mirror of
https://github.com/rust-lang/rust.git
synced 2024-12-15 18:15:01 +00:00
Make suggestion give multiple alternatives
This commit is contained in:
parent
f12edfdb53
commit
5949f762bf
@ -1,9 +1,9 @@
|
||||
use clippy_utils::{
|
||||
consts::constant, diagnostics::span_lint_and_sugg, is_from_proc_macro, path_to_local, source::snippet_opt,
|
||||
consts::constant, diagnostics::span_lint_and_then, is_from_proc_macro, path_to_local, source::snippet_opt,
|
||||
};
|
||||
use rustc_errors::Applicability;
|
||||
use rustc_hir::{BinOpKind, Expr, ExprKind};
|
||||
use rustc_lint::{LateContext, LateLintPass, LintContext};
|
||||
use rustc_lint::{LateContext, LateLintPass, Lint, LintContext};
|
||||
use rustc_middle::lint::in_external_macro;
|
||||
use rustc_session::{declare_lint_pass, declare_tool_lint};
|
||||
|
||||
@ -53,12 +53,37 @@ declare_clippy_lint! {
|
||||
}
|
||||
declare_lint_pass!(ManualFloatMethods => [MANUAL_IS_INFINITE, MANUAL_IS_FINITE]);
|
||||
|
||||
#[derive(Clone, Copy)]
|
||||
enum Variant {
|
||||
ManualIsInfinite,
|
||||
ManualIsFinite,
|
||||
}
|
||||
|
||||
impl Variant {
|
||||
pub fn lint(self) -> &'static Lint {
|
||||
match self {
|
||||
Self::ManualIsInfinite => MANUAL_IS_INFINITE,
|
||||
Self::ManualIsFinite => MANUAL_IS_FINITE,
|
||||
}
|
||||
}
|
||||
|
||||
pub fn msg(self) -> &'static str {
|
||||
match self {
|
||||
Self::ManualIsInfinite => "manually checking if a float is infinite",
|
||||
Self::ManualIsFinite => "manually checking if a float is finite",
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
impl<'tcx> LateLintPass<'tcx> for ManualFloatMethods {
|
||||
fn check_expr(&mut self, cx: &LateContext<'tcx>, expr: &'tcx Expr<'tcx>) {
|
||||
if !in_external_macro(cx.sess(), expr.span)
|
||||
&& (!cx.param_env.is_const() || cx.tcx.features().active(sym!(const_float_classify)))
|
||||
&& let ExprKind::Binary(kind, lhs, rhs) = expr.kind
|
||||
&& let ExprKind::Binary(lhs_kind, lhs_lhs, lhs_rhs) = lhs.kind
|
||||
&& let ExprKind::Binary(rhs_kind, rhs_lhs, rhs_rhs) = rhs.kind
|
||||
// Checking all possible scenarios using a function would be a hopeless task, as we have
|
||||
// 16 possible alignments of constants/operands. For now, let's use `partition`.
|
||||
&& let (operands, consts) = [lhs_lhs, lhs_rhs, rhs_lhs, rhs_rhs]
|
||||
.into_iter()
|
||||
.partition::<Vec<&Expr<'_>>, _>(|i| path_to_local(i).is_some())
|
||||
@ -74,24 +99,51 @@ impl<'tcx> LateLintPass<'tcx> for ManualFloatMethods {
|
||||
&& let Some(local_snippet) = snippet_opt(cx, first.span)
|
||||
&& !is_from_proc_macro(cx, expr)
|
||||
{
|
||||
let (msg, lint, sugg_fn) = match (kind.node, lhs_kind.node, rhs_kind.node) {
|
||||
(BinOpKind::Or, BinOpKind::Eq, BinOpKind::Eq) => {
|
||||
("manually checking if a float is infinite", MANUAL_IS_INFINITE, "is_infinite")
|
||||
},
|
||||
(BinOpKind::And, BinOpKind::Ne, BinOpKind::Ne) => {
|
||||
("manually checking if a float is finite", MANUAL_IS_FINITE, "is_finite")
|
||||
},
|
||||
let variant = match (kind.node, lhs_kind.node, rhs_kind.node) {
|
||||
(BinOpKind::Or, BinOpKind::Eq, BinOpKind::Eq) => Variant::ManualIsInfinite,
|
||||
(BinOpKind::And, BinOpKind::Ne, BinOpKind::Ne) => Variant::ManualIsFinite,
|
||||
_ => return,
|
||||
};
|
||||
|
||||
span_lint_and_sugg(
|
||||
span_lint_and_then(
|
||||
cx,
|
||||
lint,
|
||||
variant.lint(),
|
||||
expr.span,
|
||||
msg,
|
||||
"try",
|
||||
format!("{local_snippet}.{sugg_fn}()"),
|
||||
Applicability::MachineApplicable,
|
||||
variant.msg(),
|
||||
|diag| {
|
||||
match variant {
|
||||
Variant::ManualIsInfinite => {
|
||||
diag.span_suggestion(
|
||||
expr.span,
|
||||
"use the dedicated method instead",
|
||||
format!("{local_snippet}.is_infinite()"),
|
||||
Applicability::MachineApplicable,
|
||||
);
|
||||
},
|
||||
Variant::ManualIsFinite => {
|
||||
// TODO: There's probably some better way to do this, i.e., create
|
||||
// multiple suggestions with notes between each of them
|
||||
diag.span_suggestion_verbose(
|
||||
expr.span,
|
||||
"use the dedicated method instead",
|
||||
format!("{local_snippet}.is_finite()"),
|
||||
Applicability::MaybeIncorrect,
|
||||
);
|
||||
diag.span_suggestion_verbose(
|
||||
expr.span,
|
||||
"this will alter how it handles NaN; if that is a problem, use instead",
|
||||
format!("{local_snippet}.is_finite() || {local_snippet}.is_nan()"),
|
||||
Applicability::MaybeIncorrect,
|
||||
);
|
||||
diag.span_suggestion_verbose(
|
||||
expr.span,
|
||||
"or, for conciseness",
|
||||
format!("!{local_snippet}.is_infinite()"),
|
||||
Applicability::MaybeIncorrect,
|
||||
);
|
||||
}
|
||||
}
|
||||
}
|
||||
);
|
||||
}
|
||||
}
|
||||
|
@ -1,37 +0,0 @@
|
||||
//@run-rustfix
|
||||
//@aux-build:proc_macros.rs:proc-macro
|
||||
#![allow(clippy::needless_if, unused)]
|
||||
#![warn(clippy::manual_is_infinite, clippy::manual_is_finite)]
|
||||
|
||||
#[macro_use]
|
||||
extern crate proc_macros;
|
||||
|
||||
const INFINITE: f32 = f32::INFINITY;
|
||||
const NEG_INFINITE: f32 = f32::NEG_INFINITY;
|
||||
|
||||
fn main() {
|
||||
let x = 1.0f32;
|
||||
if x.is_infinite() {}
|
||||
if x.is_finite() {}
|
||||
if x.is_infinite() {}
|
||||
if x.is_finite() {}
|
||||
let x = 1.0f64;
|
||||
if x.is_infinite() {}
|
||||
if x.is_finite() {}
|
||||
// Don't lint
|
||||
if x.is_infinite() {}
|
||||
if x.is_finite() {}
|
||||
// If they're doing it this way, they probably know what they're doing
|
||||
if x.abs() < f64::INFINITY {}
|
||||
external! {
|
||||
let x = 1.0;
|
||||
if x == f32::INFINITY || x == f32::NEG_INFINITY {}
|
||||
if x != f32::INFINITY && x != f32::NEG_INFINITY {}
|
||||
}
|
||||
with_span! {
|
||||
span
|
||||
let x = 1.0;
|
||||
if x == f32::INFINITY || x == f32::NEG_INFINITY {}
|
||||
if x != f32::INFINITY && x != f32::NEG_INFINITY {}
|
||||
}
|
||||
}
|
@ -1,4 +1,3 @@
|
||||
//@run-rustfix
|
||||
//@aux-build:proc_macros.rs:proc-macro
|
||||
#![allow(clippy::needless_if, unused)]
|
||||
#![warn(clippy::manual_is_infinite, clippy::manual_is_finite)]
|
||||
|
@ -1,42 +1,80 @@
|
||||
error: manually checking if a float is infinite
|
||||
--> $DIR/manual_float_methods.rs:14:8
|
||||
--> $DIR/manual_float_methods.rs:13:8
|
||||
|
|
||||
LL | if x == f32::INFINITY || x == f32::NEG_INFINITY {}
|
||||
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: try: `x.is_infinite()`
|
||||
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: use the dedicated method instead: `x.is_infinite()`
|
||||
|
|
||||
= note: `-D clippy::manual-is-infinite` implied by `-D warnings`
|
||||
|
||||
error: manually checking if a float is finite
|
||||
--> $DIR/manual_float_methods.rs:15:8
|
||||
--> $DIR/manual_float_methods.rs:14:8
|
||||
|
|
||||
LL | if x != f32::INFINITY && x != f32::NEG_INFINITY {}
|
||||
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: try: `x.is_finite()`
|
||||
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
||||
|
|
||||
= note: `-D clippy::manual-is-finite` implied by `-D warnings`
|
||||
help: use the dedicated method instead
|
||||
|
|
||||
LL | if x.is_finite() {}
|
||||
| ~~~~~~~~~~~~~
|
||||
help: this will alter how it handles NaN; if that is a problem, use instead
|
||||
|
|
||||
LL | if x.is_finite() || x.is_nan() {}
|
||||
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~
|
||||
help: or, for conciseness
|
||||
|
|
||||
LL | if !x.is_infinite() {}
|
||||
| ~~~~~~~~~~~~~~~~
|
||||
|
||||
error: manually checking if a float is infinite
|
||||
--> $DIR/manual_float_methods.rs:16:8
|
||||
--> $DIR/manual_float_methods.rs:15:8
|
||||
|
|
||||
LL | if x == INFINITE || x == NEG_INFINITE {}
|
||||
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: try: `x.is_infinite()`
|
||||
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: use the dedicated method instead: `x.is_infinite()`
|
||||
|
||||
error: manually checking if a float is finite
|
||||
--> $DIR/manual_float_methods.rs:17:8
|
||||
--> $DIR/manual_float_methods.rs:16:8
|
||||
|
|
||||
LL | if x != INFINITE && x != NEG_INFINITE {}
|
||||
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: try: `x.is_finite()`
|
||||
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
||||
|
|
||||
help: use the dedicated method instead
|
||||
|
|
||||
LL | if x.is_finite() {}
|
||||
| ~~~~~~~~~~~~~
|
||||
help: this will alter how it handles NaN; if that is a problem, use instead
|
||||
|
|
||||
LL | if x.is_finite() || x.is_nan() {}
|
||||
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~
|
||||
help: or, for conciseness
|
||||
|
|
||||
LL | if !x.is_infinite() {}
|
||||
| ~~~~~~~~~~~~~~~~
|
||||
|
||||
error: manually checking if a float is infinite
|
||||
--> $DIR/manual_float_methods.rs:19:8
|
||||
--> $DIR/manual_float_methods.rs:18:8
|
||||
|
|
||||
LL | if x == f64::INFINITY || x == f64::NEG_INFINITY {}
|
||||
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: try: `x.is_infinite()`
|
||||
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: use the dedicated method instead: `x.is_infinite()`
|
||||
|
||||
error: manually checking if a float is finite
|
||||
--> $DIR/manual_float_methods.rs:20:8
|
||||
--> $DIR/manual_float_methods.rs:19:8
|
||||
|
|
||||
LL | if x != f64::INFINITY && x != f64::NEG_INFINITY {}
|
||||
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ help: try: `x.is_finite()`
|
||||
| ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
|
||||
|
|
||||
help: use the dedicated method instead
|
||||
|
|
||||
LL | if x.is_finite() {}
|
||||
| ~~~~~~~~~~~~~
|
||||
help: this will alter how it handles NaN; if that is a problem, use instead
|
||||
|
|
||||
LL | if x.is_finite() || x.is_nan() {}
|
||||
| ~~~~~~~~~~~~~~~~~~~~~~~~~~~
|
||||
help: or, for conciseness
|
||||
|
|
||||
LL | if !x.is_infinite() {}
|
||||
| ~~~~~~~~~~~~~~~~
|
||||
|
||||
error: aborting due to 6 previous errors
|
||||
|
||||
|
Loading…
Reference in New Issue
Block a user