rust/compiler/rustc_mir_build/src/builder/matches/mod.rs
Stuart Cook 33c90235a1
Rollup merge of #138959 - meithecatte:matchpair-place-option, r=Zalathar
Revert "Make MatchPairTree::place non-optional"

Reverts a part of #137875. Fixes #138958.

cc `@Zalathar`
2025-03-26 19:40:31 +11:00

2866 lines
119 KiB
Rust

//! Code related to match expressions. These are sufficiently complex to
//! warrant their own module and submodules. :) This main module includes the
//! high-level algorithm, the submodules contain the details.
//!
//! This also includes code for pattern bindings in `let` statements and
//! function parameters.
use std::assert_matches::assert_matches;
use std::borrow::Borrow;
use std::mem;
use std::sync::Arc;
use rustc_abi::VariantIdx;
use rustc_data_structures::fx::FxIndexMap;
use rustc_data_structures::stack::ensure_sufficient_stack;
use rustc_hir::{BindingMode, ByRef, LetStmt, LocalSource, Node};
use rustc_middle::bug;
use rustc_middle::middle::region;
use rustc_middle::mir::{self, *};
use rustc_middle::thir::{self, *};
use rustc_middle::ty::{self, CanonicalUserTypeAnnotation, Ty};
use rustc_span::{BytePos, Pos, Span, Symbol, sym};
use tracing::{debug, instrument};
use crate::builder::ForGuard::{self, OutsideGuard, RefWithinGuard};
use crate::builder::expr::as_place::PlaceBuilder;
use crate::builder::matches::user_ty::ProjectedUserTypesNode;
use crate::builder::scope::DropKind;
use crate::builder::{
BlockAnd, BlockAndExtension, Builder, GuardFrame, GuardFrameLocal, LocalsForNode,
};
// helper functions, broken out by category:
mod match_pair;
mod test;
mod user_ty;
mod util;
/// Arguments to [`Builder::then_else_break_inner`] that are usually forwarded
/// to recursive invocations.
#[derive(Clone, Copy)]
struct ThenElseArgs {
/// Used as the temp scope for lowering `expr`. If absent (for match guards),
/// `self.local_scope()` is used.
temp_scope_override: Option<region::Scope>,
variable_source_info: SourceInfo,
/// Determines how bindings should be handled when lowering `let` expressions.
///
/// Forwarded to [`Builder::lower_let_expr`] when lowering [`ExprKind::Let`].
declare_let_bindings: DeclareLetBindings,
}
/// Should lowering a `let` expression also declare its bindings?
///
/// Used by [`Builder::lower_let_expr`] when lowering [`ExprKind::Let`].
#[derive(Clone, Copy)]
pub(crate) enum DeclareLetBindings {
/// Yes, declare `let` bindings as normal for `if` conditions.
Yes,
/// No, don't declare `let` bindings, because the caller declares them
/// separately due to special requirements.
///
/// Used for match guards and let-else.
No,
/// Let expressions are not permitted in this context, so it is a bug to
/// try to lower one (e.g inside lazy-boolean-or or boolean-not).
LetNotPermitted,
}
/// Used by [`Builder::bind_matched_candidate_for_arm_body`] to determine
/// whether or not to call [`Builder::storage_live_binding`] to emit
/// [`StatementKind::StorageLive`].
#[derive(Clone, Copy)]
pub(crate) enum EmitStorageLive {
/// Yes, emit `StorageLive` as normal.
Yes,
/// No, don't emit `StorageLive`. The caller has taken responsibility for
/// emitting `StorageLive` as appropriate.
No,
}
/// Used by [`Builder::storage_live_binding`] and [`Builder::bind_matched_candidate_for_arm_body`]
/// to decide whether to schedule drops.
#[derive(Clone, Copy, Debug)]
pub(crate) enum ScheduleDrops {
/// Yes, the relevant functions should also schedule drops as appropriate.
Yes,
/// No, don't schedule drops. The caller has taken responsibility for any
/// appropriate drops.
No,
}
impl<'a, 'tcx> Builder<'a, 'tcx> {
/// Lowers a condition in a way that ensures that variables bound in any let
/// expressions are definitely initialized in the if body.
///
/// If `declare_let_bindings` is false then variables created in `let`
/// expressions will not be declared. This is for if let guards on arms with
/// an or pattern, where the guard is lowered multiple times.
pub(crate) fn then_else_break(
&mut self,
block: BasicBlock,
expr_id: ExprId,
temp_scope_override: Option<region::Scope>,
variable_source_info: SourceInfo,
declare_let_bindings: DeclareLetBindings,
) -> BlockAnd<()> {
self.then_else_break_inner(
block,
expr_id,
ThenElseArgs { temp_scope_override, variable_source_info, declare_let_bindings },
)
}
fn then_else_break_inner(
&mut self,
block: BasicBlock, // Block that the condition and branch will be lowered into
expr_id: ExprId, // Condition expression to lower
args: ThenElseArgs,
) -> BlockAnd<()> {
let this = self;
let expr = &this.thir[expr_id];
let expr_span = expr.span;
match expr.kind {
ExprKind::LogicalOp { op: op @ LogicalOp::And, lhs, rhs } => {
this.visit_coverage_branch_operation(op, expr_span);
let lhs_then_block = this.then_else_break_inner(block, lhs, args).into_block();
let rhs_then_block =
this.then_else_break_inner(lhs_then_block, rhs, args).into_block();
rhs_then_block.unit()
}
ExprKind::LogicalOp { op: op @ LogicalOp::Or, lhs, rhs } => {
this.visit_coverage_branch_operation(op, expr_span);
let local_scope = this.local_scope();
let (lhs_success_block, failure_block) =
this.in_if_then_scope(local_scope, expr_span, |this| {
this.then_else_break_inner(
block,
lhs,
ThenElseArgs {
declare_let_bindings: DeclareLetBindings::LetNotPermitted,
..args
},
)
});
let rhs_success_block = this
.then_else_break_inner(
failure_block,
rhs,
ThenElseArgs {
declare_let_bindings: DeclareLetBindings::LetNotPermitted,
..args
},
)
.into_block();
// Make the LHS and RHS success arms converge to a common block.
// (We can't just make LHS goto RHS, because `rhs_success_block`
// might contain statements that we don't want on the LHS path.)
let success_block = this.cfg.start_new_block();
this.cfg.goto(lhs_success_block, args.variable_source_info, success_block);
this.cfg.goto(rhs_success_block, args.variable_source_info, success_block);
success_block.unit()
}
ExprKind::Unary { op: UnOp::Not, arg } => {
// Improve branch coverage instrumentation by noting conditions
// nested within one or more `!` expressions.
// (Skipped if branch coverage is not enabled.)
if let Some(coverage_info) = this.coverage_info.as_mut() {
coverage_info.visit_unary_not(this.thir, expr_id);
}
let local_scope = this.local_scope();
let (success_block, failure_block) =
this.in_if_then_scope(local_scope, expr_span, |this| {
// Help out coverage instrumentation by injecting a dummy statement with
// the original condition's span (including `!`). This fixes #115468.
if this.tcx.sess.instrument_coverage() {
this.cfg.push_coverage_span_marker(block, this.source_info(expr_span));
}
this.then_else_break_inner(
block,
arg,
ThenElseArgs {
declare_let_bindings: DeclareLetBindings::LetNotPermitted,
..args
},
)
});
this.break_for_else(success_block, args.variable_source_info);
failure_block.unit()
}
ExprKind::Scope { region_scope, lint_level, value } => {
let region_scope = (region_scope, this.source_info(expr_span));
this.in_scope(region_scope, lint_level, |this| {
this.then_else_break_inner(block, value, args)
})
}
ExprKind::Use { source } => this.then_else_break_inner(block, source, args),
ExprKind::Let { expr, ref pat } => this.lower_let_expr(
block,
expr,
pat,
Some(args.variable_source_info.scope),
args.variable_source_info.span,
args.declare_let_bindings,
EmitStorageLive::Yes,
),
_ => {
let mut block = block;
let temp_scope = args.temp_scope_override.unwrap_or_else(|| this.local_scope());
let mutability = Mutability::Mut;
// Increment the decision depth, in case we encounter boolean expressions
// further down.
this.mcdc_increment_depth_if_enabled();
let place = unpack!(
block = this.as_temp(
block,
TempLifetime {
temp_lifetime: Some(temp_scope),
backwards_incompatible: None
},
expr_id,
mutability
)
);
this.mcdc_decrement_depth_if_enabled();
let operand = Operand::Move(Place::from(place));
let then_block = this.cfg.start_new_block();
let else_block = this.cfg.start_new_block();
let term = TerminatorKind::if_(operand, then_block, else_block);
// Record branch coverage info for this condition.
// (Does nothing if branch coverage is not enabled.)
this.visit_coverage_branch_condition(expr_id, then_block, else_block);
let source_info = this.source_info(expr_span);
this.cfg.terminate(block, source_info, term);
this.break_for_else(else_block, source_info);
then_block.unit()
}
}
}
/// Generates MIR for a `match` expression.
///
/// The MIR that we generate for a match looks like this.
///
/// ```text
/// [ 0. Pre-match ]
/// |
/// [ 1. Evaluate Scrutinee (expression being matched on) ]
/// [ (PlaceMention of scrutinee) ]
/// |
/// [ 2. Decision tree -- check discriminants ] <--------+
/// | |
/// | (once a specific arm is chosen) |
/// | |
/// [pre_binding_block] [otherwise_block]
/// | |
/// [ 3. Create "guard bindings" for arm ] |
/// [ (create fake borrows) ] |
/// | |
/// [ 4. Execute guard code ] |
/// [ (read fake borrows) ] --(guard is false)-----------+
/// |
/// | (guard results in true)
/// |
/// [ 5. Create real bindings and execute arm ]
/// |
/// [ Exit match ]
/// ```
///
/// All of the different arms have been stacked on top of each other to
/// simplify the diagram. For an arm with no guard the blocks marked 3 and
/// 4 and the fake borrows are omitted.
///
/// We generate MIR in the following steps:
///
/// 1. Evaluate the scrutinee and add the PlaceMention of it ([Builder::lower_scrutinee]).
/// 2. Create the decision tree ([Builder::lower_match_tree]).
/// 3. Determine the fake borrows that are needed from the places that were
/// matched against and create the required temporaries for them
/// ([util::collect_fake_borrows]).
/// 4. Create everything else: the guards and the arms ([Builder::lower_match_arms]).
///
/// ## False edges
///
/// We don't want to have the exact structure of the decision tree be visible through borrow
/// checking. Specifically we want borrowck to think that:
/// - at any point, any or none of the patterns and guards seen so far may have been tested;
/// - after the match, any of the patterns may have matched.
///
/// For example, all of these would fail to error if borrowck could see the real CFG (examples
/// taken from `tests/ui/nll/match-cfg-fake-edges.rs`):
/// ```ignore (too many errors, this is already in the test suite)
/// let x = String::new();
/// let _ = match true {
/// _ => {},
/// _ => drop(x),
/// };
/// // Borrowck must not know the second arm is never run.
/// drop(x); //~ ERROR use of moved value
///
/// let x;
/// # let y = true;
/// match y {
/// _ if { x = 2; true } => {},
/// // Borrowck must not know the guard is always run.
/// _ => drop(x), //~ ERROR used binding `x` is possibly-uninitialized
/// };
///
/// let x = String::new();
/// # let y = true;
/// match y {
/// false if { drop(x); true } => {},
/// // Borrowck must not know the guard is not run in the `true` case.
/// true => drop(x), //~ ERROR use of moved value: `x`
/// false => {},
/// };
///
/// # let mut y = (true, true);
/// let r = &mut y.1;
/// match y {
/// //~^ ERROR cannot use `y.1` because it was mutably borrowed
/// (false, true) => {}
/// // Borrowck must not know we don't test `y.1` when `y.0` is `true`.
/// (true, _) => drop(r),
/// (false, _) => {}
/// };
/// ```
///
/// We add false edges to act as if we were naively matching each arm in order. What we need is
/// a (fake) path from each candidate to the next, specifically from candidate C's pre-binding
/// block to next candidate D's pre-binding block. For maximum precision (needed for deref
/// patterns), we choose the earliest node on D's success path that doesn't also lead to C (to
/// avoid loops).
///
/// This turns out to be easy to compute: that block is the `start_block` of the first call to
/// `match_candidates` where D is the first candidate in the list.
///
/// For example:
/// ```rust
/// # let (x, y) = (true, true);
/// match (x, y) {
/// (true, true) => 1,
/// (false, true) => 2,
/// (true, false) => 3,
/// _ => 4,
/// }
/// # ;
/// ```
/// In this example, the pre-binding block of arm 1 has a false edge to the block for result
/// `false` of the first test on `x`. The other arms have false edges to the pre-binding blocks
/// of the next arm.
///
/// On top of this, we also add a false edge from the otherwise_block of each guard to the
/// aforementioned start block of the next candidate, to ensure borrock doesn't rely on which
/// guards may have run.
#[instrument(level = "debug", skip(self, arms))]
pub(crate) fn match_expr(
&mut self,
destination: Place<'tcx>,
mut block: BasicBlock,
scrutinee_id: ExprId,
arms: &[ArmId],
span: Span,
scrutinee_span: Span,
) -> BlockAnd<()> {
let scrutinee_place =
unpack!(block = self.lower_scrutinee(block, scrutinee_id, scrutinee_span));
let match_start_span = span.shrink_to_lo().to(scrutinee_span);
let patterns = arms
.iter()
.map(|&arm| {
let arm = &self.thir[arm];
let has_match_guard =
if arm.guard.is_some() { HasMatchGuard::Yes } else { HasMatchGuard::No };
(&*arm.pattern, has_match_guard)
})
.collect();
let built_tree = self.lower_match_tree(
block,
scrutinee_span,
&scrutinee_place,
match_start_span,
patterns,
false,
);
self.lower_match_arms(
destination,
scrutinee_place,
scrutinee_span,
arms,
built_tree,
self.source_info(span),
)
}
/// Evaluate the scrutinee and add the PlaceMention for it.
fn lower_scrutinee(
&mut self,
mut block: BasicBlock,
scrutinee_id: ExprId,
scrutinee_span: Span,
) -> BlockAnd<PlaceBuilder<'tcx>> {
let scrutinee_place_builder = unpack!(block = self.as_place_builder(block, scrutinee_id));
if let Some(scrutinee_place) = scrutinee_place_builder.try_to_place(self) {
let source_info = self.source_info(scrutinee_span);
self.cfg.push_place_mention(block, source_info, scrutinee_place);
}
block.and(scrutinee_place_builder)
}
/// Lower the bindings, guards and arm bodies of a `match` expression.
///
/// The decision tree should have already been created
/// (by [Builder::lower_match_tree]).
///
/// `outer_source_info` is the SourceInfo for the whole match.
fn lower_match_arms(
&mut self,
destination: Place<'tcx>,
scrutinee_place_builder: PlaceBuilder<'tcx>,
scrutinee_span: Span,
arms: &[ArmId],
built_match_tree: BuiltMatchTree<'tcx>,
outer_source_info: SourceInfo,
) -> BlockAnd<()> {
let arm_end_blocks: Vec<BasicBlock> = arms
.iter()
.map(|&arm| &self.thir[arm])
.zip(built_match_tree.branches)
.map(|(arm, branch)| {
debug!("lowering arm {:?}\ncorresponding branch = {:?}", arm, branch);
let arm_source_info = self.source_info(arm.span);
let arm_scope = (arm.scope, arm_source_info);
let match_scope = self.local_scope();
self.in_scope(arm_scope, arm.lint_level, |this| {
let old_dedup_scope =
mem::replace(&mut this.fixed_temps_scope, Some(arm.scope));
// `try_to_place` may fail if it is unable to resolve the given
// `PlaceBuilder` inside a closure. In this case, we don't want to include
// a scrutinee place. `scrutinee_place_builder` will fail to be resolved
// if the only match arm is a wildcard (`_`).
// Example:
// ```
// let foo = (0, 1);
// let c = || {
// match foo { _ => () };
// };
// ```
let scrutinee_place = scrutinee_place_builder.try_to_place(this);
let opt_scrutinee_place =
scrutinee_place.as_ref().map(|place| (Some(place), scrutinee_span));
let scope = this.declare_bindings(
None,
arm.span,
&arm.pattern,
arm.guard,
opt_scrutinee_place,
);
let arm_block = this.bind_pattern(
outer_source_info,
branch,
&built_match_tree.fake_borrow_temps,
scrutinee_span,
Some((arm, match_scope)),
EmitStorageLive::Yes,
);
this.fixed_temps_scope = old_dedup_scope;
if let Some(source_scope) = scope {
this.source_scope = source_scope;
}
this.expr_into_dest(destination, arm_block, arm.body)
})
.into_block()
})
.collect();
// all the arm blocks will rejoin here
let end_block = self.cfg.start_new_block();
let end_brace = self.source_info(
outer_source_info.span.with_lo(outer_source_info.span.hi() - BytePos::from_usize(1)),
);
for arm_block in arm_end_blocks {
let block = &self.cfg.basic_blocks[arm_block];
let last_location = block.statements.last().map(|s| s.source_info);
self.cfg.goto(arm_block, last_location.unwrap_or(end_brace), end_block);
}
self.source_scope = outer_source_info.scope;
end_block.unit()
}
/// For a top-level `match` arm or a `let` binding, binds the variables and
/// ascribes types, and also checks the match arm guard (if present).
///
/// `arm_scope` should be `Some` if and only if this is called for a
/// `match` arm.
///
/// In the presence of or-patterns, a match arm might have multiple
/// sub-branches representing different ways to match, with each sub-branch
/// requiring its own bindings and its own copy of the guard. This method
/// handles those sub-branches individually, and then has them jump together
/// to a common block.
///
/// Returns a single block that the match arm can be lowered into.
/// (For `let` bindings, this is the code that can use the bindings.)
fn bind_pattern(
&mut self,
outer_source_info: SourceInfo,
branch: MatchTreeBranch<'tcx>,
fake_borrow_temps: &[(Place<'tcx>, Local, FakeBorrowKind)],
scrutinee_span: Span,
arm_match_scope: Option<(&Arm<'tcx>, region::Scope)>,
emit_storage_live: EmitStorageLive,
) -> BasicBlock {
if branch.sub_branches.len() == 1 {
let [sub_branch] = branch.sub_branches.try_into().unwrap();
// Avoid generating another `BasicBlock` when we only have one sub branch.
self.bind_and_guard_matched_candidate(
sub_branch,
fake_borrow_temps,
scrutinee_span,
arm_match_scope,
ScheduleDrops::Yes,
emit_storage_live,
)
} else {
// It's helpful to avoid scheduling drops multiple times to save
// drop elaboration from having to clean up the extra drops.
//
// If we are in a `let` then we only schedule drops for the first
// candidate.
//
// If we're in a `match` arm then we could have a case like so:
//
// Ok(x) | Err(x) if return => { /* ... */ }
//
// In this case we don't want a drop of `x` scheduled when we
// return: it isn't bound by move until right before enter the arm.
// To handle this we instead unschedule it's drop after each time
// we lower the guard.
let target_block = self.cfg.start_new_block();
let mut schedule_drops = ScheduleDrops::Yes;
let arm = arm_match_scope.unzip().0;
// We keep a stack of all of the bindings and type ascriptions
// from the parent candidates that we visit, that also need to
// be bound for each candidate.
for sub_branch in branch.sub_branches {
if let Some(arm) = arm {
self.clear_top_scope(arm.scope);
}
let binding_end = self.bind_and_guard_matched_candidate(
sub_branch,
fake_borrow_temps,
scrutinee_span,
arm_match_scope,
schedule_drops,
emit_storage_live,
);
if arm.is_none() {
schedule_drops = ScheduleDrops::No;
}
self.cfg.goto(binding_end, outer_source_info, target_block);
}
target_block
}
}
pub(super) fn expr_into_pattern(
&mut self,
mut block: BasicBlock,
irrefutable_pat: &Pat<'tcx>,
initializer_id: ExprId,
) -> BlockAnd<()> {
match irrefutable_pat.kind {
// Optimize the case of `let x = ...` to write directly into `x`
PatKind::Binding { mode: BindingMode(ByRef::No, _), var, subpattern: None, .. } => {
let place = self.storage_live_binding(
block,
var,
irrefutable_pat.span,
OutsideGuard,
ScheduleDrops::Yes,
);
block = self.expr_into_dest(place, block, initializer_id).into_block();
// Inject a fake read, see comments on `FakeReadCause::ForLet`.
let source_info = self.source_info(irrefutable_pat.span);
self.cfg.push_fake_read(block, source_info, FakeReadCause::ForLet(None), place);
self.schedule_drop_for_binding(var, irrefutable_pat.span, OutsideGuard);
block.unit()
}
// Optimize the case of `let x: T = ...` to write directly
// into `x` and then require that `T == typeof(x)`.
PatKind::AscribeUserType {
ref subpattern,
ascription: thir::Ascription { ref annotation, variance: _ },
} if let PatKind::Binding {
mode: BindingMode(ByRef::No, _),
var,
subpattern: None,
..
} = subpattern.kind =>
{
let place = self.storage_live_binding(
block,
var,
irrefutable_pat.span,
OutsideGuard,
ScheduleDrops::Yes,
);
block = self.expr_into_dest(place, block, initializer_id).into_block();
// Inject a fake read, see comments on `FakeReadCause::ForLet`.
let pattern_source_info = self.source_info(irrefutable_pat.span);
let cause_let = FakeReadCause::ForLet(None);
self.cfg.push_fake_read(block, pattern_source_info, cause_let, place);
let ty_source_info = self.source_info(annotation.span);
let base = self.canonical_user_type_annotations.push(annotation.clone());
self.cfg.push(
block,
Statement {
source_info: ty_source_info,
kind: StatementKind::AscribeUserType(
Box::new((place, UserTypeProjection { base, projs: Vec::new() })),
// We always use invariant as the variance here. This is because the
// variance field from the ascription refers to the variance to use
// when applying the type to the value being matched, but this
// ascription applies rather to the type of the binding. e.g., in this
// example:
//
// ```
// let x: T = <expr>
// ```
//
// We are creating an ascription that defines the type of `x` to be
// exactly `T` (i.e., with invariance). The variance field, in
// contrast, is intended to be used to relate `T` to the type of
// `<expr>`.
ty::Invariant,
),
},
);
self.schedule_drop_for_binding(var, irrefutable_pat.span, OutsideGuard);
block.unit()
}
_ => {
let initializer = &self.thir[initializer_id];
let place_builder =
unpack!(block = self.lower_scrutinee(block, initializer_id, initializer.span));
self.place_into_pattern(block, irrefutable_pat, place_builder, true)
}
}
}
pub(crate) fn place_into_pattern(
&mut self,
block: BasicBlock,
irrefutable_pat: &Pat<'tcx>,
initializer: PlaceBuilder<'tcx>,
set_match_place: bool,
) -> BlockAnd<()> {
let built_tree = self.lower_match_tree(
block,
irrefutable_pat.span,
&initializer,
irrefutable_pat.span,
vec![(irrefutable_pat, HasMatchGuard::No)],
false,
);
let [branch] = built_tree.branches.try_into().unwrap();
// For matches and function arguments, the place that is being matched
// can be set when creating the variables. But the place for
// let PATTERN = ... might not even exist until we do the assignment.
// so we set it here instead.
if set_match_place {
// `try_to_place` may fail if it is unable to resolve the given `PlaceBuilder` inside a
// closure. In this case, we don't want to include a scrutinee place.
// `scrutinee_place_builder` will fail for destructured assignments. This is because a
// closure only captures the precise places that it will read and as a result a closure
// may not capture the entire tuple/struct and rather have individual places that will
// be read in the final MIR.
// Example:
// ```
// let foo = (0, 1);
// let c = || {
// let (v1, v2) = foo;
// };
// ```
if let Some(place) = initializer.try_to_place(self) {
// Because or-alternatives bind the same variables, we only explore the first one.
let first_sub_branch = branch.sub_branches.first().unwrap();
for binding in &first_sub_branch.bindings {
let local = self.var_local_id(binding.var_id, OutsideGuard);
if let LocalInfo::User(BindingForm::Var(VarBindingForm {
opt_match_place: Some((ref mut match_place, _)),
..
})) = **self.local_decls[local].local_info.as_mut().unwrap_crate_local()
{
*match_place = Some(place);
} else {
bug!("Let binding to non-user variable.")
};
}
}
}
self.bind_pattern(
self.source_info(irrefutable_pat.span),
branch,
&[],
irrefutable_pat.span,
None,
EmitStorageLive::Yes,
)
.unit()
}
/// Declares the bindings of the given patterns and returns the visibility
/// scope for the bindings in these patterns, if such a scope had to be
/// created. NOTE: Declaring the bindings should always be done in their
/// drop scope.
#[instrument(skip(self), level = "debug")]
pub(crate) fn declare_bindings(
&mut self,
mut visibility_scope: Option<SourceScope>,
scope_span: Span,
pattern: &Pat<'tcx>,
guard: Option<ExprId>,
opt_match_place: Option<(Option<&Place<'tcx>>, Span)>,
) -> Option<SourceScope> {
self.visit_primary_bindings_special(
pattern,
&ProjectedUserTypesNode::None,
&mut |this, name, mode, var, span, ty, user_tys| {
let vis_scope = *visibility_scope
.get_or_insert_with(|| this.new_source_scope(scope_span, LintLevel::Inherited));
let source_info = SourceInfo { span, scope: this.source_scope };
let user_tys = user_tys.build_user_type_projections();
this.declare_binding(
source_info,
vis_scope,
name,
mode,
var,
ty,
user_tys,
ArmHasGuard(guard.is_some()),
opt_match_place.map(|(x, y)| (x.cloned(), y)),
pattern.span,
);
},
);
if let Some(guard_expr) = guard {
self.declare_guard_bindings(guard_expr, scope_span, visibility_scope);
}
visibility_scope
}
/// Declare bindings in a guard. This has to be done when declaring bindings
/// for an arm to ensure that or patterns only have one version of each
/// variable.
pub(crate) fn declare_guard_bindings(
&mut self,
guard_expr: ExprId,
scope_span: Span,
visibility_scope: Option<SourceScope>,
) {
match self.thir.exprs[guard_expr].kind {
ExprKind::Let { expr: _, pat: ref guard_pat } => {
// FIXME: pass a proper `opt_match_place`
self.declare_bindings(visibility_scope, scope_span, guard_pat, None, None);
}
ExprKind::Scope { value, .. } => {
self.declare_guard_bindings(value, scope_span, visibility_scope);
}
ExprKind::Use { source } => {
self.declare_guard_bindings(source, scope_span, visibility_scope);
}
ExprKind::LogicalOp { op: LogicalOp::And, lhs, rhs } => {
self.declare_guard_bindings(lhs, scope_span, visibility_scope);
self.declare_guard_bindings(rhs, scope_span, visibility_scope);
}
_ => {}
}
}
/// Emits a [`StatementKind::StorageLive`] for the given var, and also
/// schedules a drop if requested (and possible).
pub(crate) fn storage_live_binding(
&mut self,
block: BasicBlock,
var: LocalVarId,
span: Span,
for_guard: ForGuard,
schedule_drop: ScheduleDrops,
) -> Place<'tcx> {
let local_id = self.var_local_id(var, for_guard);
let source_info = self.source_info(span);
self.cfg.push(block, Statement { source_info, kind: StatementKind::StorageLive(local_id) });
// Although there is almost always scope for given variable in corner cases
// like #92893 we might get variable with no scope.
if let Some(region_scope) = self.region_scope_tree.var_scope(var.0.local_id)
&& matches!(schedule_drop, ScheduleDrops::Yes)
{
self.schedule_drop(span, region_scope, local_id, DropKind::Storage);
}
Place::from(local_id)
}
pub(crate) fn schedule_drop_for_binding(
&mut self,
var: LocalVarId,
span: Span,
for_guard: ForGuard,
) {
let local_id = self.var_local_id(var, for_guard);
if let Some(region_scope) = self.region_scope_tree.var_scope(var.0.local_id) {
self.schedule_drop(span, region_scope, local_id, DropKind::Value);
}
}
/// Visits all of the "primary" bindings in a pattern, i.e. the leftmost
/// occurrence of each variable bound by the pattern.
/// See [`PatKind::Binding::is_primary`] for more context.
///
/// This variant provides only the limited subset of binding data needed
/// by its callers, and should be a "pure" visit without side-effects.
pub(super) fn visit_primary_bindings(
&mut self,
pattern: &Pat<'tcx>,
f: &mut impl FnMut(&mut Self, LocalVarId, Span),
) {
pattern.walk_always(|pat| {
if let PatKind::Binding { var, is_primary: true, .. } = pat.kind {
f(self, var, pat.span);
}
})
}
/// Visits all of the "primary" bindings in a pattern, while preparing
/// additional user-type-annotation data needed by `declare_bindings`.
///
/// This also has the side-effect of pushing all user type annotations
/// onto `canonical_user_type_annotations`, so that they end up in MIR
/// even if they aren't associated with any bindings.
#[instrument(level = "debug", skip(self, f))]
fn visit_primary_bindings_special(
&mut self,
pattern: &Pat<'tcx>,
user_tys: &ProjectedUserTypesNode<'_>,
f: &mut impl FnMut(
&mut Self,
Symbol,
BindingMode,
LocalVarId,
Span,
Ty<'tcx>,
&ProjectedUserTypesNode<'_>,
),
) {
// Avoid having to write the full method name at each recursive call.
let visit_subpat = |this: &mut Self, subpat, user_tys: &_, f: &mut _| {
this.visit_primary_bindings_special(subpat, user_tys, f)
};
match pattern.kind {
PatKind::Binding { name, mode, var, ty, ref subpattern, is_primary, .. } => {
if is_primary {
f(self, name, mode, var, pattern.span, ty, user_tys);
}
if let Some(subpattern) = subpattern.as_ref() {
visit_subpat(self, subpattern, user_tys, f);
}
}
PatKind::Array { ref prefix, ref slice, ref suffix }
| PatKind::Slice { ref prefix, ref slice, ref suffix } => {
let from = u64::try_from(prefix.len()).unwrap();
let to = u64::try_from(suffix.len()).unwrap();
for subpattern in prefix.iter() {
visit_subpat(self, subpattern, &user_tys.index(), f);
}
if let Some(subpattern) = slice {
visit_subpat(self, subpattern, &user_tys.subslice(from, to), f);
}
for subpattern in suffix.iter() {
visit_subpat(self, subpattern, &user_tys.index(), f);
}
}
PatKind::Constant { .. }
| PatKind::Range { .. }
| PatKind::Wild
| PatKind::Never
| PatKind::Error(_) => {}
PatKind::Deref { ref subpattern } => {
visit_subpat(self, subpattern, &user_tys.deref(), f);
}
PatKind::DerefPattern { ref subpattern, .. } => {
visit_subpat(self, subpattern, &ProjectedUserTypesNode::None, f);
}
PatKind::AscribeUserType {
ref subpattern,
ascription: thir::Ascription { ref annotation, variance: _ },
} => {
// This corresponds to something like
//
// ```
// let A::<'a>(_): A<'static> = ...;
// ```
//
// Note that the variance doesn't apply here, as we are tracking the effect
// of `user_ty` on any bindings contained with subpattern.
// Caution: Pushing this user type here is load-bearing even for
// patterns containing no bindings, to ensure that the type ends
// up represented in MIR _somewhere_.
let base_user_ty = self.canonical_user_type_annotations.push(annotation.clone());
let subpattern_user_tys = user_tys.push_user_type(base_user_ty);
visit_subpat(self, subpattern, &subpattern_user_tys, f)
}
PatKind::ExpandedConstant { ref subpattern, .. } => {
visit_subpat(self, subpattern, user_tys, f)
}
PatKind::Leaf { ref subpatterns } => {
for subpattern in subpatterns {
let subpattern_user_tys = user_tys.leaf(subpattern.field);
debug!("visit_primary_bindings: subpattern_user_tys={subpattern_user_tys:?}");
visit_subpat(self, &subpattern.pattern, &subpattern_user_tys, f);
}
}
PatKind::Variant { adt_def, args: _, variant_index, ref subpatterns } => {
for subpattern in subpatterns {
let subpattern_user_tys =
user_tys.variant(adt_def, variant_index, subpattern.field);
visit_subpat(self, &subpattern.pattern, &subpattern_user_tys, f);
}
}
PatKind::Or { ref pats } => {
// In cases where we recover from errors the primary bindings
// may not all be in the leftmost subpattern. For example in
// `let (x | y) = ...`, the primary binding of `y` occurs in
// the right subpattern
for subpattern in pats.iter() {
visit_subpat(self, subpattern, user_tys, f);
}
}
}
}
}
/// Data extracted from a pattern that doesn't affect which branch is taken. Collected during
/// pattern simplification and not mutated later.
#[derive(Debug, Clone)]
struct PatternExtraData<'tcx> {
/// [`Span`] of the original pattern.
span: Span,
/// Bindings that must be established.
bindings: Vec<Binding<'tcx>>,
/// Types that must be asserted.
ascriptions: Vec<Ascription<'tcx>>,
/// Whether this corresponds to a never pattern.
is_never: bool,
}
impl<'tcx> PatternExtraData<'tcx> {
fn is_empty(&self) -> bool {
self.bindings.is_empty() && self.ascriptions.is_empty()
}
}
/// A pattern in a form suitable for lowering the match tree, with all irrefutable
/// patterns simplified away.
///
/// Here, "flat" indicates that irrefutable nodes in the pattern tree have been
/// recursively replaced with their refutable subpatterns. They are not
/// necessarily flat in an absolute sense.
///
/// Will typically be incorporated into a [`Candidate`].
#[derive(Debug, Clone)]
struct FlatPat<'tcx> {
/// To match the pattern, all of these must be satisfied...
match_pairs: Vec<MatchPairTree<'tcx>>,
extra_data: PatternExtraData<'tcx>,
}
impl<'tcx> FlatPat<'tcx> {
/// Creates a `FlatPat` containing a simplified [`MatchPairTree`] list/forest
/// for the given pattern.
fn new(place: PlaceBuilder<'tcx>, pattern: &Pat<'tcx>, cx: &mut Builder<'_, 'tcx>) -> Self {
// Recursively build a tree of match pairs for the given pattern.
let mut match_pairs = vec![];
let mut extra_data = PatternExtraData {
span: pattern.span,
bindings: Vec::new(),
ascriptions: Vec::new(),
is_never: pattern.is_never_pattern(),
};
MatchPairTree::for_pattern(place, pattern, cx, &mut match_pairs, &mut extra_data);
Self { match_pairs, extra_data }
}
}
/// Candidates are a generalization of (a) top-level match arms, and
/// (b) sub-branches of or-patterns, allowing the match-lowering process to handle
/// them both in a mostly-uniform way. For example, the list of candidates passed
/// to [`Builder::match_candidates`] will often contain a mixture of top-level
/// candidates and or-pattern subcandidates.
///
/// At the start of match lowering, there is one candidate for each match arm.
/// During match lowering, arms with or-patterns will be expanded into a tree
/// of candidates, where each "leaf" candidate represents one of the ways for
/// the arm pattern to successfully match.
#[derive(Debug)]
struct Candidate<'tcx> {
/// For the candidate to match, all of these must be satisfied...
///
/// ---
/// Initially contains a list of match pairs created by [`FlatPat`], but is
/// subsequently mutated (in a queue-like way) while lowering the match tree.
/// When this list becomes empty, the candidate is fully matched and becomes
/// a leaf (see [`Builder::select_matched_candidate`]).
///
/// Key mutations include:
///
/// - When a match pair is fully satisfied by a test, it is removed from the
/// list, and its subpairs are added instead (see [`Builder::sort_candidate`]).
/// - During or-pattern expansion, any leading or-pattern is removed, and is
/// converted into subcandidates (see [`Builder::expand_and_match_or_candidates`]).
/// - After a candidate's subcandidates have been lowered, a copy of any remaining
/// or-patterns is added to each leaf subcandidate
/// (see [`Builder::test_remaining_match_pairs_after_or`]).
///
/// Invariants:
/// - All or-patterns ([`TestCase::Or`]) have been sorted to the end.
match_pairs: Vec<MatchPairTree<'tcx>>,
/// ...and if this is non-empty, one of these subcandidates also has to match...
///
/// ---
/// Initially a candidate has no subcandidates; they are added (and then immediately
/// lowered) during or-pattern expansion. Their main function is to serve as _output_
/// of match tree lowering, allowing later steps to see the leaf candidates that
/// represent a match of the entire match arm.
///
/// A candidate no subcandidates is either incomplete (if it has match pairs left),
/// or is a leaf in the match tree. A candidate with one or more subcandidates is
/// an internal node in the match tree.
///
/// Invariant: at the end of match tree lowering, this must not contain an
/// `is_never` candidate, because that would break binding consistency.
/// - See [`Builder::remove_never_subcandidates`].
subcandidates: Vec<Candidate<'tcx>>,
/// ...and if there is a guard it must be evaluated; if it's `false` then branch to `otherwise_block`.
///
/// ---
/// For subcandidates, this is copied from the parent candidate, so it indicates
/// whether the enclosing match arm has a guard.
has_guard: bool,
/// Holds extra pattern data that was prepared by [`FlatPat`], including bindings and
/// ascriptions that must be established if this candidate succeeds.
extra_data: PatternExtraData<'tcx>,
/// When setting `self.subcandidates`, we store here the span of the or-pattern they came from.
///
/// ---
/// Invariant: it is `None` iff `subcandidates.is_empty()`.
/// - FIXME: We sometimes don't unset this when clearing `subcandidates`.
or_span: Option<Span>,
/// The block before the `bindings` have been established.
///
/// After the match tree has been lowered, [`Builder::lower_match_arms`]
/// will use this as the start point for lowering bindings and guards, and
/// then jump to a shared block containing the arm body.
pre_binding_block: Option<BasicBlock>,
/// The block to branch to if the guard or a nested candidate fails to match.
otherwise_block: Option<BasicBlock>,
/// The earliest block that has only candidates >= this one as descendents. Used for false
/// edges, see the doc for [`Builder::match_expr`].
false_edge_start_block: Option<BasicBlock>,
}
impl<'tcx> Candidate<'tcx> {
fn new(
place: PlaceBuilder<'tcx>,
pattern: &Pat<'tcx>,
has_guard: HasMatchGuard,
cx: &mut Builder<'_, 'tcx>,
) -> Self {
// Use `FlatPat` to build simplified match pairs, then immediately
// incorporate them into a new candidate.
Self::from_flat_pat(
FlatPat::new(place, pattern, cx),
matches!(has_guard, HasMatchGuard::Yes),
)
}
/// Incorporates an already-simplified [`FlatPat`] into a new candidate.
fn from_flat_pat(flat_pat: FlatPat<'tcx>, has_guard: bool) -> Self {
let mut this = Candidate {
match_pairs: flat_pat.match_pairs,
extra_data: flat_pat.extra_data,
has_guard,
subcandidates: Vec::new(),
or_span: None,
otherwise_block: None,
pre_binding_block: None,
false_edge_start_block: None,
};
this.sort_match_pairs();
this
}
/// Restores the invariant that or-patterns must be sorted to the end.
fn sort_match_pairs(&mut self) {
self.match_pairs.sort_by_key(|pair| matches!(pair.test_case, TestCase::Or { .. }));
}
/// Returns whether the first match pair of this candidate is an or-pattern.
fn starts_with_or_pattern(&self) -> bool {
matches!(&*self.match_pairs, [MatchPairTree { test_case: TestCase::Or { .. }, .. }, ..])
}
/// Visit the leaf candidates (those with no subcandidates) contained in
/// this candidate.
fn visit_leaves<'a>(&'a mut self, mut visit_leaf: impl FnMut(&'a mut Self)) {
traverse_candidate(
self,
&mut (),
&mut move |c, _| visit_leaf(c),
move |c, _| c.subcandidates.iter_mut(),
|_| {},
);
}
/// Visit the leaf candidates in reverse order.
fn visit_leaves_rev<'a>(&'a mut self, mut visit_leaf: impl FnMut(&'a mut Self)) {
traverse_candidate(
self,
&mut (),
&mut move |c, _| visit_leaf(c),
move |c, _| c.subcandidates.iter_mut().rev(),
|_| {},
);
}
}
/// A depth-first traversal of the `Candidate` and all of its recursive
/// subcandidates.
///
/// This signature is very generic, to support traversing candidate trees by
/// reference or by value, and to allow a mutable "context" to be shared by the
/// traversal callbacks. Most traversals can use the simpler
/// [`Candidate::visit_leaves`] wrapper instead.
fn traverse_candidate<'tcx, C, T, I>(
candidate: C,
context: &mut T,
// Called when visiting a "leaf" candidate (with no subcandidates).
visit_leaf: &mut impl FnMut(C, &mut T),
// Called when visiting a "node" candidate (with one or more subcandidates).
// Returns an iterator over the candidate's children (by value or reference).
// Can perform setup before visiting the node's children.
get_children: impl Copy + Fn(C, &mut T) -> I,
// Called after visiting a "node" candidate's children.
complete_children: impl Copy + Fn(&mut T),
) where
C: Borrow<Candidate<'tcx>>, // Typically `Candidate` or `&mut Candidate`
I: Iterator<Item = C>,
{
if candidate.borrow().subcandidates.is_empty() {
visit_leaf(candidate, context)
} else {
for child in get_children(candidate, context) {
traverse_candidate(child, context, visit_leaf, get_children, complete_children);
}
complete_children(context)
}
}
#[derive(Clone, Debug)]
struct Binding<'tcx> {
span: Span,
source: Place<'tcx>,
var_id: LocalVarId,
binding_mode: BindingMode,
}
/// Indicates that the type of `source` must be a subtype of the
/// user-given type `user_ty`; this is basically a no-op but can
/// influence region inference.
#[derive(Clone, Debug)]
struct Ascription<'tcx> {
source: Place<'tcx>,
annotation: CanonicalUserTypeAnnotation<'tcx>,
variance: ty::Variance,
}
/// Partial summary of a [`thir::Pat`], indicating what sort of test should be
/// performed to match/reject the pattern, and what the desired test outcome is.
/// This avoids having to perform a full match on [`thir::PatKind`] in some places,
/// and helps [`TestKind::Switch`] and [`TestKind::SwitchInt`] know what target
/// values to use.
///
/// Created by [`MatchPairTree::for_pattern`], and then inspected primarily by:
/// - [`Builder::pick_test_for_match_pair`] (to choose a test)
/// - [`Builder::sort_candidate`] (to see how the test interacts with a match pair)
///
/// Note that or-patterns are not tested directly like the other variants.
/// Instead they participate in or-pattern expansion, where they are transformed into
/// subcandidates. See [`Builder::expand_and_match_or_candidates`].
#[derive(Debug, Clone)]
enum TestCase<'tcx> {
Variant { adt_def: ty::AdtDef<'tcx>, variant_index: VariantIdx },
Constant { value: mir::Const<'tcx> },
Range(Arc<PatRange<'tcx>>),
Slice { len: usize, variable_length: bool },
Deref { temp: Place<'tcx>, mutability: Mutability },
Never,
Or { pats: Box<[FlatPat<'tcx>]> },
}
impl<'tcx> TestCase<'tcx> {
fn as_range(&self) -> Option<&PatRange<'tcx>> {
if let Self::Range(v) = self { Some(v.as_ref()) } else { None }
}
}
/// Node in a tree of "match pairs", where each pair consists of a place to be
/// tested, and a test to perform on that place.
///
/// Each node also has a list of subpairs (possibly empty) that must also match,
/// and a reference to the THIR pattern it represents.
#[derive(Debug, Clone)]
pub(crate) struct MatchPairTree<'tcx> {
/// This place...
///
/// ---
/// This can be `None` if it referred to a non-captured place in a closure.
///
/// Invariant: Can only be `None` when `test_case` is `Or`.
/// Therefore this must be `Some(_)` after or-pattern expansion.
place: Option<Place<'tcx>>,
/// ... must pass this test...
test_case: TestCase<'tcx>,
/// ... and these subpairs must match.
///
/// ---
/// Subpairs typically represent tests that can only be performed after their
/// parent has succeeded. For example, the pattern `Some(3)` might have an
/// outer match pair that tests for the variant `Some`, and then a subpair
/// that tests its field for the value `3`.
subpairs: Vec<Self>,
/// Type field of the pattern this node was created from.
pattern_ty: Ty<'tcx>,
/// Span field of the pattern this node was created from.
pattern_span: Span,
}
/// See [`Test`] for more.
#[derive(Clone, Debug, PartialEq)]
enum TestKind<'tcx> {
/// Test what enum variant a value is.
///
/// The subset of expected variants is not stored here; instead they are
/// extracted from the [`TestCase`]s of the candidates participating in the
/// test.
Switch {
/// The enum type being tested.
adt_def: ty::AdtDef<'tcx>,
},
/// Test what value an integer or `char` has.
///
/// The test's target values are not stored here; instead they are extracted
/// from the [`TestCase`]s of the candidates participating in the test.
SwitchInt,
/// Test whether a `bool` is `true` or `false`.
If,
/// Test for equality with value, possibly after an unsizing coercion to
/// `ty`,
Eq {
value: Const<'tcx>,
// Integer types are handled by `SwitchInt`, and constants with ADT
// types and `&[T]` types are converted back into patterns, so this can
// only be `&str`, `f32` or `f64`.
ty: Ty<'tcx>,
},
/// Test whether the value falls within an inclusive or exclusive range.
Range(Arc<PatRange<'tcx>>),
/// Test that the length of the slice is `== len` or `>= len`.
Len { len: u64, op: BinOp },
/// Call `Deref::deref[_mut]` on the value.
Deref {
/// Temporary to store the result of `deref()`/`deref_mut()`.
temp: Place<'tcx>,
mutability: Mutability,
},
/// Assert unreachability of never patterns.
Never,
}
/// A test to perform to determine which [`Candidate`] matches a value.
///
/// [`Test`] is just the test to perform; it does not include the value
/// to be tested.
#[derive(Debug)]
pub(crate) struct Test<'tcx> {
span: Span,
kind: TestKind<'tcx>,
}
/// The branch to be taken after a test.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash)]
enum TestBranch<'tcx> {
/// Success branch, used for tests with two possible outcomes.
Success,
/// Branch corresponding to this constant.
Constant(Const<'tcx>, u128),
/// Branch corresponding to this variant.
Variant(VariantIdx),
/// Failure branch for tests with two possible outcomes, and "otherwise" branch for other tests.
Failure,
}
impl<'tcx> TestBranch<'tcx> {
fn as_constant(&self) -> Option<&Const<'tcx>> {
if let Self::Constant(v, _) = self { Some(v) } else { None }
}
}
/// `ArmHasGuard` is a wrapper around a boolean flag. It indicates whether
/// a match arm has a guard expression attached to it.
#[derive(Copy, Clone, Debug)]
pub(crate) struct ArmHasGuard(pub(crate) bool);
///////////////////////////////////////////////////////////////////////////
// Main matching algorithm
/// A sub-branch in the output of match lowering. Match lowering has generated MIR code that will
/// branch to `success_block` when the matched value matches the corresponding pattern. If there is
/// a guard, its failure must continue to `otherwise_block`, which will resume testing patterns.
#[derive(Debug)]
struct MatchTreeSubBranch<'tcx> {
span: Span,
/// The block that is branched to if the corresponding subpattern matches.
success_block: BasicBlock,
/// The block to branch to if this arm had a guard and the guard fails.
otherwise_block: BasicBlock,
/// The bindings to set up in this sub-branch.
bindings: Vec<Binding<'tcx>>,
/// The ascriptions to set up in this sub-branch.
ascriptions: Vec<Ascription<'tcx>>,
/// Whether the sub-branch corresponds to a never pattern.
is_never: bool,
}
/// A branch in the output of match lowering.
#[derive(Debug)]
struct MatchTreeBranch<'tcx> {
sub_branches: Vec<MatchTreeSubBranch<'tcx>>,
}
/// The result of generating MIR for a pattern-matching expression. Each input branch/arm/pattern
/// gives rise to an output `MatchTreeBranch`. If one of the patterns matches, we branch to the
/// corresponding `success_block`. If none of the patterns matches, we branch to `otherwise_block`.
///
/// Each branch is made of one of more sub-branches, corresponding to or-patterns. E.g.
/// ```ignore(illustrative)
/// match foo {
/// (x, false) | (false, x) => {}
/// (true, true) => {}
/// }
/// ```
/// Here the first arm gives the first `MatchTreeBranch`, which has two sub-branches, one for each
/// alternative of the or-pattern. They are kept separate because each needs to bind `x` to a
/// different place.
#[derive(Debug)]
struct BuiltMatchTree<'tcx> {
branches: Vec<MatchTreeBranch<'tcx>>,
otherwise_block: BasicBlock,
/// If any of the branches had a guard, we collect here the places and locals to fakely borrow
/// to ensure match guards can't modify the values as we match them. For more details, see
/// [`util::collect_fake_borrows`].
fake_borrow_temps: Vec<(Place<'tcx>, Local, FakeBorrowKind)>,
}
impl<'tcx> MatchTreeSubBranch<'tcx> {
fn from_sub_candidate(
candidate: Candidate<'tcx>,
parent_data: &Vec<PatternExtraData<'tcx>>,
) -> Self {
debug_assert!(candidate.match_pairs.is_empty());
MatchTreeSubBranch {
span: candidate.extra_data.span,
success_block: candidate.pre_binding_block.unwrap(),
otherwise_block: candidate.otherwise_block.unwrap(),
bindings: parent_data
.iter()
.flat_map(|d| &d.bindings)
.chain(&candidate.extra_data.bindings)
.cloned()
.collect(),
ascriptions: parent_data
.iter()
.flat_map(|d| &d.ascriptions)
.cloned()
.chain(candidate.extra_data.ascriptions)
.collect(),
is_never: candidate.extra_data.is_never,
}
}
}
impl<'tcx> MatchTreeBranch<'tcx> {
fn from_candidate(candidate: Candidate<'tcx>) -> Self {
let mut sub_branches = Vec::new();
traverse_candidate(
candidate,
&mut Vec::new(),
&mut |candidate: Candidate<'_>, parent_data: &mut Vec<PatternExtraData<'_>>| {
sub_branches.push(MatchTreeSubBranch::from_sub_candidate(candidate, parent_data));
},
|inner_candidate, parent_data| {
parent_data.push(inner_candidate.extra_data);
inner_candidate.subcandidates.into_iter()
},
|parent_data| {
parent_data.pop();
},
);
MatchTreeBranch { sub_branches }
}
}
#[derive(Debug, Clone, Copy, PartialEq, Eq)]
enum HasMatchGuard {
Yes,
No,
}
impl<'a, 'tcx> Builder<'a, 'tcx> {
/// The entrypoint of the matching algorithm. Create the decision tree for the match expression,
/// starting from `block`.
///
/// `patterns` is a list of patterns, one for each arm. The associated boolean indicates whether
/// the arm has a guard.
///
/// `refutable` indicates whether the candidate list is refutable (for `if let` and `let else`)
/// or not (for `let` and `match`). In the refutable case we return the block to which we branch
/// on failure.
fn lower_match_tree(
&mut self,
block: BasicBlock,
scrutinee_span: Span,
scrutinee_place_builder: &PlaceBuilder<'tcx>,
match_start_span: Span,
patterns: Vec<(&Pat<'tcx>, HasMatchGuard)>,
refutable: bool,
) -> BuiltMatchTree<'tcx> {
// Assemble the initial list of candidates. These top-level candidates are 1:1 with the
// input patterns, but other parts of match lowering also introduce subcandidates (for
// sub-or-patterns). So inside the algorithm, the candidates list may not correspond to
// match arms directly.
let mut candidates: Vec<Candidate<'_>> = patterns
.into_iter()
.map(|(pat, has_guard)| {
Candidate::new(scrutinee_place_builder.clone(), pat, has_guard, self)
})
.collect();
let fake_borrow_temps = util::collect_fake_borrows(
self,
&candidates,
scrutinee_span,
scrutinee_place_builder.base(),
);
// This will generate code to test scrutinee_place and branch to the appropriate arm block.
// If none of the arms match, we branch to `otherwise_block`. When lowering a `match`
// expression, exhaustiveness checking ensures that this block is unreachable.
let mut candidate_refs = candidates.iter_mut().collect::<Vec<_>>();
let otherwise_block =
self.match_candidates(match_start_span, scrutinee_span, block, &mut candidate_refs);
// Set up false edges so that the borrow-checker cannot make use of the specific CFG we
// generated. We falsely branch from each candidate to the one below it to make it as if we
// were testing match branches one by one in order. In the refutable case we also want a
// false edge to the final failure block.
let mut next_candidate_start_block = if refutable { Some(otherwise_block) } else { None };
for candidate in candidates.iter_mut().rev() {
let has_guard = candidate.has_guard;
candidate.visit_leaves_rev(|leaf_candidate| {
if let Some(next_candidate_start_block) = next_candidate_start_block {
let source_info = self.source_info(leaf_candidate.extra_data.span);
// Falsely branch to `next_candidate_start_block` before reaching pre_binding.
let old_pre_binding = leaf_candidate.pre_binding_block.unwrap();
let new_pre_binding = self.cfg.start_new_block();
self.false_edges(
old_pre_binding,
new_pre_binding,
next_candidate_start_block,
source_info,
);
leaf_candidate.pre_binding_block = Some(new_pre_binding);
if has_guard {
// Falsely branch to `next_candidate_start_block` also if the guard fails.
let new_otherwise = self.cfg.start_new_block();
let old_otherwise = leaf_candidate.otherwise_block.unwrap();
self.false_edges(
new_otherwise,
old_otherwise,
next_candidate_start_block,
source_info,
);
leaf_candidate.otherwise_block = Some(new_otherwise);
}
}
assert!(leaf_candidate.false_edge_start_block.is_some());
next_candidate_start_block = leaf_candidate.false_edge_start_block;
});
}
if !refutable {
// Match checking ensures `otherwise_block` is actually unreachable in irrefutable
// cases.
let source_info = self.source_info(scrutinee_span);
// Matching on a scrutinee place of an uninhabited type doesn't generate any memory
// reads by itself, and so if the place is uninitialized we wouldn't know. In order to
// disallow the following:
// ```rust
// let x: !;
// match x {}
// ```
// we add a dummy read on the place.
//
// NOTE: If we require never patterns for empty matches, those will check that the place
// is initialized, and so this read would no longer be needed.
let cause_matched_place = FakeReadCause::ForMatchedPlace(None);
if let Some(scrutinee_place) = scrutinee_place_builder.try_to_place(self) {
self.cfg.push_fake_read(
otherwise_block,
source_info,
cause_matched_place,
scrutinee_place,
);
}
self.cfg.terminate(otherwise_block, source_info, TerminatorKind::Unreachable);
}
BuiltMatchTree {
branches: candidates.into_iter().map(MatchTreeBranch::from_candidate).collect(),
otherwise_block,
fake_borrow_temps,
}
}
/// The main match algorithm. It begins with a set of candidates `candidates` and has the job of
/// generating code that branches to an appropriate block if the scrutinee matches one of these
/// candidates. The
/// candidates are ordered such that the first item in the list
/// has the highest priority. When a candidate is found to match
/// the value, we will set and generate a branch to the appropriate
/// pre-binding block.
///
/// If none of the candidates apply, we continue to the returned `otherwise_block`.
///
/// Note that while `match` expressions in the Rust language are exhaustive,
/// candidate lists passed to this method are often _non-exhaustive_.
/// For example, the match lowering process will frequently divide up the
/// list of candidates, and recursively call this method with a non-exhaustive
/// subset of candidates.
/// See [`Builder::test_candidates`] for more details on this
/// "backtracking automata" approach.
///
/// For an example of how we use `otherwise_block`, consider:
/// ```
/// # fn foo((x, y): (bool, bool)) -> u32 {
/// match (x, y) {
/// (true, true) => 1,
/// (_, false) => 2,
/// (false, true) => 3,
/// }
/// # }
/// ```
/// For this match, we generate something like:
/// ```
/// # fn foo((x, y): (bool, bool)) -> u32 {
/// if x {
/// if y {
/// return 1
/// } else {
/// // continue
/// }
/// } else {
/// // continue
/// }
/// if y {
/// if x {
/// // This is actually unreachable because the `(true, true)` case was handled above,
/// // but we don't know that from within the lowering algorithm.
/// // continue
/// } else {
/// return 3
/// }
/// } else {
/// return 2
/// }
/// // this is the final `otherwise_block`, which is unreachable because the match was exhaustive.
/// unreachable!()
/// # }
/// ```
///
/// Every `continue` is an instance of branching to some `otherwise_block` somewhere deep within
/// the algorithm. For more details on why we lower like this, see [`Builder::test_candidates`].
///
/// Note how we test `x` twice. This is the tradeoff of backtracking automata: we prefer smaller
/// code size so we accept non-optimal code paths.
#[instrument(skip(self), level = "debug")]
fn match_candidates(
&mut self,
span: Span,
scrutinee_span: Span,
start_block: BasicBlock,
candidates: &mut [&mut Candidate<'tcx>],
) -> BasicBlock {
ensure_sufficient_stack(|| {
self.match_candidates_inner(span, scrutinee_span, start_block, candidates)
})
}
/// Construct the decision tree for `candidates`. Don't call this, call `match_candidates`
/// instead to reserve sufficient stack space.
fn match_candidates_inner(
&mut self,
span: Span,
scrutinee_span: Span,
mut start_block: BasicBlock,
candidates: &mut [&mut Candidate<'tcx>],
) -> BasicBlock {
if let [first, ..] = candidates {
if first.false_edge_start_block.is_none() {
first.false_edge_start_block = Some(start_block);
}
}
// Process a prefix of the candidates.
let rest = match candidates {
[] => {
// If there are no candidates that still need testing, we're done.
return start_block;
}
[first, remaining @ ..] if first.match_pairs.is_empty() => {
// The first candidate has satisfied all its match pairs.
// We record the blocks that will be needed by match arm lowering,
// and then continue with the remaining candidates.
let remainder_start = self.select_matched_candidate(first, start_block);
remainder_start.and(remaining)
}
candidates if candidates.iter().any(|candidate| candidate.starts_with_or_pattern()) => {
// If any candidate starts with an or-pattern, we want to expand or-patterns
// before we do any more tests.
//
// The only candidate we strictly _need_ to expand here is the first one.
// But by expanding other candidates as early as possible, we unlock more
// opportunities to include them in test outcomes, making the match tree
// smaller and simpler.
self.expand_and_match_or_candidates(span, scrutinee_span, start_block, candidates)
}
candidates => {
// The first candidate has some unsatisfied match pairs; we proceed to do more tests.
self.test_candidates(span, scrutinee_span, candidates, start_block)
}
};
// Process any candidates that remain.
let remaining_candidates = unpack!(start_block = rest);
self.match_candidates(span, scrutinee_span, start_block, remaining_candidates)
}
/// Link up matched candidates.
///
/// For example, if we have something like this:
///
/// ```ignore (illustrative)
/// ...
/// Some(x) if cond1 => ...
/// Some(x) => ...
/// Some(x) if cond2 => ...
/// ...
/// ```
///
/// We generate real edges from:
///
/// * `start_block` to the [pre-binding block] of the first pattern,
/// * the [otherwise block] of the first pattern to the second pattern,
/// * the [otherwise block] of the third pattern to a block with an
/// [`Unreachable` terminator](TerminatorKind::Unreachable).
///
/// In addition, we later add fake edges from the otherwise blocks to the
/// pre-binding block of the next candidate in the original set of
/// candidates.
///
/// [pre-binding block]: Candidate::pre_binding_block
/// [otherwise block]: Candidate::otherwise_block
fn select_matched_candidate(
&mut self,
candidate: &mut Candidate<'tcx>,
start_block: BasicBlock,
) -> BasicBlock {
assert!(candidate.otherwise_block.is_none());
assert!(candidate.pre_binding_block.is_none());
assert!(candidate.subcandidates.is_empty());
candidate.pre_binding_block = Some(start_block);
let otherwise_block = self.cfg.start_new_block();
// Create the otherwise block for this candidate, which is the
// pre-binding block for the next candidate.
candidate.otherwise_block = Some(otherwise_block);
otherwise_block
}
/// Takes a list of candidates such that some of the candidates' first match pairs are
/// or-patterns. This expands as many or-patterns as possible and processes the resulting
/// candidates. Returns the unprocessed candidates if any.
fn expand_and_match_or_candidates<'b, 'c>(
&mut self,
span: Span,
scrutinee_span: Span,
start_block: BasicBlock,
candidates: &'b mut [&'c mut Candidate<'tcx>],
) -> BlockAnd<&'b mut [&'c mut Candidate<'tcx>]> {
// We can't expand or-patterns freely. The rule is:
// - If a candidate doesn't start with an or-pattern, we include it in
// the expansion list as-is (i.e. it "expands" to itself).
// - If a candidate has an or-pattern as its only remaining match pair,
// we can expand it.
// - If it starts with an or-pattern but also has other match pairs,
// we can expand it, but we can't process more candidates after it.
//
// If we didn't stop, the `otherwise` cases could get mixed up. E.g. in the
// following, or-pattern simplification (in `merge_trivial_subcandidates`) makes it
// so the `1` and `2` cases branch to a same block (which then tests `false`). If we
// took `(2, _)` in the same set of candidates, when we reach the block that tests
// `false` we don't know whether we came from `1` or `2`, hence we can't know where
// to branch on failure.
//
// ```ignore(illustrative)
// match (1, true) {
// (1 | 2, false) => {},
// (2, _) => {},
// _ => {}
// }
// ```
//
// We therefore split the `candidates` slice in two, expand or-patterns in the first part,
// and process the rest separately.
let expand_until = candidates
.iter()
.position(|candidate| {
// If a candidate starts with an or-pattern and has more match pairs,
// we can expand it, but we must stop expanding _after_ it.
candidate.match_pairs.len() > 1 && candidate.starts_with_or_pattern()
})
.map(|pos| pos + 1) // Stop _after_ the found candidate
.unwrap_or(candidates.len()); // Otherwise, include all candidates
let (candidates_to_expand, remaining_candidates) = candidates.split_at_mut(expand_until);
// Expand one level of or-patterns for each candidate in `candidates_to_expand`.
// We take care to preserve the relative ordering of candidates, so that
// or-patterns are expanded in their parent's relative position.
let mut expanded_candidates = Vec::new();
for candidate in candidates_to_expand.iter_mut() {
if candidate.starts_with_or_pattern() {
let or_match_pair = candidate.match_pairs.remove(0);
// Expand the or-pattern into subcandidates.
self.create_or_subcandidates(candidate, or_match_pair);
// Collect the newly created subcandidates.
for subcandidate in candidate.subcandidates.iter_mut() {
expanded_candidates.push(subcandidate);
}
// Note that the subcandidates have been added to `expanded_candidates`,
// but `candidate` itself has not. If the last candidate has more match pairs,
// they are handled separately by `test_remaining_match_pairs_after_or`.
} else {
// A candidate that doesn't start with an or-pattern has nothing to
// expand, so it is included in the post-expansion list as-is.
expanded_candidates.push(candidate);
}
}
// Recursively lower the part of the match tree represented by the
// expanded candidates. This is where subcandidates actually get lowered!
let remainder_start = self.match_candidates(
span,
scrutinee_span,
start_block,
expanded_candidates.as_mut_slice(),
);
// Postprocess subcandidates, and process any leftover match pairs.
// (Only the last candidate can possibly have more match pairs.)
debug_assert!({
let mut all_except_last = candidates_to_expand.iter().rev().skip(1);
all_except_last.all(|candidate| candidate.match_pairs.is_empty())
});
for candidate in candidates_to_expand.iter_mut() {
if !candidate.subcandidates.is_empty() {
self.merge_trivial_subcandidates(candidate);
self.remove_never_subcandidates(candidate);
}
}
// It's important to perform the above simplifications _before_ dealing
// with remaining match pairs, to avoid exponential blowup if possible
// (for trivial or-patterns), and avoid useless work (for never patterns).
if let Some(last_candidate) = candidates_to_expand.last_mut() {
self.test_remaining_match_pairs_after_or(span, scrutinee_span, last_candidate);
}
remainder_start.and(remaining_candidates)
}
/// Given a match-pair that corresponds to an or-pattern, expand each subpattern into a new
/// subcandidate. Any candidate that has been expanded this way should also be postprocessed
/// at the end of [`Self::expand_and_match_or_candidates`].
fn create_or_subcandidates(
&mut self,
candidate: &mut Candidate<'tcx>,
match_pair: MatchPairTree<'tcx>,
) {
let TestCase::Or { pats } = match_pair.test_case else { bug!() };
debug!("expanding or-pattern: candidate={:#?}\npats={:#?}", candidate, pats);
candidate.or_span = Some(match_pair.pattern_span);
candidate.subcandidates = pats
.into_vec()
.into_iter()
.map(|flat_pat| Candidate::from_flat_pat(flat_pat, candidate.has_guard))
.collect();
candidate.subcandidates[0].false_edge_start_block = candidate.false_edge_start_block;
}
/// Try to merge all of the subcandidates of the given candidate into one. This avoids
/// exponentially large CFGs in cases like `(1 | 2, 3 | 4, ...)`. The candidate should have been
/// expanded with `create_or_subcandidates`.
///
/// Given a pattern `(P | Q, R | S)` we (in principle) generate a CFG like
/// so:
///
/// ```text
/// [ start ]
/// |
/// [ match P, Q ]
/// |
/// +----------------------------------------+------------------------------------+
/// | | |
/// V V V
/// [ P matches ] [ Q matches ] [ otherwise ]
/// | | |
/// V V |
/// [ match R, S ] [ match R, S ] |
/// | | |
/// +--------------+------------+ +--------------+------------+ |
/// | | | | | | |
/// V V V V V V |
/// [ R matches ] [ S matches ] [otherwise ] [ R matches ] [ S matches ] [otherwise ] |
/// | | | | | | |
/// +--------------+------------|------------+--------------+ | |
/// | | | |
/// | +----------------------------------------+--------+
/// | |
/// V V
/// [ Success ] [ Failure ]
/// ```
///
/// In practice there are some complications:
///
/// * If there's a guard, then the otherwise branch of the first match on
/// `R | S` goes to a test for whether `Q` matches, and the control flow
/// doesn't merge into a single success block until after the guard is
/// tested.
/// * If neither `P` or `Q` has any bindings or type ascriptions and there
/// isn't a match guard, then we create a smaller CFG like:
///
/// ```text
/// ...
/// +---------------+------------+
/// | | |
/// [ P matches ] [ Q matches ] [ otherwise ]
/// | | |
/// +---------------+ |
/// | ...
/// [ match R, S ]
/// |
/// ...
/// ```
///
/// Note that this takes place _after_ the subcandidates have participated
/// in match tree lowering.
fn merge_trivial_subcandidates(&mut self, candidate: &mut Candidate<'tcx>) {
assert!(!candidate.subcandidates.is_empty());
if candidate.has_guard {
// FIXME(or_patterns; matthewjasper) Don't give up if we have a guard.
return;
}
// FIXME(or_patterns; matthewjasper) Try to be more aggressive here.
let can_merge = candidate.subcandidates.iter().all(|subcandidate| {
subcandidate.subcandidates.is_empty() && subcandidate.extra_data.is_empty()
});
if !can_merge {
return;
}
let mut last_otherwise = None;
let shared_pre_binding_block = self.cfg.start_new_block();
// This candidate is about to become a leaf, so unset `or_span`.
let or_span = candidate.or_span.take().unwrap();
let source_info = self.source_info(or_span);
if candidate.false_edge_start_block.is_none() {
candidate.false_edge_start_block = candidate.subcandidates[0].false_edge_start_block;
}
// Remove the (known-trivial) subcandidates from the candidate tree,
// so that they aren't visible after match tree lowering, and wire them
// all to join up at a single shared pre-binding block.
// (Note that the subcandidates have already had their part of the match
// tree lowered by this point, which is why we can add a goto to them.)
for subcandidate in mem::take(&mut candidate.subcandidates) {
let subcandidate_block = subcandidate.pre_binding_block.unwrap();
self.cfg.goto(subcandidate_block, source_info, shared_pre_binding_block);
last_otherwise = subcandidate.otherwise_block;
}
candidate.pre_binding_block = Some(shared_pre_binding_block);
assert!(last_otherwise.is_some());
candidate.otherwise_block = last_otherwise;
}
/// Never subcandidates may have a set of bindings inconsistent with their siblings,
/// which would break later code. So we filter them out. Note that we can't filter out
/// top-level candidates this way.
fn remove_never_subcandidates(&mut self, candidate: &mut Candidate<'tcx>) {
if candidate.subcandidates.is_empty() {
return;
}
let false_edge_start_block = candidate.subcandidates[0].false_edge_start_block;
candidate.subcandidates.retain_mut(|candidate| {
if candidate.extra_data.is_never {
candidate.visit_leaves(|subcandidate| {
let block = subcandidate.pre_binding_block.unwrap();
// That block is already unreachable but needs a terminator to make the MIR well-formed.
let source_info = self.source_info(subcandidate.extra_data.span);
self.cfg.terminate(block, source_info, TerminatorKind::Unreachable);
});
false
} else {
true
}
});
if candidate.subcandidates.is_empty() {
// If `candidate` has become a leaf candidate, ensure it has a `pre_binding_block` and `otherwise_block`.
let next_block = self.cfg.start_new_block();
candidate.pre_binding_block = Some(next_block);
candidate.otherwise_block = Some(next_block);
// In addition, if `candidate` doesn't have `false_edge_start_block`, it should be assigned here.
if candidate.false_edge_start_block.is_none() {
candidate.false_edge_start_block = false_edge_start_block;
}
}
}
/// If more match pairs remain, test them after each subcandidate.
/// We could have added them to the or-candidates during or-pattern expansion, but that
/// would make it impossible to detect simplifiable or-patterns. That would guarantee
/// exponentially large CFGs for cases like `(1 | 2, 3 | 4, ...)`.
fn test_remaining_match_pairs_after_or(
&mut self,
span: Span,
scrutinee_span: Span,
candidate: &mut Candidate<'tcx>,
) {
if candidate.match_pairs.is_empty() {
return;
}
let or_span = candidate.or_span.unwrap_or(candidate.extra_data.span);
let source_info = self.source_info(or_span);
let mut last_otherwise = None;
candidate.visit_leaves(|leaf_candidate| {
last_otherwise = leaf_candidate.otherwise_block;
});
let remaining_match_pairs = mem::take(&mut candidate.match_pairs);
// We're testing match pairs that remained after an `Or`, so the remaining
// pairs should all be `Or` too, due to the sorting invariant.
debug_assert!(
remaining_match_pairs
.iter()
.all(|match_pair| matches!(match_pair.test_case, TestCase::Or { .. }))
);
// Visit each leaf candidate within this subtree, add a copy of the remaining
// match pairs to it, and then recursively lower the rest of the match tree
// from that point.
candidate.visit_leaves(|leaf_candidate| {
// At this point the leaf's own match pairs have all been lowered
// and removed, so `extend` and assignment are equivalent,
// but extending can also recycle any existing vector capacity.
assert!(leaf_candidate.match_pairs.is_empty());
leaf_candidate.match_pairs.extend(remaining_match_pairs.iter().cloned());
let or_start = leaf_candidate.pre_binding_block.unwrap();
let otherwise =
self.match_candidates(span, scrutinee_span, or_start, &mut [leaf_candidate]);
// In a case like `(P | Q, R | S)`, if `P` succeeds and `R | S` fails, we know `(Q,
// R | S)` will fail too. If there is no guard, we skip testing of `Q` by branching
// directly to `last_otherwise`. If there is a guard,
// `leaf_candidate.otherwise_block` can be reached by guard failure as well, so we
// can't skip `Q`.
let or_otherwise = if leaf_candidate.has_guard {
leaf_candidate.otherwise_block.unwrap()
} else {
last_otherwise.unwrap()
};
self.cfg.goto(otherwise, source_info, or_otherwise);
});
}
/// Pick a test to run. Which test doesn't matter as long as it is guaranteed to fully match at
/// least one match pair. We currently simply pick the test corresponding to the first match
/// pair of the first candidate in the list.
///
/// *Note:* taking the first match pair is somewhat arbitrary, and we might do better here by
/// choosing more carefully what to test.
///
/// For example, consider the following possible match-pairs:
///
/// 1. `x @ Some(P)` -- we will do a [`Switch`] to decide what variant `x` has
/// 2. `x @ 22` -- we will do a [`SwitchInt`] to decide what value `x` has
/// 3. `x @ 3..5` -- we will do a [`Range`] test to decide what range `x` falls in
/// 4. etc.
///
/// [`Switch`]: TestKind::Switch
/// [`SwitchInt`]: TestKind::SwitchInt
/// [`Range`]: TestKind::Range
fn pick_test(&mut self, candidates: &[&mut Candidate<'tcx>]) -> (Place<'tcx>, Test<'tcx>) {
// Extract the match-pair from the highest priority candidate
let match_pair = &candidates[0].match_pairs[0];
let test = self.pick_test_for_match_pair(match_pair);
// Unwrap is ok after simplification.
let match_place = match_pair.place.unwrap();
debug!(?test, ?match_pair);
(match_place, test)
}
/// Given a test, we partition the input candidates into several buckets.
/// If a candidate matches in exactly one of the branches of `test`
/// (and no other branches), we put it into the corresponding bucket.
/// If it could match in more than one of the branches of `test`, the test
/// doesn't usefully apply to it, and we stop partitioning candidates.
///
/// Importantly, we also **mutate** the branched candidates to remove match pairs
/// that are entailed by the outcome of the test, and add any sub-pairs of the
/// removed pairs.
///
/// This returns a pair of
/// - the candidates that weren't sorted;
/// - for each possible outcome of the test, the candidates that match in that outcome.
///
/// For example:
/// ```
/// # let (x, y, z) = (true, true, true);
/// match (x, y, z) {
/// (true , _ , true ) => true, // (0)
/// (false, false, _ ) => false, // (1)
/// (_ , true , _ ) => true, // (2)
/// (true , _ , false) => false, // (3)
/// }
/// # ;
/// ```
///
/// Assume we are testing on `x`. Conceptually, there are 2 overlapping candidate sets:
/// - If the outcome is that `x` is true, candidates {0, 2, 3} are possible
/// - If the outcome is that `x` is false, candidates {1, 2} are possible
///
/// Following our algorithm:
/// - Candidate 0 is sorted into outcome `x == true`
/// - Candidate 1 is sorted into outcome `x == false`
/// - Candidate 2 remains unsorted, because testing `x` has no effect on it
/// - Candidate 3 remains unsorted, because a previous candidate (2) was unsorted
/// - This helps preserve the illusion that candidates are tested "in order"
///
/// The sorted candidates are mutated to remove entailed match pairs:
/// - candidate 0 becomes `[z @ true]` since we know that `x` was `true`;
/// - candidate 1 becomes `[y @ false]` since we know that `x` was `false`.
fn sort_candidates<'b, 'c>(
&mut self,
match_place: Place<'tcx>,
test: &Test<'tcx>,
mut candidates: &'b mut [&'c mut Candidate<'tcx>],
) -> (
&'b mut [&'c mut Candidate<'tcx>],
FxIndexMap<TestBranch<'tcx>, Vec<&'b mut Candidate<'tcx>>>,
) {
// For each of the possible outcomes, collect vector of candidates that apply if the test
// has that particular outcome.
let mut target_candidates: FxIndexMap<_, Vec<&mut Candidate<'_>>> = Default::default();
let total_candidate_count = candidates.len();
// Sort the candidates into the appropriate vector in `target_candidates`. Note that at some
// point we may encounter a candidate where the test is not relevant; at that point, we stop
// sorting.
while let Some(candidate) = candidates.first_mut() {
let Some(branch) =
self.sort_candidate(match_place, test, candidate, &target_candidates)
else {
break;
};
let (candidate, rest) = candidates.split_first_mut().unwrap();
target_candidates.entry(branch).or_insert_with(Vec::new).push(candidate);
candidates = rest;
}
// At least the first candidate ought to be tested
assert!(
total_candidate_count > candidates.len(),
"{total_candidate_count}, {candidates:#?}"
);
debug!("tested_candidates: {}", total_candidate_count - candidates.len());
debug!("untested_candidates: {}", candidates.len());
(candidates, target_candidates)
}
/// This is the most subtle part of the match lowering algorithm. At this point, there are
/// no fully-satisfied candidates, and no or-patterns to expand, so we actually need to
/// perform some sort of test to make progress.
///
/// Once we pick what sort of test we are going to perform, this test will help us winnow down
/// our candidates. So we walk over the candidates (from high to low priority) and check. We
/// compute, for each outcome of the test, a list of (modified) candidates. If a candidate
/// matches in exactly one branch of our test, we add it to the corresponding outcome. We also
/// **mutate its list of match pairs** if appropriate, to reflect the fact that we know which
/// outcome occurred.
///
/// For example, if we are testing `x.0`'s variant, and we have a candidate `(x.0 @ Some(v), x.1
/// @ 22)`, then we would have a resulting candidate of `((x.0 as Some).0 @ v, x.1 @ 22)` in the
/// branch corresponding to `Some`. To ensure we make progress, we always pick a test that
/// results in simplifying the first candidate.
///
/// But there may also be candidates that the test doesn't
/// apply to. The classical example is wildcards:
///
/// ```
/// # let (x, y, z) = (true, true, true);
/// match (x, y, z) {
/// (true , _ , true ) => true, // (0)
/// (false, false, _ ) => false, // (1)
/// (_ , true , _ ) => true, // (2)
/// (true , _ , false) => false, // (3)
/// }
/// # ;
/// ```
///
/// Here, the traditional "decision tree" method would generate 2 separate code-paths for the 2
/// possible values of `x`. This would however duplicate some candidates, which would need to be
/// lowered several times.
///
/// In some cases, this duplication can create an exponential amount of
/// code. This is most easily seen by noticing that this method terminates
/// with precisely the reachable arms being reachable - but that problem
/// is trivially NP-complete:
///
/// ```ignore (illustrative)
/// match (var0, var1, var2, var3, ...) {
/// (true , _ , _ , false, true, ...) => false,
/// (_ , true, true , false, _ , ...) => false,
/// (false, _ , false, false, _ , ...) => false,
/// ...
/// _ => true
/// }
/// ```
///
/// Here the last arm is reachable only if there is an assignment to
/// the variables that does not match any of the literals. Therefore,
/// compilation would take an exponential amount of time in some cases.
///
/// In rustc, we opt instead for the "backtracking automaton" approach. This guarantees we never
/// duplicate a candidate (except in the presence of or-patterns). In fact this guarantee is
/// ensured by the fact that we carry around `&mut Candidate`s which can't be duplicated.
///
/// To make this work, whenever we decide to perform a test, if we encounter a candidate that
/// could match in more than one branch of the test, we stop. We generate code for the test and
/// for the candidates in its branches; the remaining candidates will be tested if the
/// candidates in the branches fail to match.
///
/// For example, if we test on `x` in the following:
/// ```
/// # fn foo((x, y, z): (bool, bool, bool)) -> u32 {
/// match (x, y, z) {
/// (true , _ , true ) => 0,
/// (false, false, _ ) => 1,
/// (_ , true , _ ) => 2,
/// (true , _ , false) => 3,
/// }
/// # }
/// ```
/// this function generates code that looks more of less like:
/// ```
/// # fn foo((x, y, z): (bool, bool, bool)) -> u32 {
/// if x {
/// match (y, z) {
/// (_, true) => return 0,
/// _ => {} // continue matching
/// }
/// } else {
/// match (y, z) {
/// (false, _) => return 1,
/// _ => {} // continue matching
/// }
/// }
/// // the block here is `remainder_start`
/// match (x, y, z) {
/// (_ , true , _ ) => 2,
/// (true , _ , false) => 3,
/// _ => unreachable!(),
/// }
/// # }
/// ```
///
/// We return the unprocessed candidates.
fn test_candidates<'b, 'c>(
&mut self,
span: Span,
scrutinee_span: Span,
candidates: &'b mut [&'c mut Candidate<'tcx>],
start_block: BasicBlock,
) -> BlockAnd<&'b mut [&'c mut Candidate<'tcx>]> {
// Choose a match pair from the first candidate, and use it to determine a
// test to perform that will confirm or refute that match pair.
let (match_place, test) = self.pick_test(candidates);
// For each of the N possible test outcomes, build the vector of candidates that applies if
// the test has that particular outcome. This also mutates the candidates to remove match
// pairs that are fully satisfied by the relevant outcome.
let (remaining_candidates, target_candidates) =
self.sort_candidates(match_place, &test, candidates);
// The block that we should branch to if none of the `target_candidates` match.
let remainder_start = self.cfg.start_new_block();
// For each outcome of the test, recursively lower the rest of the match tree
// from that point. (Note that we haven't lowered the actual test yet!)
let target_blocks: FxIndexMap<_, _> = target_candidates
.into_iter()
.map(|(branch, mut candidates)| {
let branch_start = self.cfg.start_new_block();
// Recursively lower the rest of the match tree after the relevant outcome.
let branch_otherwise =
self.match_candidates(span, scrutinee_span, branch_start, &mut *candidates);
// Link up the `otherwise` block of the subtree to `remainder_start`.
let source_info = self.source_info(span);
self.cfg.goto(branch_otherwise, source_info, remainder_start);
(branch, branch_start)
})
.collect();
// Perform the chosen test, branching to one of the N subtrees prepared above
// (or to `remainder_start` if no outcome was satisfied).
self.perform_test(
span,
scrutinee_span,
start_block,
remainder_start,
match_place,
&test,
target_blocks,
);
remainder_start.and(remaining_candidates)
}
}
///////////////////////////////////////////////////////////////////////////
// Pat binding - used for `let` and function parameters as well.
impl<'a, 'tcx> Builder<'a, 'tcx> {
/// Lowers a `let` expression that appears in a suitable context
/// (e.g. an `if` condition or match guard).
///
/// Also used for lowering let-else statements, since they have similar
/// needs despite not actually using `let` expressions.
///
/// Use [`DeclareLetBindings`] to control whether the `let` bindings are
/// declared or not.
pub(crate) fn lower_let_expr(
&mut self,
mut block: BasicBlock,
expr_id: ExprId,
pat: &Pat<'tcx>,
source_scope: Option<SourceScope>,
scope_span: Span,
declare_let_bindings: DeclareLetBindings,
emit_storage_live: EmitStorageLive,
) -> BlockAnd<()> {
let expr_span = self.thir[expr_id].span;
let scrutinee = unpack!(block = self.lower_scrutinee(block, expr_id, expr_span));
let built_tree = self.lower_match_tree(
block,
expr_span,
&scrutinee,
pat.span,
vec![(pat, HasMatchGuard::No)],
true,
);
let [branch] = built_tree.branches.try_into().unwrap();
self.break_for_else(built_tree.otherwise_block, self.source_info(expr_span));
match declare_let_bindings {
DeclareLetBindings::Yes => {
let expr_place = scrutinee.try_to_place(self);
let opt_expr_place = expr_place.as_ref().map(|place| (Some(place), expr_span));
self.declare_bindings(
source_scope,
pat.span.to(scope_span),
pat,
None,
opt_expr_place,
);
}
DeclareLetBindings::No => {} // Caller is responsible for bindings.
DeclareLetBindings::LetNotPermitted => {
self.tcx.dcx().span_bug(expr_span, "let expression not expected in this context")
}
}
let success = self.bind_pattern(
self.source_info(pat.span),
branch,
&[],
expr_span,
None,
emit_storage_live,
);
// If branch coverage is enabled, record this branch.
self.visit_coverage_conditional_let(pat, success, built_tree.otherwise_block);
success.unit()
}
/// Initializes each of the bindings from the candidate by
/// moving/copying/ref'ing the source as appropriate. Tests the guard, if
/// any, and then branches to the arm. Returns the block for the case where
/// the guard succeeds.
///
/// Note: we do not check earlier that if there is a guard,
/// there cannot be move bindings. We avoid a use-after-move by only
/// moving the binding once the guard has evaluated to true (see below).
fn bind_and_guard_matched_candidate(
&mut self,
sub_branch: MatchTreeSubBranch<'tcx>,
fake_borrows: &[(Place<'tcx>, Local, FakeBorrowKind)],
scrutinee_span: Span,
arm_match_scope: Option<(&Arm<'tcx>, region::Scope)>,
schedule_drops: ScheduleDrops,
emit_storage_live: EmitStorageLive,
) -> BasicBlock {
debug!("bind_and_guard_matched_candidate(subbranch={:?})", sub_branch);
let block = sub_branch.success_block;
if sub_branch.is_never {
// This arm has a dummy body, we don't need to generate code for it. `block` is already
// unreachable (except via false edge).
let source_info = self.source_info(sub_branch.span);
self.cfg.terminate(block, source_info, TerminatorKind::Unreachable);
return self.cfg.start_new_block();
}
self.ascribe_types(block, sub_branch.ascriptions);
// Lower an instance of the arm guard (if present) for this candidate,
// and then perform bindings for the arm body.
if let Some((arm, match_scope)) = arm_match_scope
&& let Some(guard) = arm.guard
{
let tcx = self.tcx;
// Bindings for guards require some extra handling to automatically
// insert implicit references/dereferences.
self.bind_matched_candidate_for_guard(
block,
schedule_drops,
sub_branch.bindings.iter(),
);
let guard_frame = GuardFrame {
locals: sub_branch
.bindings
.iter()
.map(|b| GuardFrameLocal::new(b.var_id))
.collect(),
};
debug!("entering guard building context: {:?}", guard_frame);
self.guard_context.push(guard_frame);
let re_erased = tcx.lifetimes.re_erased;
let scrutinee_source_info = self.source_info(scrutinee_span);
for &(place, temp, kind) in fake_borrows {
let borrow = Rvalue::Ref(re_erased, BorrowKind::Fake(kind), place);
self.cfg.push_assign(block, scrutinee_source_info, Place::from(temp), borrow);
}
let mut guard_span = rustc_span::DUMMY_SP;
let (post_guard_block, otherwise_post_guard_block) =
self.in_if_then_scope(match_scope, guard_span, |this| {
guard_span = this.thir[guard].span;
this.then_else_break(
block,
guard,
None, // Use `self.local_scope()` as the temp scope
this.source_info(arm.span),
DeclareLetBindings::No, // For guards, `let` bindings are declared separately
)
});
let source_info = self.source_info(guard_span);
let guard_end = self.source_info(tcx.sess.source_map().end_point(guard_span));
let guard_frame = self.guard_context.pop().unwrap();
debug!("Exiting guard building context with locals: {:?}", guard_frame);
for &(_, temp, _) in fake_borrows {
let cause = FakeReadCause::ForMatchGuard;
self.cfg.push_fake_read(post_guard_block, guard_end, cause, Place::from(temp));
}
self.cfg.goto(otherwise_post_guard_block, source_info, sub_branch.otherwise_block);
// We want to ensure that the matched candidates are bound
// after we have confirmed this candidate *and* any
// associated guard; Binding them on `block` is too soon,
// because that would be before we've checked the result
// from the guard.
//
// But binding them on the arm is *too late*, because
// then all of the candidates for a single arm would be
// bound in the same place, that would cause a case like:
//
// ```rust
// match (30, 2) {
// (mut x, 1) | (2, mut x) if { true } => { ... }
// ... // ^^^^^^^ (this is `arm_block`)
// }
// ```
//
// would yield an `arm_block` something like:
//
// ```
// StorageLive(_4); // _4 is `x`
// _4 = &mut (_1.0: i32); // this is handling `(mut x, 1)` case
// _4 = &mut (_1.1: i32); // this is handling `(2, mut x)` case
// ```
//
// and that is clearly not correct.
let by_value_bindings = sub_branch
.bindings
.iter()
.filter(|binding| matches!(binding.binding_mode.0, ByRef::No));
// Read all of the by reference bindings to ensure that the
// place they refer to can't be modified by the guard.
for binding in by_value_bindings.clone() {
let local_id = self.var_local_id(binding.var_id, RefWithinGuard);
let cause = FakeReadCause::ForGuardBinding;
self.cfg.push_fake_read(post_guard_block, guard_end, cause, Place::from(local_id));
}
assert_matches!(
schedule_drops,
ScheduleDrops::Yes,
"patterns with guards must schedule drops"
);
self.bind_matched_candidate_for_arm_body(
post_guard_block,
ScheduleDrops::Yes,
by_value_bindings,
emit_storage_live,
);
post_guard_block
} else {
// (Here, it is not too early to bind the matched
// candidate on `block`, because there is no guard result
// that we have to inspect before we bind them.)
self.bind_matched_candidate_for_arm_body(
block,
schedule_drops,
sub_branch.bindings.iter(),
emit_storage_live,
);
block
}
}
/// Append `AscribeUserType` statements onto the end of `block`
/// for each ascription
fn ascribe_types(
&mut self,
block: BasicBlock,
ascriptions: impl IntoIterator<Item = Ascription<'tcx>>,
) {
for ascription in ascriptions {
let source_info = self.source_info(ascription.annotation.span);
let base = self.canonical_user_type_annotations.push(ascription.annotation);
self.cfg.push(
block,
Statement {
source_info,
kind: StatementKind::AscribeUserType(
Box::new((
ascription.source,
UserTypeProjection { base, projs: Vec::new() },
)),
ascription.variance,
),
},
);
}
}
/// Binding for guards is a bit different from binding for the arm body,
/// because an extra layer of implicit reference/dereference is added.
///
/// The idea is that any pattern bindings of type T will map to a `&T` within
/// the context of the guard expression, but will continue to map to a `T`
/// in the context of the arm body. To avoid surfacing this distinction in
/// the user source code (which would be a severe change to the language and
/// require far more revision to the compiler), any occurrence of the
/// identifier in the guard expression will automatically get a deref op
/// applied to it. (See the caller of [`Self::is_bound_var_in_guard`].)
///
/// So an input like:
///
/// ```ignore (illustrative)
/// let place = Foo::new();
/// match place { foo if inspect(foo)
/// => feed(foo), ... }
/// ```
///
/// will be treated as if it were really something like:
///
/// ```ignore (illustrative)
/// let place = Foo::new();
/// match place { Foo { .. } if { let tmp1 = &place; inspect(*tmp1) }
/// => { let tmp2 = place; feed(tmp2) }, ... }
/// ```
///
/// And an input like:
///
/// ```ignore (illustrative)
/// let place = Foo::new();
/// match place { ref mut foo if inspect(foo)
/// => feed(foo), ... }
/// ```
///
/// will be treated as if it were really something like:
///
/// ```ignore (illustrative)
/// let place = Foo::new();
/// match place { Foo { .. } if { let tmp1 = & &mut place; inspect(*tmp1) }
/// => { let tmp2 = &mut place; feed(tmp2) }, ... }
/// ```
/// ---
///
/// ## Implementation notes
///
/// To encode the distinction above, we must inject the
/// temporaries `tmp1` and `tmp2`.
///
/// There are two cases of interest: binding by-value, and binding by-ref.
///
/// 1. Binding by-value: Things are simple.
///
/// * Establishing `tmp1` creates a reference into the
/// matched place. This code is emitted by
/// [`Self::bind_matched_candidate_for_guard`].
///
/// * `tmp2` is only initialized "lazily", after we have
/// checked the guard. Thus, the code that can trigger
/// moves out of the candidate can only fire after the
/// guard evaluated to true. This initialization code is
/// emitted by [`Self::bind_matched_candidate_for_arm_body`].
///
/// 2. Binding by-reference: Things are tricky.
///
/// * Here, the guard expression wants a `&&` or `&&mut`
/// into the original input. This means we need to borrow
/// the reference that we create for the arm.
/// * So we eagerly create the reference for the arm and then take a
/// reference to that.
///
/// ---
///
/// See these PRs for some historical context:
/// - <https://github.com/rust-lang/rust/pull/49870> (introduction of autoref)
/// - <https://github.com/rust-lang/rust/pull/59114> (always use autoref)
fn bind_matched_candidate_for_guard<'b>(
&mut self,
block: BasicBlock,
schedule_drops: ScheduleDrops,
bindings: impl IntoIterator<Item = &'b Binding<'tcx>>,
) where
'tcx: 'b,
{
debug!("bind_matched_candidate_for_guard(block={:?})", block);
// Assign each of the bindings. Since we are binding for a
// guard expression, this will never trigger moves out of the
// candidate.
let re_erased = self.tcx.lifetimes.re_erased;
for binding in bindings {
debug!("bind_matched_candidate_for_guard(binding={:?})", binding);
let source_info = self.source_info(binding.span);
// For each pattern ident P of type T, `ref_for_guard` is
// a reference R: &T pointing to the location matched by
// the pattern, and every occurrence of P within a guard
// denotes *R.
let ref_for_guard = self.storage_live_binding(
block,
binding.var_id,
binding.span,
RefWithinGuard,
schedule_drops,
);
match binding.binding_mode.0 {
ByRef::No => {
// The arm binding will be by value, so for the guard binding
// just take a shared reference to the matched place.
let rvalue = Rvalue::Ref(re_erased, BorrowKind::Shared, binding.source);
self.cfg.push_assign(block, source_info, ref_for_guard, rvalue);
}
ByRef::Yes(mutbl) => {
// The arm binding will be by reference, so eagerly create it now.
let value_for_arm = self.storage_live_binding(
block,
binding.var_id,
binding.span,
OutsideGuard,
schedule_drops,
);
let rvalue =
Rvalue::Ref(re_erased, util::ref_pat_borrow_kind(mutbl), binding.source);
self.cfg.push_assign(block, source_info, value_for_arm, rvalue);
// For the guard binding, take a shared reference to that reference.
let rvalue = Rvalue::Ref(re_erased, BorrowKind::Shared, value_for_arm);
self.cfg.push_assign(block, source_info, ref_for_guard, rvalue);
}
}
}
}
fn bind_matched_candidate_for_arm_body<'b>(
&mut self,
block: BasicBlock,
schedule_drops: ScheduleDrops,
bindings: impl IntoIterator<Item = &'b Binding<'tcx>>,
emit_storage_live: EmitStorageLive,
) where
'tcx: 'b,
{
debug!("bind_matched_candidate_for_arm_body(block={:?})", block);
let re_erased = self.tcx.lifetimes.re_erased;
// Assign each of the bindings. This may trigger moves out of the candidate.
for binding in bindings {
let source_info = self.source_info(binding.span);
let local = match emit_storage_live {
// Here storages are already alive, probably because this is a binding
// from let-else.
// We just need to schedule drop for the value.
EmitStorageLive::No => self.var_local_id(binding.var_id, OutsideGuard).into(),
EmitStorageLive::Yes => self.storage_live_binding(
block,
binding.var_id,
binding.span,
OutsideGuard,
schedule_drops,
),
};
if matches!(schedule_drops, ScheduleDrops::Yes) {
self.schedule_drop_for_binding(binding.var_id, binding.span, OutsideGuard);
}
let rvalue = match binding.binding_mode.0 {
ByRef::No => Rvalue::Use(self.consume_by_copy_or_move(binding.source)),
ByRef::Yes(mutbl) => {
Rvalue::Ref(re_erased, util::ref_pat_borrow_kind(mutbl), binding.source)
}
};
self.cfg.push_assign(block, source_info, local, rvalue);
}
}
/// Each binding (`ref mut var`/`ref var`/`mut var`/`var`, where the bound
/// `var` has type `T` in the arm body) in a pattern maps to 2 locals. The
/// first local is a binding for occurrences of `var` in the guard, which
/// will have type `&T`. The second local is a binding for occurrences of
/// `var` in the arm body, which will have type `T`.
#[instrument(skip(self), level = "debug")]
fn declare_binding(
&mut self,
source_info: SourceInfo,
visibility_scope: SourceScope,
name: Symbol,
mode: BindingMode,
var_id: LocalVarId,
var_ty: Ty<'tcx>,
user_ty: Option<Box<UserTypeProjections>>,
has_guard: ArmHasGuard,
opt_match_place: Option<(Option<Place<'tcx>>, Span)>,
pat_span: Span,
) {
let tcx = self.tcx;
let debug_source_info = SourceInfo { span: source_info.span, scope: visibility_scope };
let local = LocalDecl {
mutability: mode.1,
ty: var_ty,
user_ty,
source_info,
local_info: ClearCrossCrate::Set(Box::new(LocalInfo::User(BindingForm::Var(
VarBindingForm {
binding_mode: mode,
// hypothetically, `visit_primary_bindings` could try to unzip
// an outermost hir::Ty as we descend, matching up
// idents in pat; but complex w/ unclear UI payoff.
// Instead, just abandon providing diagnostic info.
opt_ty_info: None,
opt_match_place,
pat_span,
},
)))),
};
let for_arm_body = self.local_decls.push(local);
if self.should_emit_debug_info_for_binding(name, var_id) {
self.var_debug_info.push(VarDebugInfo {
name,
source_info: debug_source_info,
value: VarDebugInfoContents::Place(for_arm_body.into()),
composite: None,
argument_index: None,
});
}
let locals = if has_guard.0 {
let ref_for_guard = self.local_decls.push(LocalDecl::<'tcx> {
// This variable isn't mutated but has a name, so has to be
// immutable to avoid the unused mut lint.
mutability: Mutability::Not,
ty: Ty::new_imm_ref(tcx, tcx.lifetimes.re_erased, var_ty),
user_ty: None,
source_info,
local_info: ClearCrossCrate::Set(Box::new(LocalInfo::User(
BindingForm::RefForGuard,
))),
});
if self.should_emit_debug_info_for_binding(name, var_id) {
self.var_debug_info.push(VarDebugInfo {
name,
source_info: debug_source_info,
value: VarDebugInfoContents::Place(ref_for_guard.into()),
composite: None,
argument_index: None,
});
}
LocalsForNode::ForGuard { ref_for_guard, for_arm_body }
} else {
LocalsForNode::One(for_arm_body)
};
debug!(?locals);
self.var_indices.insert(var_id, locals);
}
/// Some bindings are introduced when producing HIR from the AST and don't
/// actually exist in the source. Skip producing debug info for those when
/// we can recognize them.
fn should_emit_debug_info_for_binding(&self, name: Symbol, var_id: LocalVarId) -> bool {
// For now we only recognize the output of desugaring assigns.
if name != sym::lhs {
return true;
}
let tcx = self.tcx;
for (_, node) in tcx.hir_parent_iter(var_id.0) {
// FIXME(khuey) at what point is it safe to bail on the iterator?
// Can we stop at the first non-Pat node?
if matches!(node, Node::LetStmt(&LetStmt { source: LocalSource::AssignDesugar(_), .. }))
{
return false;
}
}
true
}
}