rust/clippy_lints/src/pass_by_ref_or_value.rs
2022-05-30 01:46:15 +00:00

286 lines
11 KiB
Rust

use std::cmp;
use std::iter;
use clippy_utils::diagnostics::span_lint_and_sugg;
use clippy_utils::source::snippet;
use clippy_utils::ty::is_copy;
use clippy_utils::{is_self, is_self_ty};
use if_chain::if_chain;
use rustc_ast::attr;
use rustc_errors::Applicability;
use rustc_hir as hir;
use rustc_hir::intravisit::FnKind;
use rustc_hir::{BindingAnnotation, Body, FnDecl, HirId, Impl, ItemKind, MutTy, Mutability, Node, PatKind};
use rustc_lint::{LateContext, LateLintPass};
use rustc_middle::ty;
use rustc_middle::ty::layout::LayoutOf;
use rustc_session::{declare_tool_lint, impl_lint_pass};
use rustc_span::def_id::LocalDefId;
use rustc_span::{sym, Span};
use rustc_target::spec::abi::Abi;
use rustc_target::spec::Target;
declare_clippy_lint! {
/// ### What it does
/// Checks for functions taking arguments by reference, where
/// the argument type is `Copy` and small enough to be more efficient to always
/// pass by value.
///
/// ### Why is this bad?
/// In many calling conventions instances of structs will
/// be passed through registers if they fit into two or less general purpose
/// registers.
///
/// ### Known problems
/// This lint is target register size dependent, it is
/// limited to 32-bit to try and reduce portability problems between 32 and
/// 64-bit, but if you are compiling for 8 or 16-bit targets then the limit
/// will be different.
///
/// The configuration option `trivial_copy_size_limit` can be set to override
/// this limit for a project.
///
/// This lint attempts to allow passing arguments by reference if a reference
/// to that argument is returned. This is implemented by comparing the lifetime
/// of the argument and return value for equality. However, this can cause
/// false positives in cases involving multiple lifetimes that are bounded by
/// each other.
///
/// Also, it does not take account of other similar cases where getting memory addresses
/// matters; namely, returning the pointer to the argument in question,
/// and passing the argument, as both references and pointers,
/// to a function that needs the memory address. For further details, refer to
/// [this issue](https://github.com/rust-lang/rust-clippy/issues/5953)
/// that explains a real case in which this false positive
/// led to an **undefined behavior** introduced with unsafe code.
///
/// ### Example
///
/// ```rust
/// // Bad
/// fn foo(v: &u32) {}
/// ```
///
/// ```rust
/// // Better
/// fn foo(v: u32) {}
/// ```
#[clippy::version = "pre 1.29.0"]
pub TRIVIALLY_COPY_PASS_BY_REF,
pedantic,
"functions taking small copyable arguments by reference"
}
declare_clippy_lint! {
/// ### What it does
/// Checks for functions taking arguments by value, where
/// the argument type is `Copy` and large enough to be worth considering
/// passing by reference. Does not trigger if the function is being exported,
/// because that might induce API breakage, if the parameter is declared as mutable,
/// or if the argument is a `self`.
///
/// ### Why is this bad?
/// Arguments passed by value might result in an unnecessary
/// shallow copy, taking up more space in the stack and requiring a call to
/// `memcpy`, which can be expensive.
///
/// ### Example
/// ```rust
/// #[derive(Clone, Copy)]
/// struct TooLarge([u8; 2048]);
///
/// // Bad
/// fn foo(v: TooLarge) {}
/// ```
/// ```rust
/// #[derive(Clone, Copy)]
/// struct TooLarge([u8; 2048]);
///
/// // Good
/// fn foo(v: &TooLarge) {}
/// ```
#[clippy::version = "1.49.0"]
pub LARGE_TYPES_PASSED_BY_VALUE,
pedantic,
"functions taking large arguments by value"
}
#[derive(Copy, Clone)]
pub struct PassByRefOrValue {
ref_min_size: u64,
value_max_size: u64,
avoid_breaking_exported_api: bool,
}
impl<'tcx> PassByRefOrValue {
pub fn new(
ref_min_size: Option<u64>,
value_max_size: u64,
avoid_breaking_exported_api: bool,
target: &Target,
) -> Self {
let ref_min_size = ref_min_size.unwrap_or_else(|| {
let bit_width = u64::from(target.pointer_width);
// Cap the calculated bit width at 32-bits to reduce
// portability problems between 32 and 64-bit targets
let bit_width = cmp::min(bit_width, 32);
#[expect(clippy::integer_division)]
let byte_width = bit_width / 8;
// Use a limit of 2 times the register byte width
byte_width * 2
});
Self {
ref_min_size,
value_max_size,
avoid_breaking_exported_api,
}
}
fn check_poly_fn(&mut self, cx: &LateContext<'tcx>, def_id: LocalDefId, decl: &FnDecl<'_>, span: Option<Span>) {
if self.avoid_breaking_exported_api && cx.access_levels.is_exported(def_id) {
return;
}
let fn_sig = cx.tcx.fn_sig(def_id);
let fn_sig = cx.tcx.erase_late_bound_regions(fn_sig);
let fn_body = cx.enclosing_body.map(|id| cx.tcx.hir().body(id));
for (index, (input, &ty)) in iter::zip(decl.inputs, fn_sig.inputs()).enumerate() {
// All spans generated from a proc-macro invocation are the same...
match span {
Some(s) if s == input.span => return,
_ => (),
}
match ty.kind() {
ty::Ref(input_lt, ty, Mutability::Not) => {
// Use lifetimes to determine if we're returning a reference to the
// argument. In that case we can't switch to pass-by-value as the
// argument will not live long enough.
let output_lts = match *fn_sig.output().kind() {
ty::Ref(output_lt, _, _) => vec![output_lt],
ty::Adt(_, substs) => substs.regions().collect(),
_ => vec![],
};
if_chain! {
if !output_lts.contains(input_lt);
if is_copy(cx, *ty);
if let Some(size) = cx.layout_of(*ty).ok().map(|l| l.size.bytes());
if size <= self.ref_min_size;
if let hir::TyKind::Rptr(_, MutTy { ty: decl_ty, .. }) = input.kind;
then {
let value_type = if fn_body.and_then(|body| body.params.get(index)).map_or(false, is_self) {
"self".into()
} else {
snippet(cx, decl_ty.span, "_").into()
};
span_lint_and_sugg(
cx,
TRIVIALLY_COPY_PASS_BY_REF,
input.span,
&format!("this argument ({} byte) is passed by reference, but would be more efficient if passed by value (limit: {} byte)", size, self.ref_min_size),
"consider passing by value instead",
value_type,
Applicability::Unspecified,
);
}
}
},
ty::Adt(_, _) | ty::Array(_, _) | ty::Tuple(_) => {
// if function has a body and parameter is annotated with mut, ignore
if let Some(param) = fn_body.and_then(|body| body.params.get(index)) {
match param.pat.kind {
PatKind::Binding(BindingAnnotation::Unannotated, _, _, _) => {},
_ => continue,
}
}
if_chain! {
if is_copy(cx, ty);
if !is_self_ty(input);
if let Some(size) = cx.layout_of(ty).ok().map(|l| l.size.bytes());
if size > self.value_max_size;
then {
span_lint_and_sugg(
cx,
LARGE_TYPES_PASSED_BY_VALUE,
input.span,
&format!("this argument ({} byte) is passed by value, but might be more efficient if passed by reference (limit: {} byte)", size, self.value_max_size),
"consider passing by reference instead",
format!("&{}", snippet(cx, input.span, "_")),
Applicability::MaybeIncorrect,
);
}
}
},
_ => {},
}
}
}
}
impl_lint_pass!(PassByRefOrValue => [TRIVIALLY_COPY_PASS_BY_REF, LARGE_TYPES_PASSED_BY_VALUE]);
impl<'tcx> LateLintPass<'tcx> for PassByRefOrValue {
fn check_trait_item(&mut self, cx: &LateContext<'tcx>, item: &'tcx hir::TraitItem<'_>) {
if item.span.from_expansion() {
return;
}
if let hir::TraitItemKind::Fn(method_sig, _) = &item.kind {
self.check_poly_fn(cx, item.def_id, method_sig.decl, None);
}
}
fn check_fn(
&mut self,
cx: &LateContext<'tcx>,
kind: FnKind<'tcx>,
decl: &'tcx FnDecl<'_>,
_body: &'tcx Body<'_>,
span: Span,
hir_id: HirId,
) {
if span.from_expansion() {
return;
}
match kind {
FnKind::ItemFn(.., header) => {
if header.abi != Abi::Rust {
return;
}
let attrs = cx.tcx.hir().attrs(hir_id);
for a in attrs {
if let Some(meta_items) = a.meta_item_list() {
if a.has_name(sym::proc_macro_derive)
|| (a.has_name(sym::inline) && attr::list_contains_name(&meta_items, sym::always))
{
return;
}
}
}
},
FnKind::Method(..) => (),
FnKind::Closure => return,
}
// Exclude non-inherent impls
if let Some(Node::Item(item)) = cx.tcx.hir().find(cx.tcx.hir().get_parent_node(hir_id)) {
if matches!(
item.kind,
ItemKind::Impl(Impl { of_trait: Some(_), .. }) | ItemKind::Trait(..)
) {
return;
}
}
self.check_poly_fn(cx, cx.tcx.hir().local_def_id(hir_id), decl, Some(span));
}
}