() { pub trait LateLintPass<'tcx>: LintPass { // Trait methods } }
By far the most common method used for Clippy lints is check_expr
method, this is because Rust is an expression language and,
more often than not, the lint we want to work on must examine expressions.
Note: If you don't fully understand what expressions are in Rust, take a look at the official documentation on expressions
Other common ones include the check_fn
method and the
check_item
method.
Emitting a lint
Inside the trait method that we implement, we can write down the lint logic and emit the lint with suggestions.
Clippy's diagnostics provides quite a few diagnostic functions that we can use to emit lints. Take a look at the documentation to pick one that suits your lint's needs the best. Some common ones you will encounter in the Clippy repository includes:
span_lint
: Emits a lint without providing any other informationspan_lint_and_note
: Emits a lint and adds a notespan_lint_and_help
: Emits a lint and provides a helpful messagespan_lint_and_sugg
: Emits a lint and provides a suggestion to fix the codespan_lint_and_then
: Likespan_lint
, but allows for a lot of output customization.
#![allow(unused)] fn main() { impl<'tcx> LateLintPass<'tcx> for LintName { fn check_expr(&mut self, cx: &LateContext<'tcx>, expr: &'tcx Expr<'_>) { // Imagine that `some_lint_expr_logic` checks for requirements for emitting the lint if some_lint_expr_logic(expr) { span_lint_and_help( cx, // < The context LINT_NAME, // < The name of the lint in ALL CAPS expr.span, // < The span to lint "message on why the lint is emitted", None, // < An optional help span (to highlight something in the lint) "message that provides a helpful suggestion", ); } } } }
Note: The message should be matter of fact and avoid capitalization and punctuation. If multiple sentences are needed, the messages should probably be split up into an error + a help / note / suggestion message.
Suggestions: Automatic fixes
Some lints know what to change in order to fix the code. For example, the lint
range_plus_one
warns for ranges where the user wrote x..y + 1
instead of using an inclusive range (x..=y
). The fix to
this code would be changing the x..y + 1
expression to x..=y
. This is
where suggestions come in.
A suggestion is a change that the lint provides to fix the issue it is linting. The output looks something like this (from the example earlier):
error: an inclusive range would be more readable
--> $DIR/range_plus_minus_one.rs:37:14
|
LL | for _ in 1..1 + 1 {}
| ^^^^^^^^ help: use: `1..=1`
Not all suggestions are always right, some of them require human supervision, that's why we have Applicability.
Applicability indicates confidence in the correctness of the suggestion, some
are always right (Applicability::MachineApplicable
), but we use
Applicability::MaybeIncorrect
and others when talking about a suggestion that
may be incorrect.
Example
The same lint LINT_NAME
but that emits a suggestion would look something like this:
#![allow(unused)] fn main() { impl<'tcx> LateLintPass<'tcx> for LintName { fn check_expr(&mut self, cx: &LateContext<'tcx>, expr: &'tcx Expr<'_>) { // Imagine that `some_lint_expr_logic` checks for requirements for emitting the lint if some_lint_expr_logic(expr) { span_lint_and_sugg( // < Note this change cx, LINT_NAME, span, "message on why the lint is emitted", "use", format!("foo + {} * bar", snippet(cx, expr.span, "<default>")), // < Suggestion Applicability::MachineApplicable, ); } } } }
Suggestions generally use the format!
macro to interpolate the
old values with the new ones. To get code snippets, use one of the snippet*
functions from clippy_utils::source
.
How to choose between notes, help messages and suggestions
Notes are presented separately from the main lint message, they provide useful information that the user needs to understand why the lint was activated. They are the most helpful when attached to a span.
Examples:
Notes
error: calls to `std::mem::forget` with a reference instead of an owned value. Forgetting a reference does nothing.
--> $DIR/drop_forget_ref.rs:10:5
|
10 | forget(&SomeStruct);
| ^^^^^^^^^^^^^^^^^^^
|
= note: `-D clippy::forget-ref` implied by `-D warnings`
note: argument has type &SomeStruct
--> $DIR/drop_forget_ref.rs:10:12
|
10 | forget(&SomeStruct);
| ^^^^^^^^^^^
Help Messages
Help messages are specifically to help the user. These are used in situation where you can't provide a specific machine applicable suggestion. They can also be attached to a span.
Example:
error: constant division of 0.0 with 0.0 will always result in NaN
--> $DIR/zero_div_zero.rs:6:25
|
6 | let other_f64_nan = 0.0f64 / 0.0;
| ^^^^^^^^^^^^
|
= help: consider using `f64::NAN` if you would like a constant representing NaN
Suggestions
Suggestions are the most helpful, they are changes to the source code to fix the
error. The magic in suggestions is that tools like rustfix
can detect them and
automatically fix your code.
Example:
error: This `.fold` can be more succinctly expressed as `.any`
--> $DIR/methods.rs:390:13
|
390 | let _ = (0..3).fold(false, |acc, x| acc || x > 2);
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: try: `.any(|x| x > 2)`
|
Snippets
Snippets are pieces of the source code (as a string), they are extracted
generally using the snippet
function.
For example, if you want to know how an item looks (and you know the item's
span), you could use snippet(cx, span, "..")
.
Final: Run UI Tests to Emit the Lint
Now, if we run our UI test, we should see that Clippy now produces output that contains the lint message we designed.
The next step is to implement the logic properly, which is a detail that we will cover in the next chapters.
Type Checking
When we work on a new lint or improve an existing lint, we might want
to retrieve the type Ty
of an expression Expr
for a variety of
reasons. This can be achieved by utilizing the LateContext
that is available for LateLintPass
.
LateContext
and TypeckResults
The lint context LateContext
and TypeckResults
(returned by LateContext::typeck_results
) are the two most useful data structures
in LateLintPass
. They allow us to jump to type definitions and other compilation
stages such as HIR.
Note:
LateContext.typeck_results
's return value isTypeckResults
and is created in the type checking step, it includes useful information such as types of expressions, ways to resolve methods and so on.
TypeckResults
contains useful methods such as expr_ty
,
which gives us access to the underlying structure Ty
of a given expression.
#![allow(unused)] fn main() { pub fn expr_ty(&self, expr: &Expr<'_>) -> Ty<'tcx> }
As a side note, besides expr_ty
, TypeckResults
contains a
pat_ty()
method that is useful for retrieving a type from a pattern.
Ty
Ty
struct contains the type information of an expression.
Let's take a look at rustc_middle
's Ty
struct to examine this struct:
#![allow(unused)] fn main() { pub struct Ty<'tcx>(Interned<'tcx, WithStableHash<TyS<'tcx>>>); }
At a first glance, this struct looks quite esoteric. But at a closer look, we will see that this struct contains many useful methods for type checking.
For instance, is_char
checks if the given Ty
struct corresponds
to the primitive character type.
is_*
Usage
In some scenarios, all we need to do is check if the Ty
of an expression
is a specific type, such as char
type, so we could write the following:
#![allow(unused)] fn main() { impl LateLintPass<'_> for MyStructLint { fn check_expr(&mut self, cx: &LateContext<'_>, expr: &Expr<'_>) { // Get type of `expr` let ty = cx.typeck_results().expr_ty(expr); // Check if the `Ty` of this expression is of character type if ty.is_char() { println!("Our expression is a char!"); } } } }
Furthermore, if we examine the source code for is_char
,
we find something very interesting:
#![allow(unused)] fn main() { #[inline] pub fn is_char(self) -> bool { matches!(self.kind(), Char) } }
Indeed, we just discovered Ty
's kind()
method, which provides us
with TyKind
of a Ty
.
TyKind
TyKind
defines the kinds of types in Rust's type system.
Peeking into TyKind
documentation, we will see that it is an
enum of over 25 variants, including items such as Bool
, Int
, Ref
, etc.
kind
Usage
The TyKind
of Ty
can be returned by calling Ty.kind()
method.
We often use this method to perform pattern matching in Clippy.
For instance, if we want to check for a struct
, we could examine if the
ty.kind
corresponds to an Adt
(algebraic data type) and if its
AdtDef
is a struct:
#![allow(unused)] fn main() { impl LateLintPass<'_> for MyStructLint { fn check_expr(&mut self, cx: &LateContext<'_>, expr: &Expr<'_>) { // Get type of `expr` let ty = cx.typeck_results().expr_ty(expr); // Match its kind to enter the type match ty.kind { ty::Adt(adt_def, _) if adt_def.is_struct() => println!("Our `expr` is a struct!"), _ => () } } } }
hir::Ty
and ty::Ty
We've been talking about ty::Ty
this whole time without addressing hir::Ty
, but the latter
is also important to understand.
hir::Ty
would represent what the user wrote, while ty::Ty
is how the compiler sees the type and has more
information. Example:
#![allow(unused)] fn main() { fn foo(x: u32) -> u32 { x } }
Here the HIR sees the types without "thinking" about them, it knows that the function takes an u32
and returns
an u32
. As far as hir::Ty
is concerned those might be different types. But at the ty::Ty
level the compiler
understands that they're the same type, in-depth lifetimes, etc...
To get from a hir::Ty
to a ty::Ty
, you can use the hir_ty_to_ty
function outside of bodies or
outside of bodies the TypeckResults::node_type()
method.
Warning: Don't use
hir_ty_to_ty
inside of bodies, because this can cause ICEs.
Useful Links
Below are some useful links to further explore the concepts covered in this chapter:
Trait Checking
Besides type checking, we might want to examine if
a specific type Ty
implements certain trait when implementing a lint.
There are three approaches to achieve this, depending on if the target trait
that we want to examine has a diagnostic item,
lang item, or neither.
Using Diagnostic Items
As explained in the Rust Compiler Development Guide, diagnostic items are introduced for identifying types via Symbols.
For instance, if we want to examine whether an expression implements
the Iterator
trait, we could simply write the following code,
providing the LateContext
(cx
), our expression at hand, and
the symbol of the trait in question:
#![allow(unused)] fn main() { use clippy_utils::is_trait_method; use rustc_hir::Expr; use rustc_lint::{LateContext, LateLintPass}; use rustc_span::symbol::sym; impl LateLintPass<'_> for CheckIteratorTraitLint { fn check_expr(&mut self, cx: &LateContext<'_>, expr: &Expr<'_>) { let implements_iterator = cx.tcx.get_diagnostic_item(sym::Iterator).map_or(false, |id| { implements_trait(cx, cx.typeck_results().expr_ty(arg), id, &[]) }); if implements_iterator { // [...] } } } }
Note: Refer to this index for all the defined
Symbol
s.
Using Lang Items
Besides diagnostic items, we can also use lang_items
.
Take a look at the documentation to find that LanguageItems
contains
all language items defined in the compiler.
Using one of its *_trait
method, we could obtain the DefId of any
specific item, such as Clone
, Copy
, Drop
, Eq
, which are familiar
to many Rustaceans.
For instance, if we want to examine whether an expression expr
implements
Drop
trait, we could access LanguageItems
via our LateContext
's
TyCtxt, which provides a lang_items
method that will return the id of
Drop
trait to us. Then, by calling Clippy utils function implements_trait
we can check that the Ty
of the expr
implements the trait:
#![allow(unused)] fn main() { use clippy_utils::implements_trait; use rustc_hir::Expr; use rustc_lint::{LateContext, LateLintPass}; impl LateLintPass<'_> for CheckDropTraitLint { fn check_expr(&mut self, cx: &LateContext<'_>, expr: &Expr<'_>) { let ty = cx.typeck_results().expr_ty(expr); if cx.tcx.lang_items() .drop_trait() .map_or(false, |id| implements_trait(cx, ty, id, &[])) { println!("`expr` implements `Drop` trait!"); } } } }
Using Type Path
If neither diagnostic item nor a language item is available, we can use
clippy_utils::paths
with the match_trait_method
to determine trait
implementation.
Note: This approach should be avoided if possible, the best thing to do would be to make a PR to
rust-lang/rust
adding a diagnostic item.
Below, we check if the given expr
implements the Iterator
's trait method cloned
:
#![allow(unused)] fn main() { use clippy_utils::{match_trait_method, paths}; use rustc_hir::Expr; use rustc_lint::{LateContext, LateLintPass}; impl LateLintPass<'_> for CheckTokioAsyncReadExtTrait { fn check_expr(&mut self, cx: &LateContext<'_>, expr: &Expr<'_>) { if match_trait_method(cx, expr, &paths::CORE_ITER_CLONED) { println!("`expr` implements `CORE_ITER_CLONED` trait!"); } } } }
Method Checking
In some scenarios we might want to check for methods when developing a lint. There are two kinds of questions that we might be curious about:
- Invocation: Does an expression call a specific method?
- Definition: Does an
impl
define a method?
Checking if an expr
is calling a specific method
Suppose we have an expr
, we can check whether it calls a specific
method, e.g. our_fancy_method
, by performing a pattern match on
the ExprKind
that we can access from expr.kind
:
#![allow(unused)] fn main() { use rustc_hir as hir; use rustc_lint::{LateContext, LateLintPass}; use rustc_span::sym; use clippy_utils::is_trait_method; impl<'tcx> LateLintPass<'tcx> for OurFancyMethodLint { fn check_expr(&mut self, cx: &LateContext<'tcx>, expr: &'tcx hir::Expr<'_>) { // Check our expr is calling a method with pattern matching if let hir::ExprKind::MethodCall(path, _, [self_arg, ..]) = &expr.kind // Check if the name of this method is `our_fancy_method` && path.ident.name == sym!(our_fancy_method) // We can check the type of the self argument whenever necessary. // (It's necessary if we want to check that method is specifically belonging to a specific trait, // for example, a `map` method could belong to user-defined trait instead of to `Iterator`) // See the next section for more information. && is_trait_method(cx, self_arg, sym::OurFancyTrait) { println!("`expr` is a method call for `our_fancy_method`"); } } } }
Take a closer look at the ExprKind
enum variant MethodCall
for more
information on the pattern matching. As mentioned in Define
Lints, the methods
lint type is full of pattern
matching with MethodCall
in case the reader wishes to explore more.
Additionally, we use the clippy_utils::sym!
macro to conveniently
convert an input our_fancy_method
into a Symbol
and compare that symbol to
the Ident
's name in the PathSegment
in the MethodCall
.
Checking if a impl
block implements a method
While sometimes we want to check whether a method is being called or not, other
times we want to know if our Ty
defines a method.
To check if our impl
block defines a method our_fancy_method
, we will
utilize the check_impl_item
method that is available in our beloved
LateLintPass
(for more information, refer to the "Lint
Passes" chapter in the Clippy book). This method provides us
with an ImplItem
struct, which represents anything within an impl
block.
Let us take a look at how we might check for the implementation of
our_fancy_method
on a type:
#![allow(unused)] fn main() { use clippy_utils::ty::is_type_diagnostic_item; use clippy_utils::return_ty; use rustc_hir::{ImplItem, ImplItemKind}; use rustc_lint::{LateContext, LateLintPass}; use rustc_span::symbol::sym; impl<'tcx> LateLintPass<'tcx> for MyTypeImpl { fn check_impl_item(&mut self, cx: &LateContext<'tcx>, impl_item: &'tcx ImplItem<'_>) { // Check if item is a method/function if let ImplItemKind::Fn(ref signature, _) = impl_item.kind // Check the method is named `our_fancy_method` && impl_item.ident.name == sym!(our_fancy_method) // We can also check it has a parameter `self` && signature.decl.implicit_self.has_implicit_self() // We can go even further and even check if its return type is `String` && is_type_diagnostic_item(cx, return_ty(cx, impl_item.hir_id), sym::String) { println!("`our_fancy_method` is implemented!"); } } } }
Dealing with macros and expansions
Sometimes we might encounter Rust macro expansions while working with Clippy. While macro expansions are not as dramatic and profound as the expansion of our universe, they can certainly bring chaos to the orderly world of code and logic.
The general rule of thumb is that we should ignore code with macro expansions when working with Clippy because the code can be dynamic in ways that are difficult or impossible for us to foresee.
False Positives
What exactly do we mean by dynamic in ways that are difficult to foresee?
Macros are expanded in the EarlyLintPass
level,
so the Abstract Syntax Tree (AST) is generated in place of macros.
This means the code which we work with in Clippy is already expanded.
If we wrote a new lint, there is a possibility that the lint is triggered in macro-generated code. Since this expanded macro code is not written by the macro's user but really by the macro's author, the user cannot and should not be responsible for fixing the issue that triggers the lint.
Besides, a Span in a macro can be changed by the macro author. Therefore, any lint check related to lines or columns should be avoided since they might be changed at any time and become unreliable or incorrect information.
Because of these unforeseeable or unstable behaviors, macro expansion should often not be regarded as a part of the stable API. This is also why most lints check if they are inside a macro or not before emitting suggestions to the end user to avoid false positives.
How to Work with Macros
Several functions are available for working with macros.
The Span.from_expansion
method
We could utilize a span
's from_expansion
method, which
detects if the span
is from a macro expansion / desugaring.
This is a very common first step in a lint:
#![allow(unused)] fn main() { if expr.span.from_expansion() { // We most likely want to ignore it. return; } }
Span.ctxt
method
The span
's context, given by the method ctxt
and returning SpanContext,
represents if the span is from a macro expansion and, if it is, which
macro call expanded this span.
Sometimes, it is useful to check if the context of two spans are equal.
For instance, suppose we have the following line of code that would
expand into 1 + 0
:
#![allow(unused)] fn main() { // The following code expands to `1 + 0` for both `EarlyLintPass` and `LateLintPass` 1 + mac!() }
Assuming that we'd collect the 1
expression as a variable left
and the
0
/mac!()
expression as a variable right
, we can simply compare their
contexts. If the context is different, then we most likely are dealing with a
macro expansion and should just ignore it:
#![allow(unused)] fn main() { if left.span.ctxt() != right.span.ctxt() { // The code author most likely cannot modify this expression return; } }
Note: Code that is not from expansion is in the "root" context. So any spans whose
from_expansion
returnsfalse
can be assumed to have the same context. Because of this, usingspan.from_expansion()
is often sufficient.
Going a bit deeper, in a simple expression such as a == b
,
a
and b
have the same context.
However, in a macro_rules!
with a == $b
, $b
is expanded to
an expression that contains a different context from a
.
Take a look at the following macro m
:
#![allow(unused)] fn main() { macro_rules! m { ($a:expr, $b:expr) => { if $a.is_some() { $b; } } } let x: Option<u32> = Some(42); m!(x, x.unwrap()); }
If the m!(x, x.unwrapp());
line is expanded, we would get two expanded
expressions:
x.is_some()
(from the$a.is_some()
line in them
macro)x.unwrap()
(corresponding to$b
in them
macro)
Suppose x.is_some()
expression's span is associated with the x_is_some_span
variable
and x.unwrap()
expression's span is associated with x_unwrap_span
variable,
we could assume that these two spans do not share the same context:
#![allow(unused)] fn main() { // x.is_some() is from inside the macro // x.unwrap() is from outside the macro assert_ne!(x_is_some_span.ctxt(), x_unwrap_span.ctxt()); }
The in_external_macro
function
rustc_middle::lint
provides a function (in_external_macro
) that can
detect if the given span is from a macro defined in a foreign crate.
Therefore, if we really want a new lint to work with macro-generated code, this is the next line of defense to avoid macros not defined inside the current crate since it is unfair to the user if Clippy lints code which the user cannot change.
For example, assume we have the following code that is being examined by Clippy:
#![allow(unused)] fn main() { #[macro_use] extern crate a_foreign_crate_with_macros; // `foo` macro is defined in `a_foreign_crate_with_macros` foo!("bar"); }
Also assume that we get the corresponding variable foo_span
for the
foo
macro call, we could decide not to lint if in_external_macro
results in true
(note that cx
can be EarlyContext
or LateContext
):
#![allow(unused)] fn main() { if in_external_macro(cx.sess(), foo_span) { // We should ignore macro from a foreign crate. return; } }
Common tools for writing lints
You may need following tooltips to catch up with common operations.
Useful Rustc dev guide links:
Retrieving the type of expression
Sometimes you may want to retrieve the type Ty
of an expression Expr
, for
example to answer following questions:
- which type does this expression correspond to (using its
TyKind
)? - is it a sized type?
- is it a primitive type?
- does it implement a trait?
This operation is performed using the expr_ty()
method from the
TypeckResults
struct, that gives you access to the underlying
structure Ty
.
Example of use:
#![allow(unused)] fn main() { impl LateLintPass<'_> for MyStructLint { fn check_expr(&mut self, cx: &LateContext<'_>, expr: &Expr<'_>) { // Get type of `expr` let ty = cx.typeck_results().expr_ty(expr); // Match its kind to enter its type match ty.kind { ty::Adt(adt_def, _) if adt_def.is_struct() => println!("Our `expr` is a struct!"), _ => () } } } }
Similarly, in TypeckResults
methods, you have the
pat_ty()
method to retrieve a type from a pattern.
Two noticeable items here:
cx
is the lint contextLateContext
. The two most useful data structures in this context aretcx
and theTypeckResults
returned byLateContext::typeck_results
, allowing us to jump to type definitions and other compilation stages such as HIR.typeck_results
's return value isTypeckResults
and is created by type checking step, it includes useful information such as types of expressions, ways to resolve methods and so on.
Checking if an expr is calling a specific method
Starting with an expr
, you can check whether it is calling a specific method
some_method
:
#![allow(unused)] fn main() { impl<'tcx> LateLintPass<'tcx> for MyStructLint { fn check_expr(&mut self, cx: &LateContext<'tcx>, expr: &'tcx hir::Expr<'_>) { // Check our expr is calling a method if let hir::ExprKind::MethodCall(path, _, _self_arg, ..) = &expr.kind // Check the name of this method is `some_method` && path.ident.name == sym!(some_method) // Optionally, check the type of the self argument. // - See "Checking for a specific type" { // ... } } } }
Checking for a specific type
There are three ways to check if an expression type is a specific type we want to check for. All of these methods only check for the base type, generic arguments have to be checked separately.
#![allow(unused)] fn main() { use clippy_utils::ty::{is_type_diagnostic_item, is_type_lang_item}; use clippy_utils::{paths, match_def_path}; use rustc_span::symbol::sym; use rustc_hir::LangItem; impl LateLintPass<'_> for MyStructLint { fn check_expr(&mut self, cx: &LateContext<'_>, expr: &Expr<'_>) { // Getting the expression type let ty = cx.typeck_results().expr_ty(expr); // 1. Using diagnostic items // The last argument is the diagnostic item to check for if is_type_diagnostic_item(cx, ty, sym::Option) { // The type is an `Option` } // 2. Using lang items if is_type_lang_item(cx, ty, LangItem::RangeFull) { // The type is a full range like `.drain(..)` } // 3. Using the type path // This method should be avoided if possible if match_def_path(cx, def_id, &paths::RESULT) { // The type is a `core::result::Result` } } } }
Prefer using diagnostic items and lang items where possible.
Checking if a type implements a specific trait
There are three ways to do this, depending on if the target trait has a diagnostic item, lang item or neither.
#![allow(unused)] fn main() { use clippy_utils::ty::implements_trait; use clippy_utils::is_trait_method; use rustc_span::symbol::sym; impl LateLintPass<'_> for MyStructLint { fn check_expr(&mut self, cx: &LateContext<'_>, expr: &Expr<'_>) { // 1. Using diagnostic items with the expression // we use `is_trait_method` function from Clippy's utils if is_trait_method(cx, expr, sym::Iterator) { // method call in `expr` belongs to `Iterator` trait } // 2. Using lang items with the expression type let ty = cx.typeck_results().expr_ty(expr); if cx.tcx.lang_items() // we are looking for the `DefId` of `Drop` trait in lang items .drop_trait() // then we use it with our type `ty` by calling `implements_trait` from Clippy's utils .map_or(false, |id| implements_trait(cx, ty, id, &[])) { // `expr` implements `Drop` trait } } } }
Prefer using diagnostic and lang items, if the target trait has one.
We access lang items through the type context tcx
. tcx
is of type
TyCtxt
and is defined in the rustc_middle
crate. A list of defined
paths for Clippy can be found in paths.rs
Checking if a type defines a specific method
To check if our type defines a method called some_method
:
#![allow(unused)] fn main() { use clippy_utils::ty::is_type_diagnostic_item; use clippy_utils::return_ty; impl<'tcx> LateLintPass<'tcx> for MyTypeImpl { fn check_impl_item(&mut self, cx: &LateContext<'tcx>, impl_item: &'tcx ImplItem<'_>) { // Check if item is a method/function if let ImplItemKind::Fn(ref signature, _) = impl_item.kind // Check the method is named `some_method` && impl_item.ident.name == sym!(some_method) // We can also check it has a parameter `self` && signature.decl.implicit_self.has_implicit_self() // We can go further and even check if its return type is `String` && is_type_diagnostic_item(cx, return_ty(cx, impl_item.hir_id), sym!(string_type)) { // ... } } } }
Dealing with macros and expansions
Keep in mind that macros are already expanded and desugaring is already applied to the code representation that you are working with in Clippy. This unfortunately causes a lot of false positives because macro expansions are "invisible" unless you actively check for them. Generally speaking, code with macro expansions should just be ignored by Clippy because that code can be dynamic in ways that are difficult or impossible to see. Use the following functions to deal with macros:
-
span.from_expansion()
: detects if a span is from macro expansion or desugaring. Checking this is a common first step in a lint.if expr.span.from_expansion() { // just forget it return; }
-
span.ctxt()
: the span's context represents whether it is from expansion, and if so, which macro call expanded it. It is sometimes useful to check if the context of two spans are equal.// expands to `1 + 0`, but don't lint 1 + mac!()
if left.span.ctxt() != right.span.ctxt() { // the coder most likely cannot modify this expression return; }
Note: Code that is not from expansion is in the "root" context. So any spans where
from_expansion
returnstrue
can be assumed to have the same context. And so just usingspan.from_expansion()
is often good enough. -
in_external_macro(span)
: detect if the given span is from a macro defined in a foreign crate. If you want the lint to work with macro-generated code, this is the next line of defense to avoid macros not defined in the current crate. It doesn't make sense to lint code that the coder can't change.You may want to use it for example to not start linting in macros from other crates
#![allow(unused)] fn main() { use rustc_middle::lint::in_external_macro; use a_crate_with_macros::foo; // `foo` is defined in `a_crate_with_macros` foo!("bar"); // if we lint the `match` of `foo` call and test its span assert_eq!(in_external_macro(cx.sess(), match_span), true); }
-
span.ctxt()
: the span's context represents whether it is from expansion, and if so, what expanded itOne thing
SpanContext
is useful for is to check if two spans are in the same context. For example, ina == b
,a
andb
have the same context. In amacro_rules!
witha == $b
,$b
is expanded to some expression with a different context froma
.macro_rules! m { ($a:expr, $b:expr) => { if $a.is_some() { $b; } } } let x: Option<u32> = Some(42); m!(x, x.unwrap()); // These spans are not from the same context // x.is_some() is from inside the macro // x.unwrap() is from outside the macro assert_eq!(x_is_some_span.ctxt(), x_unwrap_span.ctxt());
Infrastructure
In order to deploy Clippy over rustup
, some infrastructure is necessary. This
chapter describes the different parts of the Clippy infrastructure that need to
be maintained to make this possible.
The most important part is the sync between the rust-lang/rust
repository and
the Clippy repository that takes place every two weeks. This process is
described in the Syncing changes between Clippy and rust-lang/rust
section.
A new Clippy release is done together with every Rust release, so every six weeks. The release process is described in the Release a new Clippy Version section. During a release cycle a changelog entry for the next release has to be written. The format of that and how to do that is documented in the Changelog Update section.
Note: The Clippy CI should also be described in this chapter, but for now is left as a TODO.
Syncing changes between Clippy and rust-lang/rust
Clippy currently gets built with a pinned nightly version.
In the rust-lang/rust
repository, where rustc resides, there's a copy of
Clippy that compiler hackers modify from time to time to adapt to changes in the
unstable API of the compiler.
We need to sync these changes back to this repository periodically, and the
changes made to this repository in the meantime also need to be synced to the
rust-lang/rust
repository.
To avoid flooding the rust-lang/rust
PR queue, this two-way sync process is
done in a bi-weekly basis if there's no urgent changes. This is done starting on
the day of the Rust stable release and then every other week. That way we
guarantee that we keep this repo up to date with the latest compiler API, and
every feature in Clippy is available for 2 weeks in nightly, before it can get
to beta. For reference, the first sync following this cadence was performed the
2020-08-27.
This process is described in detail in the following sections. For general
information about subtree
s in the Rust repository see the rustc-dev-guide.
Patching git-subtree to work with big repos
Currently, there's a bug in git-subtree
that prevents it from working properly
with the rust-lang/rust
repo. There's an open PR to fix that, but it's
stale. Before continuing with the following steps, we need to manually apply
that fix to our local copy of git-subtree
.
You can get the patched version of git-subtree
from here.
Put this file under /usr/lib/git-core
(making a backup of the previous file)
and make sure it has the proper permissions:
sudo cp --backup /path/to/patched/git-subtree.sh /usr/lib/git-core/git-subtree
sudo chmod --reference=/usr/lib/git-core/git-subtree~ /usr/lib/git-core/git-subtree
sudo chown --reference=/usr/lib/git-core/git-subtree~ /usr/lib/git-core/git-subtree
Note: The first time running
git subtree push
a cache has to be built. This involves going through the complete Clippy history once. For this you have to increase the stack limit though, which you can do withulimit -s 60000
. Make sure to run theulimit
command from the same session you call git subtree.
Note: If you are a Debian user,
dash
is the shell used by default for scripts instead ofsh
. This shell has a hardcoded recursion limit set to 1,000. In order to make this process work, you need to force the script to runbash
instead. You can do this by editing the first line of thegit-subtree
script and changingsh
tobash
.
Defining remotes
You may want to define remotes, so you don't have to type out the remote
addresses on every sync. You can do this with the following commands (these
commands still have to be run inside the rust
directory):
# Set clippy-upstream remote for pulls
$ git remote add clippy-upstream https://github.com/rust-lang/rust-clippy
# Make sure to not push to the upstream repo
$ git remote set-url --push clippy-upstream DISABLED
# Set a local remote
$ git remote add clippy-local /path/to/rust-clippy
Note: The following sections assume that you have set those remotes with the above remote names.
Performing the sync from rust-lang/rust
to Clippy
Here is a TL;DR version of the sync process (all the following commands have
to be run inside the rust
directory):
-
Clone the
rust-lang/rust
repository or make sure it is up-to-date. -
Checkout the commit from the latest available nightly. You can get it using
rustup check
. -
Sync the changes to the rust-copy of Clippy to your Clippy fork:
# Be sure to either use a net-new branch, e.g. `sync-from-rust`, or delete the branch beforehand # because changes cannot be fast forwarded and you have to run this command again. git subtree push -P src/tools/clippy clippy-local sync-from-rust
Note: Most of the time you have to create a merge commit in the
rust-clippy
repo (this has to be done in the Clippy repo, not in the rust-copy of Clippy):git fetch upstream # assuming upstream is the rust-lang/rust remote git checkout sync-from-rust git merge upstream/master --no-ff
Note: This is one of the few instances where a merge commit is allowed in a PR.
-
Bump the nightly version in the Clippy repository by changing the date in the rust-toolchain file to the current date and committing it with the message:
git commit -m "Bump nightly version -> YYYY-MM-DD"
-
Open a PR to
rust-lang/rust-clippy
and wait for it to get merged (to accelerate the process ping the@rust-lang/clippy
team in your PR and/or ask them in the Zulip stream.)
Performing the sync from Clippy to rust-lang/rust
All the following commands have to be run inside the rust
directory.
- Make sure you have checked out the latest
master
ofrust-lang/rust
. - Sync the
rust-lang/rust-clippy
master to the rust-copy of Clippy:git checkout -b sync-from-clippy git subtree pull -P src/tools/clippy clippy-upstream master
- Open a PR to
rust-lang/rust
Backport Changes
Sometimes it is necessary to backport changes to the beta release of Clippy. Backports in Clippy are rare and should be approved by the Clippy team. For example, a backport is done, if a crucial ICE was fixed or a lint is broken to a point, that it has to be disabled, before landing on stable.
Backports are done to the beta
branch of Clippy. Backports to stable Clippy
releases basically don't exist, since this would require a Rust point release,
which is almost never justifiable for a Clippy fix.
Backport the changes
Backports are done on the beta branch of the Clippy repository.
# Assuming the current directory corresponds to the Clippy repository
$ git checkout beta
$ git checkout -b backport
$ git cherry-pick <SHA> # `<SHA>` is the commit hash of the commit(s), that should be backported
$ git push origin backport
Now you should test that the backport passes all the tests in the Rust repository. You can do this with:
# Assuming the current directory corresponds to the Rust repository
$ git checkout beta
# Make sure to change `your-github-name` to your github name in the following command
$ git subtree pull -p src/tools/clippy https://github.com/<your-github-name>/rust-clippy backport
$ ./x.py test src/tools/clippy
Should the test fail, you can fix Clippy directly in the Rust repository. This has to be first applied to the Clippy beta branch and then again synced to the Rust repository, though. The easiest way to do this is:
# In the Rust repository
$ git diff --patch --relative=src/tools/clippy > clippy.patch
# In the Clippy repository
$ git apply /path/to/clippy.patch
$ git add -u
$ git commit -m "Fix rustup fallout"
$ git push origin backport
After this, you can open a PR to the beta
branch of the Clippy repository.
Update Clippy in the Rust Repository
This step must be done, after the PR of the previous step was merged.
After the backport landed in the Clippy repository, the branch has to be synced back to the beta branch of the Rust repository.
# Assuming the current directory corresponds to the Rust repository
$ git checkout beta
$ git checkout -b clippy_backport
$ git subtree pull -p src/tools/clippy https://github.com/rust-lang/rust-clippy beta
$ git push origin clippy_backport
Make sure to test the backport in the Rust repository before opening a PR. This
is done with ./x.py test src/tools/clippy
. If that passes all tests, open a PR
to the beta
branch of the Rust repository. In this PR you should tag the
Clippy team member, that agreed to the backport or the @rust-lang/clippy
team.
Make sure to add [beta]
to the title of the PR.
Changelog Update
If you want to help with updating the changelog, you're in the right place.
When to update
Typos and other small fixes/additions are always welcome.
Special care needs to be taken when it comes to updating the changelog for a new Rust release. For that purpose, the changelog is ideally updated during the week before an upcoming stable release. You can find the release dates on the Rust Forge.
Most of the time we only need to update the changelog for minor Rust releases. It's been very rare that Clippy changes were included in a patch release.
Changelog update walkthrough
1. Finding the relevant Clippy commits
Each Rust release ships with its own version of Clippy. The Clippy subtree can
be found in the tools
directory of the Rust repository.
Depending on the current time and what exactly you want to update, the following bullet points might be helpful:
- When writing the release notes for the upcoming stable release you need to
check out the Clippy commit of the current Rust
beta
branch. Link - When writing the release notes for the upcoming beta release, you need to
check out the Clippy commit of the current Rust
master
. Link - When writing the (forgotten) release notes for a past stable release, you need to check out the Rust release tag of the stable release. Link
Usually you want to write the changelog of the upcoming stable release. Make
sure though, that beta
was already branched in the Rust repository.
To find the commit hash, issue the following command when in a rust-lang/rust
checkout:
git log --oneline -- src/tools/clippy/ | grep -o "Merge commit '[a-f0-9]*' into .*" | head -1 | sed -e "s/Merge commit '\([a-f0-9]*\)' into .*/\1/g"
2. Fetching the PRs between those commits
Once you've got the correct commit range, run
util/fetch_prs_between.sh commit1 commit2 > changes.txt
and open that file in your editor of choice.
When updating the changelog it's also a good idea to make sure that commit1
is
already correct in the current changelog.
3. Authoring the final changelog
The above script should have dumped all the relevant PRs to the file you specified. It should have filtered out most of the irrelevant PRs already, but it's a good idea to do a manual cleanup pass where you look for more irrelevant PRs. If you're not sure about some PRs, just leave them in for the review and ask for feedback.
With the PRs filtered, you can start to take each PR and move the changelog:
content to CHANGELOG.md
. Adapt the wording as you see fit but try to keep it
somewhat coherent.
The order should roughly be:
- New lints
- Moves or deprecations of lints
- Changes that expand what code existing lints cover
- False positive fixes
- Suggestion fixes/improvements
- ICE fixes
- Documentation improvements
- Others
As section headers, we use:
### New Lints
### Moves and Deprecations
### Enhancements
### False Positive Fixes
### Suggestion Fixes/Improvements
### ICE Fixes
### Documentation Improvements
### Others
Please also be sure to update the Beta/Unreleased sections at the top with the relevant commit ranges.
3.1 Include beta-accepted
PRs
Look for the beta-accepted
label and make sure to also include the PRs with
that label in the changelog. If you can, remove the beta-accepted
labels
after the changelog PR was merged.
Note: Some of those PRs might even get backported to the previous
beta
. Those have to be included in the changelog of the previous release.
4. Update clippy::version
attributes
Next, make sure to check that the #[clippy::version]
attributes for the added
lints contain the correct version.
Release a new Clippy Version
NOTE: This document is probably only relevant to you, if you're a member of the Clippy team.
Clippy is released together with stable Rust releases. The dates for these releases can be found at the Rust Forge. This document explains the necessary steps to create a Clippy release.
- Remerge the
beta
branch - Update the
beta
branch - Find the Clippy commit
- Tag the stable commit
- Update
CHANGELOG.md
NOTE: This document is for stable Rust releases, not for point releases. For point releases, step 1. and 2. should be enough.
Remerge the beta
branch
This step is only necessary, if since the last release something was backported to the beta Rust release. The remerge is then necessary, to make sure that the Clippy commit, that was used by the now stable Rust release, persists in the tree of the Clippy repository.
To find out if this step is necessary run
# Assumes that the local master branch of rust-lang/rust-clippy is up-to-date
$ git fetch upstream
$ git branch master --contains upstream/beta
If this command outputs master
, this step is not necessary.
# Assuming `HEAD` is the current `master` branch of rust-lang/rust-clippy
$ git checkout -b backport_remerge
$ git merge upstream/beta
$ git diff # This diff has to be empty, otherwise something with the remerge failed
$ git push origin backport_remerge # This can be pushed to your fork
After this, open a PR to the master branch. In this PR, the commit hash of the
HEAD
of the beta
branch must exist. In addition to that, no files should be
changed by this PR.
Update the beta
branch
This step must be done after the PR of the previous step was merged.
First, the Clippy commit of the beta
branch of the Rust repository has to be
determined.
# Assuming the current directory corresponds to the Rust repository
$ git fetch upstream
$ git checkout upstream/beta
$ BETA_SHA=$(git log --oneline -- src/tools/clippy/ | grep -o "Merge commit '[a-f0-9]*' into .*" | head -1 | sed -e "s/Merge commit '\([a-f0-9]*\)' into .*/\1/g")
After finding the Clippy commit, the beta
branch in the Clippy repository can
be updated.
# Assuming the current directory corresponds to the Clippy repository
$ git checkout beta
$ git reset --hard $BETA_SHA
$ git push upstream beta
Find the Clippy commit
The first step is to tag the Clippy commit, that is included in the stable Rust release. This commit can be found in the Rust repository.
# Assuming the current directory corresponds to the Rust repository
$ git fetch upstream # `upstream` is the `rust-lang/rust` remote
$ git checkout 1.XX.0 # XX should be exchanged with the corresponding version
$ SHA=$(git log --oneline -- src/tools/clippy/ | grep -o "Merge commit '[a-f0-9]*' into .*" | head -1 | sed -e "s/Merge commit '\([a-f0-9]*\)' into .*/\1/g")
Tag the stable commit
After finding the Clippy commit, it can be tagged with the release number.
# Assuming the current directory corresponds to the Clippy repository
$ git checkout $SHA
$ git tag rust-1.XX.0 # XX should be exchanged with the corresponding version
$ git push upstream rust-1.XX.0 # `upstream` is the `rust-lang/rust-clippy` remote
After this, the release should be available on the Clippy release page.
Update the stable
branch
At this step you should have already checked out the commit of the rust-1.XX.0
tag. Updating the stable branch from here is as easy as:
# Assuming the current directory corresponds to the Clippy repository and the
# commit of the just created rust-1.XX.0 tag is checked out.
$ git push upstream rust-1.XX.0:stable # `upstream` is the `rust-lang/rust-clippy` remote
NOTE: Usually there are no stable backports for Clippy, so this update should be possible without force pushing or anything like this. If there should have happened a stable backport, make sure to re-merge those changes just as with the
beta
branch.
Update CHANGELOG.md
For this see the document on how to update the changelog.
If you don't have time to do a complete changelog update right away, just update the following parts:
-
Remove the
(beta)
from the new stable version:## Rust 1.XX (beta) -> ## Rust 1.XX
-
Update the release date line of the new stable version:
Current beta, release 20YY-MM-DD -> Current stable, released 20YY-MM-DD
-
Update the release date line of the previous stable version:
Current stable, released 20YY-MM-DD -> Released 20YY-MM-DD
The Clippy Book
This document explains how to make additions and changes to the Clippy book, the guide to Clippy that you're reading right now. The Clippy book is formatted with Markdown and generated by mdBook.
Get mdBook
While not strictly necessary since the book source is simply Markdown text
files, having mdBook locally will allow you to build, test and serve the book
locally to view changes before you commit them to the repository. You likely
already have cargo
installed, so the easiest option is to:
cargo install mdbook
See the mdBook installation instructions for other options.
Make changes
The book's
src
directory contains all the markdown files used to generate the book. If you
want to see your changes in real time, you can use the mdBook serve
command to
run a web server locally that will automatically update changes as they are
made. From the top level of your rust-clippy
directory:
mdbook serve book --open
Then navigate to http://localhost:3000
to see the generated book. While the
server is running, changes you make will automatically be updated.
For more information, see the mdBook guide.
Proposals
This chapter is about accepted proposals for changes that should be worked on in or around Clippy in the long run.
Besides adding more and more lints and improve the lints that Clippy already has, Clippy is also interested in making the experience of its users, developers and maintainers better over time. Projects that address bigger picture things like this usually take more time, and it is useful to have a proposal for those first. This is the place where such proposals are collected, so that we can refer to them when working on them.
Roadmap 2021
Summary
This Roadmap lays out the plans for Clippy in 2021:
- Improving usability and reliability
- Improving experience of contributors and maintainers
- Develop and specify processes
Members of the Clippy team will be assigned tasks from one or more of these topics. The team member is then responsible to complete the assigned tasks. This can either be done by implementing them or by providing mentorship to interested contributors.
Motivation
With the ongoing growth of the Rust language and with that of the whole ecosystem, also Clippy gets more and more users and contributors. This is good for the project, but also brings challenges along. Some of these challenges are:
- More issues about reliability or usability are popping up
- Traffic is hard to handle for a small team
- Bigger projects don't get completed due to the lack of processes and/or time of the team members
Additionally, according to the Rust Roadmap 2021, clear processes should be defined by every team and unified across teams. This Roadmap is the first step towards this.
Explanation
This section will explain the things that should be done in 2021. It is important to note, that this document focuses on the "What?", not the "How?". The later will be addressed in follow-up tracking issue, with an assigned team member.
The following is split up in two major sections. The first section covers the user facing plans, the second section the internal plans.
User Facing
Clippy should be as pleasant to use and configure as possible. This section covers plans that should be implemented to improve the situation of Clippy in this regard.
Usability
In the following, plans to improve the usability are covered.
No Output After cargo check
Currently, when cargo clippy
is run after cargo check
, it does not produce
any output. This is especially problematic since rust-analyzer
is on the rise,
and it uses cargo check
for checking code. A fix is already implemented, but
it still has to be pushed over the finish line. This also includes the
stabilization of the cargo clippy --fix
command or the support of multi-span
suggestions in rustfix
.
lints.toml
Configuration
This is something that comes up every now and then: a reusable configuration file, where lint levels can be defined. Discussions about this often lead to nothing specific or to "we need an RFC for this". And this is exactly what needs to be done. Get together with the cargo team and write an RFC and implement such a configuration file somehow and somewhere.
Lint Groups
There are more and more issues about managing lints in Clippy popping up. Lints
are hard to implement with a guarantee of no/few false positives (FPs). One way
to address this might be to introduce more lint groups to give users the ability
to better manage lints, or improve the process of classifying lints, so that
disabling lints due to FPs becomes rare. It is important to note, that Clippy
lints are less conservative than rustc
lints, which won't change in the
future.
Reliability
In the following, plans to improve the reliability are covered.
False Positive Rate
In the worst case, new lints are only available in nightly for 2 weeks, before hitting beta and ultimately stable. This and the fact that fewer people use nightly Rust nowadays makes it more probable that a lint with many FPs hits stable. This leads to annoyed users, that will disable these new lints in the best case and to more annoyed users, that will stop using Clippy in the worst. A process should be developed and implemented to prevent this from happening.
Internal
(The end of) 2020 has shown, that Clippy has to think about the available resources, especially regarding management and maintenance of the project. This section address issues affecting team members and contributors.
Management
In 2020 Clippy achieved over 1000 open issues with regularly between 25-35 open PRs. This is simultaneously a win and a loss. More issues and PRs means more people are interested in Clippy and in contributing to it. On the other hand, it means for team members more work and for contributors longer wait times for reviews. The following will describe plans how to improve the situation for both team members and contributors.
Clear Expectations for Team Members
According to the Rust Roadmap 2021, a document specifying what it means to be a member of the team should be produced. This should not put more pressure on the team members, but rather help them and interested folks to know what the expectations are. With this it should also be easier to recruit new team members and may encourage people to get in touch, if they're interested to join.
Scaling up the Team
More people means less work for each individual. Together with the document about expectations for team members, a document defining the process of how to join the team should be produced. This can also increase the stability of the team, in case of current members dropping out (temporarily). There can also be different roles in the team, like people triaging vs. people reviewing.
Regular Meetings
Other teams have regular meetings. Clippy is big enough that it might be worth to also do them. Especially if more people join the team, this can be important for sync-ups. Besides the asynchronous communication, that works well for working on separate lints, a meeting adds a synchronous alternative at a known time. This is especially helpful if there are bigger things that need to be discussed (like the projects in this roadmap). For starters bi-weekly meetings before Rust syncs might make sense.
Triaging
To get a handle on the influx of open issues, a process for triaging issues and PRs should be developed. Officially, Clippy follows the Rust triage process, but currently no one enforces it. This can be improved by sharing triage teams across projects or by implementing dashboards / tools which simplify triaging.
Development
Improving the developer and contributor experience is something the Clippy team works on regularly. Though, some things might need special attention and planing. These topics are listed in the following.
Process for New and Existing Lints
As already mentioned above, classifying new lints gets quite hard, because the probability of a buggy lint getting into stable is quite high. A process should be implemented on how to classify lints. In addition, a test system should be developed to find out which lints are currently problematic in real world code to fix or disable them.
Processes
Related to the point before, a process for suggesting and discussing major changes should be implemented. It's also not clearly defined when a lint should be enabled or disabled by default. This can also be improved by the test system mentioned above.
Dev-Tools
There's already cargo dev
which makes Clippy development easier and more
pleasant. This can still be expanded, so that it covers more areas of the
development process.
Contributor Guide
Similar to a Clippy Book, which describes how to use Clippy, a book about how to
contribute to Clippy might be helpful for new and existing contributors. There's
already the doc
directory in the Clippy repo, this can be turned into a
mdbook
.
rustc
integration
Recently Clippy was integrated with git subtree
into the rust-lang/rust
repository. This made syncing between the two repositories easier. A
#[non_exhaustive]
list of things that still can be improved is:
- Use the same
rustfmt
version and configuration asrustc
. - Make
cargo dev
work in the Rust repo, just as it works in the Clippy repo. E.g.cargo dev bless
orcargo dev update_lints
. And even add more things to it that might be useful for the Rust repo, e.g.cargo dev deprecate
. - Easier sync process. The
subtree
situation is not ideal.
Prioritization
The most pressing issues for users of Clippy are of course the user facing issues. So there should be a priority on those issues, but without losing track of the internal issues listed in this document.
Getting the FP rate of warn/deny-by-default lints under control should have the highest priority. Other user facing issues should also get a high priority, but shouldn't be in the way of addressing internal issues.
To better manage the upcoming projects, the basic internal processes, like meetings, tracking issues and documentation, should be established as soon as possible. They might even be necessary to properly manage the projects, regarding the user facing issues.
Prior Art
Rust Roadmap
Rust's roadmap process was established by RFC 1728 in 2016. Since then every year a roadmap was published, that defined the bigger plans for the coming years. This year roadmap can be found here.
Drawbacks
Big Roadmap
This roadmap is pretty big and not all items listed in this document might be addressed during 2021. Because this is the first roadmap for Clippy, having open tasks at the end of 2021 is fine, but they should be revisited in the 2022 roadmap.
- Feature Name: syntax-tree-patterns
- Start Date: 2019-03-12
- RFC PR: #3875
Summary
Introduce a domain-specific language (similar to regular expressions) that allows to describe lints using syntax tree patterns.
Motivation
Finding parts of a syntax tree (AST, HIR, ...) that have certain properties (e.g. "an if that has a block as its condition") is a major task when writing lints. For non-trivial lints, it often requires nested pattern matching of AST / HIR nodes. For example, testing that an expression is a boolean literal requires the following checks:
if let ast::ExprKind::Lit(lit) = &expr.node {
if let ast::LitKind::Bool(_) = &lit.node {
...
}
}
Writing this kind of matching code quickly becomes a complex task and the
resulting code is often hard to comprehend. The code below shows a simplified
version of the pattern matching required by the collapsible_if
lint:
// simplified version of the collapsible_if lint
if let ast::ExprKind::If(check, then, None) = &expr.node {
if then.stmts.len() == 1 {
if let ast::StmtKind::Expr(inner) | ast::StmtKind::Semi(inner) = &then.stmts[0].node {
if let ast::ExprKind::If(check_inner, content, None) = &inner.node {
...
}
}
}
}
The if_chain
macro can improve readability by flattening the nested if
statements, but the resulting code is still quite hard to read:
#![allow(unused)] fn main() { // simplified version of the collapsible_if lint if_chain! { if let ast::ExprKind::If(check, then, None) = &expr.node; if then.stmts.len() == 1; if let ast::StmtKind::Expr(inner) | ast::StmtKind::Semi(inner) = &then.stmts[0].node; if let ast::ExprKind::If(check_inner, content, None) = &inner.node; then { ... } } }
The code above matches if expressions that contain only another if expression (where both ifs don't have an else branch). While it's easy to explain what the lint does, it's hard to see that from looking at the code samples above.
Following the motivation above, the first goal this RFC is to simplify writing and reading lints.
The second part of the motivation is clippy's dependence on unstable compiler-internal data structures. Clippy lints are currently written against the compiler's AST / HIR which means that even small changes in these data structures might break a lot of lints. The second goal of this RFC is to make lints independent of the compiler's AST / HIR data structures.
Approach
A lot of complexity in writing lints currently seems to come from having to manually implement the matching logic (see code samples above). It's an imperative style that describes how to match a syntax tree node instead of specifying what should be matched against declaratively. In other areas, it's common to use declarative patterns to describe desired information and let the implementation do the actual matching. A well-known example of this approach are regular expressions. Instead of writing code that detects certain character sequences, one can describe a search pattern using a domain-specific language and search for matches using that pattern. The advantage of using a declarative domain-specific language is that its limited domain (e.g. matching character sequences in the case of regular expressions) allows to express entities in that domain in a very natural and expressive way.
While regular expressions are very useful when searching for patterns in flat character sequences, they cannot easily be applied to hierarchical data structures like syntax trees. This RFC therefore proposes a pattern matching system that is inspired by regular expressions and designed for hierarchical syntax trees.
Guide-level explanation
This proposal adds a pattern!
macro that can be used to specify a syntax tree
pattern to search for. A simple pattern is shown below:
#![allow(unused)] fn main() { pattern!{ my_pattern: Expr = Lit(Bool(false)) } }
This macro call defines a pattern named my_pattern
that can be matched against
an Expr
syntax tree node. The actual pattern (Lit(Bool(false))
in this case)
defines which syntax trees should match the pattern. This pattern matches
expressions that are boolean literals with value false
.
The pattern can then be used to implement lints in the following way:
...
impl EarlyLintPass for MyAwesomeLint {
fn check_expr(&mut self, cx: &EarlyContext, expr: &syntax::ast::Expr) {
if my_pattern(expr).is_some() {
cx.span_lint(
MY_AWESOME_LINT,
expr.span,
"This is a match for a simple pattern. Well done!",
);
}
}
}
The pattern!
macro call expands to a function my_pattern
that expects a
syntax tree expression as its argument and returns an Option
that indicates
whether the pattern matched.
Note: The result type is explained in more detail in a later section. For now, it's enough to know that the result is
Some
if the pattern matched andNone
otherwise.
Pattern syntax
The following examples demonstrate the pattern syntax:
Any (_
)
The simplest pattern is the any pattern. It matches anything and is therefore
similar to regex's *
.
#![allow(unused)] fn main() { pattern!{ // matches any expression my_pattern: Expr = _ } }
Node (<node-name>(<args>)
)
Nodes are used to match a specific variant of an AST node. A node has a name and
a number of arguments that depends on the node type. For example, the Lit
node
has a single argument that describes the type of the literal. As another
example, the If
node has three arguments describing the if's condition, then
block and else block.
#![allow(unused)] fn main() { pattern!{ // matches any expression that is a literal my_pattern: Expr = Lit(_) } pattern!{ // matches any expression that is a boolean literal my_pattern: Expr = Lit(Bool(_)) } pattern!{ // matches if expressions that have a boolean literal in their condition // Note: The `_?` syntax here means that the else branch is optional and can be anything. // This is discussed in more detail in the section `Repetition`. my_pattern: Expr = If( Lit(Bool(_)) , _, _?) } }
Literal (<lit>
)
A pattern can also contain Rust literals. These literals match themselves.
#![allow(unused)] fn main() { pattern!{ // matches the boolean literal false my_pattern: Expr = Lit(Bool(false)) } pattern!{ // matches the character literal 'x' my_pattern: Expr = Lit(Char('x')) } }
Alternations (a | b
)
#![allow(unused)] fn main() { pattern!{ // matches if the literal is a boolean or integer literal my_pattern: Lit = Bool(_) | Int(_) } pattern!{ // matches if the expression is a char literal with value 'x' or 'y' my_pattern: Expr = Lit( Char('x' | 'y') ) } }
Empty (()
)
The empty pattern represents an empty sequence or the None
variant of an
optional.
#![allow(unused)] fn main() { pattern!{ // matches if the expression is an empty array my_pattern: Expr = Array( () ) } pattern!{ // matches if expressions that don't have an else clause my_pattern: Expr = If(_, _, ()) } }
Sequence (<a> <b>
)
#![allow(unused)] fn main() { pattern!{ // matches the array [true, false] my_pattern: Expr = Array( Lit(Bool(true)) Lit(Bool(false)) ) } }
Repetition (<a>*
, <a>+
, <a>?
, <a>{n}
, <a>{n,m}
, <a>{n,}
)
Elements may be repeated. The syntax for specifying repetitions is identical to regex's syntax.
#![allow(unused)] fn main() { pattern!{ // matches arrays that contain 2 'x's as their last or second-last elements // Examples: // ['x', 'x'] match // ['x', 'x', 'y'] match // ['a', 'b', 'c', 'x', 'x', 'y'] match // ['x', 'x', 'y', 'z'] no match my_pattern: Expr = Array( _* Lit(Char('x')){2} _? ) } pattern!{ // matches if expressions that **may or may not** have an else block // Attn: `If(_, _, _)` matches only ifs that **have** an else block // // | if with else block | if without else block // If(_, _, _) | match | no match // If(_, _, _?) | match | match // If(_, _, ()) | no match | match my_pattern: Expr = If(_, _, _?) } }
Named submatch (<a>#<name>
)
#![allow(unused)] fn main() { pattern!{ // matches character literals and gives the literal the name foo my_pattern: Expr = Lit(Char(_)#foo) } pattern!{ // matches character literals and gives the char the name bar my_pattern: Expr = Lit(Char(_#bar)) } pattern!{ // matches character literals and gives the expression the name baz my_pattern: Expr = Lit(Char(_))#baz } }
The reason for using named submatches is described in the section The result type.
Summary
The following table gives an summary of the pattern syntax:
Syntax | Concept | Examples |
---|---|---|
_ | Any | _ |
<node-name>(<args>) | Node | Lit(Bool(true)) , If(_, _, _) |
<lit> | Literal | 'x' , false , 101 |
<a> \| <b> | Alternation | Char(_) \| Bool(_) |
() | Empty | Array( () ) |
<a> <b> | Sequence | Tuple( Lit(Bool(_)) Lit(Int(_)) Lit(_) ) |
<a>* <a>+ <a>? <a>{n} <a>{n,m} <a>{n,} | Repetition | Array( _* ) , Block( Semi(_)+ ) , If(_, _, Block(_)?) , Array( Lit(_){10} ) , Lit(_){5,10} , Lit(Bool(_)){10,} |
<a>#<name> | Named submatch | Lit(Int(_))#foo Lit(Int(_#bar)) |
The result type
A lot of lints require checks that go beyond what the pattern syntax described
above can express. For example, a lint might want to check whether a node was
created as part of a macro expansion or whether there's no comment above a node.
Another example would be a lint that wants to match two nodes that have the same
value (as needed by lints like almost_swapped
). Instead of allowing users to
write these checks into the pattern directly (which might make patterns hard to
read), the proposed solution allows users to assign names to parts of a pattern
expression. When matching a pattern against a syntax tree node, the return value
will contain references to all nodes that were matched by these named
subpatterns. This is similar to capture groups in regular expressions.
For example, given the following pattern
#![allow(unused)] fn main() { pattern!{ // matches character literals my_pattern: Expr = Lit(Char(_#val_inner)#val)#val_outer } }
one could get references to the nodes that matched the subpatterns in the following way:
...
fn check_expr(expr: &syntax::ast::Expr) {
if let Some(result) = my_pattern(expr) {
result.val_inner // type: &char
result.val // type: &syntax::ast::Lit
result.val_outer // type: &syntax::ast::Expr
}
}
The types in the result
struct depend on the pattern. For example, the
following pattern
#![allow(unused)] fn main() { pattern!{ // matches arrays of character literals my_pattern_seq: Expr = Array( Lit(_)*#foo ) } }
matches arrays that consist of any number of literal expressions. Because those
expressions are named foo
, the result struct contains a foo
attribute which
is a vector of expressions:
...
if let Some(result) = my_pattern_seq(expr) {
result.foo // type: Vec<&syntax::ast::Expr>
}
Another result type occurs when a name is only defined in one branch of an alternation:
#![allow(unused)] fn main() { pattern!{ // matches if expression is a boolean or integer literal my_pattern_alt: Expr = Lit( Bool(_#bar) | Int(_) ) } }
In the pattern above, the bar
name is only defined if the pattern matches a
boolean literal. If it matches an integer literal, the name isn't set. To
account for this, the result struct's bar
attribute is an option type:
...
if let Some(result) = my_pattern_alt(expr) {
result.bar // type: Option<&bool>
}
It's also possible to use a name in multiple alternation branches if they have compatible types:
pattern!{
// matches if expression is a boolean or integer literal
my_pattern_mult: Expr =
Lit(_#baz) | Array( Lit(_#baz) )
}
...
if let Some(result) = my_pattern_mult(expr) {
result.baz // type: &syntax::ast::Lit
}
Named submatches are a flat namespace and this is intended. In the example above, two different sub-structures are assigned to a flat name. I expect that for most lints, a flat namespace is sufficient and easier to work with than a hierarchical one.
Two stages
Using named subpatterns, users can write lints in two stages. First, a coarse selection of possible matches is produced by the pattern syntax. In the second stage, the named subpattern references can be used to do additional tests like asserting that a node hasn't been created as part of a macro expansion.
Implementing clippy lints using patterns
As a "real-world" example, I re-implemented the collapsible_if
lint using
patterns. The code can be found
here.
The pattern-based version passes all test cases that were written for
collapsible_if
.
Reference-level explanation
Overview
The following diagram shows the dependencies between the main parts of the proposed solution:
Pattern syntax
|
| parsing / lowering
v
PatternTree
^
|
|
IsMatch trait
|
|
+---------------+-----------+---------+
| | | |
v v v v
syntax::ast rustc::hir syn ...
The pattern syntax described in the previous section is parsed / lowered into the so-called PatternTree data structure that represents a valid syntax tree pattern. Matching a PatternTree against an actual syntax tree (e.g. rust ast / hir or the syn ast, ...) is done using the IsMatch trait.
The PatternTree and the IsMatch trait are introduced in more detail in the following sections.
PatternTree
The core data structure of this RFC is the PatternTree.
It's a data structure similar to rust's AST / HIR, but with the following differences:
- The PatternTree doesn't contain parsing information like
Span
s - The PatternTree can represent alternatives, sequences and optionals
The code below shows a simplified version of the current PatternTree:
Note: The current implementation can be found here.
#![allow(unused)] fn main() { pub enum Expr { Lit(Alt<Lit>), Array(Seq<Expr>), Block_(Alt<BlockType>), If(Alt<Expr>, Alt<BlockType>, Opt<Expr>), IfLet( Alt<BlockType>, Opt<Expr>, ), } pub enum Lit { Char(Alt<char>), Bool(Alt<bool>), Int(Alt<u128>), } pub enum Stmt { Expr(Alt<Expr>), Semi(Alt<Expr>), } pub enum BlockType { Block(Seq<Stmt>), } }
The Alt
, Seq
and Opt
structs look like these:
Note: The current implementation can be found here.
pub enum Alt<T> {
Any,
Elmt(Box<T>),
Alt(Box<Self>, Box<Self>),
Named(Box<Self>, ...)
}
pub enum Opt<T> {
Any, // anything, but not None
Elmt(Box<T>),
None,
Alt(Box<Self>, Box<Self>),
Named(Box<Self>, ...)
}
pub enum Seq<T> {
Any,
Empty,
Elmt(Box<T>),
Repeat(Box<Self>, RepeatRange),
Seq(Box<Self>, Box<Self>),
Alt(Box<Self>, Box<Self>),
Named(Box<Self>, ...)
}
pub struct RepeatRange {
pub start: usize,
pub end: Option<usize> // exclusive
}
Parsing / Lowering
The input of a pattern!
macro call is parsed into a ParseTree
first and then
lowered to a PatternTree
.
Valid patterns depend on the PatternTree definitions. For example, the pattern
Lit(Bool(_)*)
isn't valid because the parameter type of the Lit
variant of
the Expr
enum is Any<Lit>
and therefore doesn't support repetition (*
). As
another example, Array( Lit(_)* )
is a valid pattern because the parameter of
Array
is of type Seq<Expr>
which allows sequences and repetitions.
Note: names in the pattern syntax correspond to PatternTree enum variants. For example, the
Lit
in the pattern above refers to theLit
variant of theExpr
enum (Expr::Lit
), not theLit
enum.
The IsMatch Trait
The pattern syntax and the PatternTree are independent of specific syntax tree implementations (rust ast / hir, syn, ...). When looking at the different pattern examples in the previous sections, it can be seen that the patterns don't contain any information specific to a certain syntax tree implementation. In contrast, clippy lints currently match against ast / hir syntax tree nodes and therefore directly depend on their implementation.
The connection between the PatternTree and specific syntax tree
implementations is the IsMatch
trait. It defines how to match PatternTree
nodes against specific syntax tree nodes. A simplified implementation of the
IsMatch
trait is shown below:
pub trait IsMatch<O> {
fn is_match(&self, other: &'o O) -> bool;
}
This trait needs to be implemented on each enum of the PatternTree (for the
corresponding syntax tree types). For example, the IsMatch
implementation for
matching ast::LitKind
against the PatternTree's Lit
enum might look like
this:
#![allow(unused)] fn main() { impl IsMatch<ast::LitKind> for Lit { fn is_match(&self, other: &ast::LitKind) -> bool { match (self, other) { (Lit::Char(i), ast::LitKind::Char(j)) => i.is_match(j), (Lit::Bool(i), ast::LitKind::Bool(j)) => i.is_match(j), (Lit::Int(i), ast::LitKind::Int(j, _)) => i.is_match(j), _ => false, } } } }
All IsMatch
implementations for matching the current PatternTree against
syntax::ast
can be found
here.
Drawbacks
Performance
The pattern matching code is currently not optimized for performance, so it might be slower than hand-written matching code. Additionally, the two-stage approach (matching against the coarse pattern first and checking for additional properties later) might be slower than the current practice of checking for structure and additional properties in one pass. For example, the following lint
pattern!{
pat_if_without_else: Expr =
If(
_,
Block(
Expr( If(_, _, ())#inner )
| Semi( If(_, _, ())#inner )
)#then,
()
)
}
...
fn check_expr(&mut self, cx: &EarlyContext<'_>, expr: &ast::Expr) {
if let Some(result) = pat_if_without_else(expr) {
if !block_starts_with_comment(cx, result.then) {
...
}
}
first matches against the pattern and then checks that the then
block doesn't
start with a comment. Using clippy's current approach, it's possible to check
for these conditions earlier:
fn check_expr(&mut self, cx: &EarlyContext<'_>, expr: &ast::Expr) {
if_chain! {
if let ast::ExprKind::If(ref check, ref then, None) = expr.node;
if !block_starts_with_comment(cx, then);
if let Some(inner) = expr_block(then);
if let ast::ExprKind::If(ref check_inner, ref content, None) = inner.node;
then {
...
}
}
}
Whether or not this causes performance regressions depends on actual patterns. If it turns out to be a problem, the pattern matching algorithms could be extended to allow "early filtering" (see the Early Filtering section in Future Possibilities).
That being said, I don't see any conceptual limitations regarding pattern matching performance.
Applicability
Even though I'd expect that a lot of lints can be written using the proposed pattern syntax, it's unlikely that all lints can be expressed using patterns. I suspect that there will still be lints that need to be implemented by writing custom pattern matching code. This would lead to mix within clippy's codebase where some lints are implemented using patterns and others aren't. This inconsistency might be considered a drawback.
Rationale and alternatives
Specifying lints using syntax tree patterns has a couple of advantages compared to the current approach of manually writing matching code. First, syntax tree patterns allow users to describe patterns in a simple and expressive way. This makes it easier to write new lints for both novices and experts and also makes reading / modifying existing lints simpler.
Another advantage is that lints are independent of specific syntax tree
implementations (e.g. AST / HIR, ...). When these syntax tree implementations
change, only the IsMatch
trait implementations need to be adapted and existing
lints can remain unchanged. This also means that if the IsMatch
trait
implementations were integrated into the compiler, updating the IsMatch
implementations would be required for the compiler to compile successfully. This
could reduce the number of times clippy breaks because of changes in the
compiler. Another advantage of the pattern's independence is that converting an
EarlyLintPass
lint into a LatePassLint
wouldn't require rewriting the whole
pattern matching code. In fact, the pattern might work just fine without any
adaptions.
Alternatives
Rust-like pattern syntax
The proposed pattern syntax requires users to know the structure of the
PatternTree
(which is very similar to the AST's / HIR's structure) and also
the pattern syntax. An alternative would be to introduce a pattern syntax that
is similar to actual Rust syntax (probably like the quote!
macro). For
example, a pattern that matches if
expressions that have false
in their
condition could look like this:
if false {
#[*]
}
Problems
Extending Rust syntax (which is quite complex by itself) with additional syntax needed for specifying patterns (alternations, sequences, repetitions, named submatches, ...) might become difficult to read and really hard to parse properly.
For example, a pattern that matches a binary operation that has 0
on both
sides might look like this:
0 #[*:BinOpKind] 0
Now consider this slightly more complex example:
1 + 0 #[*:BinOpKind] 0
The parser would need to know the precedence of #[*:BinOpKind]
because it
affects the structure of the resulting AST. 1 + 0 + 0
is parsed as (1 + 0) + 0
while 1 + 0 * 0
is parsed as 1 + (0 * 0)
. Since the pattern could be any
BinOpKind
, the precedence cannot be known in advance.
Another example of a problem would be named submatches. Take a look at this pattern:
fn test() {
1 #foo
}
Which node is #foo
referring to? int
, ast::Lit
, ast::Expr
, ast::Stmt
?
Naming subpatterns in a rust-like syntax is difficult because a lot of AST nodes
don't have a syntactic element that can be used to put the name tag on. In these
situations, the only sensible option would be to assign the name tag to the
outermost node (ast::Stmt
in the example above), because the information of
all child nodes can be retrieved through the outermost node. The problem with
this then would be that accessing inner nodes (like ast::Lit
) would again
require manual pattern matching.
In general, Rust syntax contains a lot of code structure implicitly. This structure is reconstructed during parsing (e.g. binary operations are reconstructed using operator precedence and left-to-right) and is one of the reasons why parsing is a complex task. The advantage of this approach is that writing code is simpler for users.
When writing syntax tree patterns, each element of the hierarchy might have alternatives, repetitions, etc.. Respecting that while still allowing human-friendly syntax that contains structure implicitly seems to be really complex, if not impossible.
Developing such a syntax would also require to maintain a custom parser that is at least as complex as the Rust parser itself. Additionally, future changes in the Rust syntax might be incompatible with such a syntax.
In summary, I think that developing such a syntax would introduce a lot of complexity to solve a relatively minor problem.
The issue of users not knowing about the PatternTree structure could be solved by a tool that, given a rust program, generates a pattern that matches only this program (similar to the clippy author lint).
For some simple cases (like the first example above), it might be possible to successfully mix Rust and pattern syntax. This space could be further explored in a future extension.
Prior art
The pattern syntax is heavily inspired by regular expressions (repetitions, alternatives, sequences, ...).
From what I've seen until now, other linters also implement lints that directly work on syntax tree data structures, just like clippy does currently. I would therefore consider the pattern syntax to be new, but please correct me if I'm wrong.
Unresolved questions
How to handle multiple matches?
When matching a syntax tree node against a pattern, there are possibly multiple ways in which the pattern can be matched. A simple example of this would be the following pattern:
#![allow(unused)] fn main() { pattern!{ my_pattern: Expr = Array( _* Lit(_)+#literals) } }
This pattern matches arrays that end with at least one literal. Now given the
array [x, 1, 2]
, should 1
be matched as part of the _*
or the Lit(_)+
part of the pattern? The difference is important because the named submatch
#literals
would contain 1 or 2 elements depending how the pattern is matched.
In regular expressions, this problem is solved by matching "greedy" by default
and "non-greedy" optionally.
I haven't looked much into this yet because I don't know how relevant it is for most lints. The current implementation simply returns the first match it finds.
Future possibilities
Implement rest of Rust Syntax
The current project only implements a small part of the Rust syntax. In the
future, this should incrementally be extended to more syntax to allow
implementing more lints. Implementing more of the Rust syntax requires extending
the PatternTree
and IsMatch
implementations, but should be relatively
straight-forward.
Early filtering
As described in the Drawbacks/Performance section, allowing additional checks during the pattern matching might be beneficial.
The pattern below shows how this could look like:
#![allow(unused)] fn main() { pattern!{ pat_if_without_else: Expr = If( _, Block( Expr( If(_, _, ())#inner ) | Semi( If(_, _, ())#inner ) )#then, () ) where !in_macro(#then.span); } }
The difference compared to the currently proposed two-stage filtering is that
using early filtering, the condition (!in_macro(#then.span)
in this case)
would be evaluated as soon as the Block(_)#then
was matched.
Another idea in this area would be to introduce a syntax for backreferences.
They could be used to require that multiple parts of a pattern should match the
same value. For example, the assign_op_pattern
lint that searches for a = a op b
and recommends changing it to a op= b
requires that both occurrences of
a
are the same. Using =#...
as syntax for backreferences, the lint could be
implemented like this:
pattern!{
assign_op_pattern: Expr =
Assign(_#target, Binary(_, =#target, _)
}
Match descendant
A lot of lints currently implement custom visitors that check whether any subtree (which might not be a direct descendant) of the current node matches some properties. This cannot be expressed with the proposed pattern syntax. Extending the pattern syntax to allow patterns like "a function that contains at least two return statements" could be a practical addition.
Negation operator for alternatives
For patterns like "a literal that is not a boolean literal" one currently needs
to list all alternatives except the boolean case. Introducing a negation
operator that allows to write Lit(!Bool(_))
might be a good idea. This pattern
would be equivalent to Lit( Char(_) | Int(_) )
(given that currently only three
literal types are implemented).
Functional composition
Patterns currently don't have any concept of composition. This leads to repetitions within patterns. For example, one of the collapsible-if patterns currently has to be written like this:
#![allow(unused)] fn main() { pattern!{ pat_if_else: Expr = If( _, _, Block_( Block( Expr((If(_, _, _?) | IfLet(_, _?))#else_) | Semi((If(_, _, _?) | IfLet(_, _?))#else_) )#block_inner )#block ) | IfLet( _, Block_( Block( Expr((If(_, _, _?) | IfLet(_, _?))#else_) | Semi((If(_, _, _?) | IfLet(_, _?))#else_) )#block_inner )#block ) } }
If patterns supported defining functions of subpatterns, the code could be simplified as follows:
#![allow(unused)] fn main() { pattern!{ fn expr_or_semi(expr: Expr) -> Stmt { Expr(expr) | Semi(expr) } fn if_or_if_let(then: Block, else: Opt<Expr>) -> Expr { If(_, then, else) | IfLet(then, else) } pat_if_else: Expr = if_or_if_let( _, Block_( Block( expr_or_semi( if_or_if_let(_, _?)#else_ ) )#block_inner )#block ) } }
Additionally, common patterns like expr_or_semi
could be shared between
different lints.
Clippy Pattern Author
Another improvement could be to create a tool that, given some valid Rust syntax, generates a pattern that matches this syntax exactly. This would make starting to write a pattern easier. A user could take a look at the patterns generated for a couple of Rust code examples and use that information to write a pattern that matches all of them.
This is similar to clippy's author lint.
Supporting other syntaxes
Most of the proposed system is language-agnostic. For example, the pattern syntax could also be used to describe patterns for other programming languages.
In order to support other languages' syntaxes, one would need to implement
another PatternTree
that sufficiently describes the languages' AST and
implement IsMatch
for this PatternTree
and the languages' AST.
One aspect of this is that it would even be possible to write lints that work on the pattern syntax itself. For example, when writing the following pattern
#![allow(unused)] fn main() { pattern!{ my_pattern: Expr = Array( Lit(Bool(false)) Lit(Bool(false)) ) } }
a lint that works on the pattern syntax's AST could suggest using this pattern instead:
#![allow(unused)] fn main() { pattern!{ my_pattern: Expr = Array( Lit(Bool(false)){2} ) } }
In the future, clippy could use this system to also provide lints for custom syntaxes like those found in macros.