1 //! Note: most of the tests relevant to this file can be found (at the time of writing) in
2 //! src/tests/ui/pattern/usefulness.
4 //! This file includes the logic for exhaustiveness and usefulness checking for
5 //! pattern-matching. Specifically, given a list of patterns for a type, we can
7 //! (a) the patterns cover every possible constructor for the type (exhaustiveness)
8 //! (b) each pattern is necessary (usefulness)
10 //! The algorithm implemented here is a modified version of the one described in:
11 //! http://moscova.inria.fr/~maranget/papers/warn/index.html
12 //! However, to save future implementors from reading the original paper, we
13 //! summarise the algorithm here to hopefully save time and be a little clearer
14 //! (without being so rigorous).
18 //! The core of the algorithm revolves about a "usefulness" check. In particular, we
19 //! are trying to compute a predicate `U(P, p)` where `P` is a list of patterns (we refer to this as
20 //! a matrix). `U(P, p)` represents whether, given an existing list of patterns
21 //! `P_1 ..= P_m`, adding a new pattern `p` will be "useful" (that is, cover previously-
22 //! uncovered values of the type).
24 //! If we have this predicate, then we can easily compute both exhaustiveness of an
25 //! entire set of patterns and the individual usefulness of each one.
26 //! (a) the set of patterns is exhaustive iff `U(P, _)` is false (i.e., adding a wildcard
27 //! match doesn't increase the number of values we're matching)
28 //! (b) a pattern `P_i` is not useful if `U(P[0..=(i-1), P_i)` is false (i.e., adding a
29 //! pattern to those that have come before it doesn't increase the number of values
34 //! The idea that powers everything that is done in this file is the following: a value is made
35 //! from a constructor applied to some fields. Examples of constructors are `Some`, `None`, `(,)`
36 //! (the 2-tuple constructor), `Foo {..}` (the constructor for a struct `Foo`), and `2` (the
37 //! constructor for the number `2`). Fields are just a (possibly empty) list of values.
39 //! Some of the constructors listed above might feel weird: `None` and `2` don't take any
40 //! arguments. This is part of what makes constructors so general: we will consider plain values
41 //! like numbers and string literals to be constructors that take no arguments, also called "0-ary
42 //! constructors"; they are the simplest case of constructors. This allows us to see any value as
43 //! made up from a tree of constructors, each having a given number of children. For example:
44 //! `(None, Ok(0))` is made from 4 different constructors.
46 //! This idea can be extended to patterns: a pattern captures a set of possible values, and we can
47 //! describe this set using constructors. For example, `Err(_)` captures all values of the type
48 //! `Result<T, E>` that start with the `Err` constructor (for some choice of `T` and `E`). The
49 //! wildcard `_` captures all values of the given type starting with any of the constructors for
52 //! We use this to compute whether different patterns might capture a same value. Do the patterns
53 //! `Ok("foo")` and `Err(_)` capture a common value? The answer is no, because the first pattern
54 //! captures only values starting with the `Ok` constructor and the second only values starting
55 //! with the `Err` constructor. Do the patterns `Some(42)` and `Some(1..10)` intersect? They might,
56 //! since they both capture values starting with `Some`. To be certain, we need to dig under the
57 //! `Some` constructor and continue asking the question. This is the main idea behind the
58 //! exhaustiveness algorithm: by looking at patterns constructor-by-constructor, we can efficiently
59 //! figure out if some new pattern might capture a value that hadn't been captured by previous
62 //! Constructors are represented by the `Constructor` enum, and its fields by the `Fields` enum.
63 //! Most of the complexity of this file resides in transforming between patterns and
64 //! (`Constructor`, `Fields`) pairs, handling all the special cases correctly.
66 //! Caveat: this constructors/fields distinction doesn't quite cover every Rust value. For example
67 //! a value of type `Rc<u64>` doesn't fit this idea very well, nor do various other things.
68 //! However, this idea covers most of the cases that are relevant to exhaustiveness checking.
73 //! Recall that `U(P, p)` represents whether, given an existing list of patterns (aka matrix) `P`,
74 //! adding a new pattern `p` will cover previously-uncovered values of the type.
75 //! During the course of the algorithm, the rows of the matrix won't just be individual patterns,
76 //! but rather partially-deconstructed patterns in the form of a list of fields. The paper
77 //! calls those pattern-vectors, and we will call them pattern-stacks. The same holds for the
80 //! For example, say we have the following:
83 //! // x: (Option<bool>, Result<()>)
85 //! (Some(true), _) => {}
86 //! (None, Err(())) => {}
87 //! (None, Err(_)) => {}
91 //! Here, the matrix `P` starts as:
95 //! [(Some(true), _)],
96 //! [(None, Err(()))],
101 //! We can tell it's not exhaustive, because `U(P, _)` is true (we're not covering
102 //! `[(Some(false), _)]`, for instance). In addition, row 3 is not useful, because
103 //! all the values it covers are already covered by row 2.
105 //! A list of patterns can be thought of as a stack, because we are mainly interested in the top of
106 //! the stack at any given point, and we can pop or apply constructors to get new pattern-stacks.
107 //! To match the paper, the top of the stack is at the beginning / on the left.
109 //! There are two important operations on pattern-stacks necessary to understand the algorithm:
111 //! 1. We can pop a given constructor off the top of a stack. This operation is called
112 //! `specialize`, and is denoted `S(c, p)` where `c` is a constructor (like `Some` or
113 //! `None`) and `p` a pattern-stack.
114 //! If the pattern on top of the stack can cover `c`, this removes the constructor and
115 //! pushes its arguments onto the stack. It also expands OR-patterns into distinct patterns.
116 //! Otherwise the pattern-stack is discarded.
117 //! This essentially filters those pattern-stacks whose top covers the constructor `c` and
118 //! discards the others.
120 //! For example, the first pattern above initially gives a stack `[(Some(true), _)]`. If we
121 //! pop the tuple constructor, we are left with `[Some(true), _]`, and if we then pop the
122 //! `Some` constructor we get `[true, _]`. If we had popped `None` instead, we would get
125 //! This returns zero or more new pattern-stacks, as follows. We look at the pattern `p_1`
126 //! on top of the stack, and we have four cases:
127 //! 1.1. `p_1 = c(r_1, .., r_a)`, i.e. the top of the stack has constructor `c`. We
128 //! push onto the stack the arguments of this constructor, and return the result:
129 //! r_1, .., r_a, p_2, .., p_n
130 //! 1.2. `p_1 = c'(r_1, .., r_a')` where `c ≠ c'`. We discard the current stack and
132 //! 1.3. `p_1 = _`. We push onto the stack as many wildcards as the constructor `c` has
133 //! arguments (its arity), and return the resulting stack:
134 //! _, .., _, p_2, .., p_n
135 //! 1.4. `p_1 = r_1 | r_2`. We expand the OR-pattern and then recurse on each resulting
137 //! S(c, (r_1, p_2, .., p_n))
138 //! S(c, (r_2, p_2, .., p_n))
140 //! 2. We can pop a wildcard off the top of the stack. This is called `D(p)`, where `p` is
142 //! This is used when we know there are missing constructor cases, but there might be
143 //! existing wildcard patterns, so to check the usefulness of the matrix, we have to check
144 //! all its *other* components.
146 //! It is computed as follows. We look at the pattern `p_1` on top of the stack,
147 //! and we have three cases:
148 //! 2.1. `p_1 = c(r_1, .., r_a)`. We discard the current stack and return nothing.
149 //! 2.2. `p_1 = _`. We return the rest of the stack:
151 //! 2.3. `p_1 = r_1 | r_2`. We expand the OR-pattern and then recurse on each resulting
153 //! D((r_1, p_2, .., p_n))
154 //! D((r_2, p_2, .., p_n))
156 //! Note that the OR-patterns are not always used directly in Rust, but are used to derive the
157 //! exhaustive integer matching rules, so they're written here for posterity.
159 //! Both those operations extend straightforwardly to a list or pattern-stacks, i.e. a matrix, by
160 //! working row-by-row. Popping a constructor ends up keeping only the matrix rows that start with
161 //! the given constructor, and popping a wildcard keeps those rows that start with a wildcard.
164 //! The algorithm for computing `U`
165 //! -------------------------------
166 //! The algorithm is inductive (on the number of columns: i.e., components of tuple patterns).
167 //! That means we're going to check the components from left-to-right, so the algorithm
168 //! operates principally on the first component of the matrix and new pattern-stack `p`.
169 //! This algorithm is realised in the `is_useful` function.
171 //! Base case. (`n = 0`, i.e., an empty tuple pattern)
172 //! - If `P` already contains an empty pattern (i.e., if the number of patterns `m > 0`),
173 //! then `U(P, p)` is false.
174 //! - Otherwise, `P` must be empty, so `U(P, p)` is true.
176 //! Inductive step. (`n > 0`, i.e., whether there's at least one column
177 //! [which may then be expanded into further columns later])
178 //! We're going to match on the top of the new pattern-stack, `p_1`.
179 //! - If `p_1 == c(r_1, .., r_a)`, i.e. we have a constructor pattern.
180 //! Then, the usefulness of `p_1` can be reduced to whether it is useful when
181 //! we ignore all the patterns in the first column of `P` that involve other constructors.
182 //! This is where `S(c, P)` comes in:
183 //! `U(P, p) := U(S(c, P), S(c, p))`
184 //! This special case is handled in `is_useful_specialized`.
186 //! For example, if `P` is:
195 //! and `p` is [Some(false), 0], then we don't care about row 2 since we know `p` only
196 //! matches values that row 2 doesn't. For row 1 however, we need to dig into the
197 //! arguments of `Some` to know whether some new value is covered. So we compute
198 //! `U([[true, _]], [false, 0])`.
200 //! - If `p_1 == _`, then we look at the list of constructors that appear in the first
201 //! component of the rows of `P`:
202 //! + If there are some constructors that aren't present, then we might think that the
203 //! wildcard `_` is useful, since it covers those constructors that weren't covered
205 //! That's almost correct, but only works if there were no wildcards in those first
206 //! components. So we need to check that `p` is useful with respect to the rows that
207 //! start with a wildcard, if there are any. This is where `D` comes in:
208 //! `U(P, p) := U(D(P), D(p))`
210 //! For example, if `P` is:
215 //! [None, false, 1],
219 //! and `p` is [_, false, _], the `Some` constructor doesn't appear in `P`. So if we
220 //! only had row 2, we'd know that `p` is useful. However row 1 starts with a
221 //! wildcard, so we need to check whether `U([[true, _]], [false, 1])`.
223 //! + Otherwise, all possible constructors (for the relevant type) are present. In this
224 //! case we must check whether the wildcard pattern covers any unmatched value. For
225 //! that, we can think of the `_` pattern as a big OR-pattern that covers all
226 //! possible constructors. For `Option`, that would mean `_ = None | Some(_)` for
227 //! example. The wildcard pattern is useful in this case if it is useful when
228 //! specialized to one of the possible constructors. So we compute:
229 //! `U(P, p) := ∃(k ϵ constructors) U(S(k, P), S(k, p))`
231 //! For example, if `P` is:
240 //! and `p` is [_, false], both `None` and `Some` constructors appear in the first
241 //! components of `P`. We will therefore try popping both constructors in turn: we
242 //! compute `U([[true, _]], [_, false])` for the `Some` constructor, and `U([[false]],
243 //! [false])` for the `None` constructor. The first case returns true, so we know that
244 //! `p` is useful for `P`. Indeed, it matches `[Some(false), _]` that wasn't matched
247 //! - If `p_1 == r_1 | r_2`, then the usefulness depends on each `r_i` separately:
248 //! `U(P, p) := U(P, (r_1, p_2, .., p_n))
249 //! || U(P, (r_2, p_2, .., p_n))`
251 //! Modifications to the algorithm
252 //! ------------------------------
253 //! The algorithm in the paper doesn't cover some of the special cases that arise in Rust, for
254 //! example uninhabited types and variable-length slice patterns. These are drawn attention to
255 //! throughout the code below. I'll make a quick note here about how exhaustive integer matching is
256 //! accounted for, though.
258 //! Exhaustive integer matching
259 //! ---------------------------
260 //! An integer type can be thought of as a (huge) sum type: 1 | 2 | 3 | ...
261 //! So to support exhaustive integer matching, we can make use of the logic in the paper for
262 //! OR-patterns. However, we obviously can't just treat ranges x..=y as individual sums, because
263 //! they are likely gigantic. So we instead treat ranges as constructors of the integers. This means
264 //! that we have a constructor *of* constructors (the integers themselves). We then need to work
265 //! through all the inductive step rules above, deriving how the ranges would be treated as
266 //! OR-patterns, and making sure that they're treated in the same way even when they're ranges.
267 //! There are really only four special cases here:
268 //! - When we match on a constructor that's actually a range, we have to treat it as if we would
270 //! + It turns out that we can simply extend the case for single-value patterns in
271 //! `specialize` to either be *equal* to a value constructor, or *contained within* a range
273 //! + When the pattern itself is a range, you just want to tell whether any of the values in
274 //! the pattern range coincide with values in the constructor range, which is precisely
276 //! Since when encountering a range pattern for a value constructor, we also use inclusion, it
277 //! means that whenever the constructor is a value/range and the pattern is also a value/range,
278 //! we can simply use intersection to test usefulness.
279 //! - When we're testing for usefulness of a pattern and the pattern's first component is a
281 //! + If all the constructors appear in the matrix, we have a slight complication. By default,
282 //! the behaviour (i.e., a disjunction over specialised matrices for each constructor) is
283 //! invalid, because we want a disjunction over every *integer* in each range, not just a
284 //! disjunction over every range. This is a bit more tricky to deal with: essentially we need
285 //! to form equivalence classes of subranges of the constructor range for which the behaviour
286 //! of the matrix `P` and new pattern `p` are the same. This is described in more
287 //! detail in `Constructor::split`.
288 //! + If some constructors are missing from the matrix, it turns out we don't need to do
289 //! anything special (because we know none of the integers are actually wildcards: i.e., we
290 //! can't span wildcards using ranges).
291 use self::Constructor::*;
292 use self::SliceKind::*;
293 use self::Usefulness::*;
294 use self::WitnessPreference::*;
296 use rustc_data_structures::captures::Captures;
297 use rustc_data_structures::fx::{FxHashMap, FxHashSet};
298 use rustc_index::vec::Idx;
300 use super::{compare_const_vals, PatternFoldable, PatternFolder};
301 use super::{FieldPat, Pat, PatKind, PatRange};
303 use rustc_arena::TypedArena;
304 use rustc_attr::{SignedInt, UnsignedInt};
305 use rustc_hir::def_id::DefId;
306 use rustc_hir::{HirId, RangeEnd};
307 use rustc_middle::mir::interpret::{truncate, ConstValue};
308 use rustc_middle::mir::Field;
309 use rustc_middle::ty::layout::IntegerExt;
310 use rustc_middle::ty::{self, Const, Ty, TyCtxt};
311 use rustc_session::lint;
312 use rustc_span::{Span, DUMMY_SP};
313 use rustc_target::abi::{Integer, Size, VariantIdx};
315 use smallvec::{smallvec, SmallVec};
316 use std::cmp::{self, max, min, Ordering};
318 use std::iter::{FromIterator, IntoIterator};
319 use std::ops::RangeInclusive;
321 crate fn expand_pattern<'tcx>(pat: Pat<'tcx>) -> Pat<'tcx> {
322 LiteralExpander.fold_pattern(&pat)
325 struct LiteralExpander;
327 impl<'tcx> PatternFolder<'tcx> for LiteralExpander {
328 fn fold_pattern(&mut self, pat: &Pat<'tcx>) -> Pat<'tcx> {
329 debug!("fold_pattern {:?} {:?} {:?}", pat, pat.ty.kind(), pat.kind);
330 match (pat.ty.kind(), &*pat.kind) {
331 (_, &PatKind::Binding { subpattern: Some(ref s), .. }) => s.fold_with(self),
332 (_, &PatKind::AscribeUserType { subpattern: ref s, .. }) => s.fold_with(self),
333 _ => pat.super_fold_with(self),
338 impl<'tcx> Pat<'tcx> {
339 pub(super) fn is_wildcard(&self) -> bool {
341 PatKind::Binding { subpattern: None, .. } | PatKind::Wild => true,
347 /// A row of a matrix. Rows of len 1 are very common, which is why `SmallVec[_; 2]`
349 #[derive(Debug, Clone, PartialEq)]
350 crate struct PatStack<'p, 'tcx>(SmallVec<[&'p Pat<'tcx>; 2]>);
352 impl<'p, 'tcx> PatStack<'p, 'tcx> {
353 crate fn from_pattern(pat: &'p Pat<'tcx>) -> Self {
354 PatStack(smallvec![pat])
357 fn from_vec(vec: SmallVec<[&'p Pat<'tcx>; 2]>) -> Self {
361 fn from_slice(s: &[&'p Pat<'tcx>]) -> Self {
362 PatStack(SmallVec::from_slice(s))
365 fn is_empty(&self) -> bool {
369 fn len(&self) -> usize {
373 fn head(&self) -> &'p Pat<'tcx> {
377 fn to_tail(&self) -> Self {
378 PatStack::from_slice(&self.0[1..])
381 fn iter(&self) -> impl Iterator<Item = &Pat<'tcx>> {
382 self.0.iter().copied()
385 // If the first pattern is an or-pattern, expand this pattern. Otherwise, return `None`.
386 fn expand_or_pat(&self) -> Option<Vec<Self>> {
389 } else if let PatKind::Or { pats } = &*self.head().kind {
393 let mut new_patstack = PatStack::from_pattern(pat);
394 new_patstack.0.extend_from_slice(&self.0[1..]);
404 /// This computes `D(self)`. See top of the file for explanations.
405 fn specialize_wildcard(&self) -> Option<Self> {
406 if self.head().is_wildcard() { Some(self.to_tail()) } else { None }
409 /// This computes `S(constructor, self)`. See top of the file for explanations.
410 fn specialize_constructor(
412 cx: &MatchCheckCtxt<'p, 'tcx>,
413 constructor: &Constructor<'tcx>,
414 ctor_wild_subpatterns: &Fields<'p, 'tcx>,
415 is_my_head_ctor: bool,
416 ) -> Option<PatStack<'p, 'tcx>> {
417 let new_fields = specialize_one_pattern(
421 ctor_wild_subpatterns,
424 Some(new_fields.push_on_patstack(&self.0[1..]))
428 impl<'p, 'tcx> Default for PatStack<'p, 'tcx> {
429 fn default() -> Self {
430 PatStack(smallvec![])
434 impl<'p, 'tcx> FromIterator<&'p Pat<'tcx>> for PatStack<'p, 'tcx> {
435 fn from_iter<T>(iter: T) -> Self
437 T: IntoIterator<Item = &'p Pat<'tcx>>,
439 PatStack(iter.into_iter().collect())
443 /// Depending on the match patterns, the specialization process might be able to use a fast path.
444 /// Tracks whether we can use the fast path and the lookup table needed in those cases.
445 #[derive(Clone, Debug, PartialEq)]
446 enum SpecializationCache {
447 /// Patterns consist of only enum variants.
448 /// Variant patterns does not intersect with each other (in contrast to range patterns),
449 /// so it is possible to precompute the result of `Matrix::specialize_constructor` at a
450 /// lower computational complexity.
451 /// `lookup` is responsible for holding the precomputed result of
452 /// `Matrix::specialize_constructor`, while `wilds` is used for two purposes: the first one is
453 /// the precomputed result of `Matrix::specialize_wildcard`, and the second is to be used as a
454 /// fallback for `Matrix::specialize_constructor` when it tries to apply a constructor that
455 /// has not been seen in the `Matrix`. See `update_cache` for further explanations.
456 Variants { lookup: FxHashMap<DefId, SmallVec<[usize; 1]>>, wilds: SmallVec<[usize; 1]> },
457 /// Does not belong to the cases above, use the slow path.
462 #[derive(Clone, PartialEq)]
463 crate struct Matrix<'p, 'tcx> {
464 patterns: Vec<PatStack<'p, 'tcx>>,
465 cache: SpecializationCache,
468 impl<'p, 'tcx> Matrix<'p, 'tcx> {
469 crate fn empty() -> Self {
470 // Use `SpecializationCache::Incompatible` as a placeholder; we will initialize it on the
471 // first call to `push`. See the first half of `update_cache`.
472 Matrix { patterns: vec![], cache: SpecializationCache::Incompatible }
475 /// Pushes a new row to the matrix. If the row starts with an or-pattern, this expands it.
476 crate fn push(&mut self, row: PatStack<'p, 'tcx>) {
477 if let Some(rows) = row.expand_or_pat() {
479 // We recursively expand the or-patterns of the new rows.
480 // This is necessary as we might have `0 | (1 | 2)` or e.g., `x @ 0 | x @ (1 | 2)`.
484 self.patterns.push(row);
485 self.update_cache(self.patterns.len() - 1);
489 fn update_cache(&mut self, idx: usize) {
490 let row = &self.patterns[idx];
491 // We don't know which kind of cache could be used until we see the first row; therefore an
492 // empty `Matrix` is initialized with `SpecializationCache::Empty`, then the cache is
493 // assigned the appropriate variant below on the first call to `push`.
494 if self.patterns.is_empty() {
495 self.cache = if row.is_empty() {
496 SpecializationCache::Incompatible
498 match *row.head().kind {
499 PatKind::Variant { .. } => SpecializationCache::Variants {
500 lookup: FxHashMap::default(),
501 wilds: SmallVec::new(),
503 // Note: If the first pattern is a wildcard, then all patterns after that is not
504 // useful. The check is simple enough so we treat it as the same as unsupported
506 _ => SpecializationCache::Incompatible,
511 match &mut self.cache {
512 SpecializationCache::Variants { ref mut lookup, ref mut wilds } => {
513 let head = row.head();
515 _ if head.is_wildcard() => {
516 // Per rule 1.3 in the top-level comments, a wildcard pattern is included in
517 // the result of `specialize_constructor` for *any* `Constructor`.
518 // We push the wildcard pattern to the precomputed result for constructors
519 // that we have seen before; results for constructors we have not yet seen
520 // defaults to `wilds`, which is updated right below.
521 for (_, v) in lookup.iter_mut() {
524 // Per rule 2.1 and 2.2 in the top-level comments, only wildcard patterns
525 // are included in the result of `specialize_wildcard`.
526 // What we do here is to track the wildcards we have seen; so in addition to
527 // acting as the precomputed result of `specialize_wildcard`, `wilds` also
528 // serves as the default value of `specialize_constructor` for constructors
529 // that are not in `lookup`.
532 PatKind::Variant { adt_def, variant_index, .. } => {
533 // Handle the cases of rule 1.1 and 1.2 in the top-level comments.
534 // A variant pattern can only be included in the results of
535 // `specialize_constructor` for a particular constructor, therefore we are
536 // using a HashMap to track that.
538 .entry(adt_def.variants[variant_index].def_id)
539 // Default to `wilds` for absent keys. See above for an explanation.
540 .or_insert_with(|| wilds.clone())
544 self.cache = SpecializationCache::Incompatible;
548 SpecializationCache::Incompatible => {}
552 /// Iterate over the first component of each row
553 fn heads<'a>(&'a self) -> impl Iterator<Item = &'a Pat<'tcx>> + Captures<'p> {
554 self.patterns.iter().map(|r| r.head())
557 /// This computes `D(self)`. See top of the file for explanations.
558 fn specialize_wildcard(&self) -> Self {
560 SpecializationCache::Variants { wilds, .. } => {
562 wilds.iter().filter_map(|&i| self.patterns[i].specialize_wildcard()).collect();
563 // When debug assertions are enabled, check the results against the "slow path"
568 patterns: self.patterns.clone(),
569 cache: SpecializationCache::Incompatible
571 .specialize_wildcard()
575 SpecializationCache::Incompatible => {
576 self.patterns.iter().filter_map(|r| r.specialize_wildcard()).collect()
581 /// This computes `S(constructor, self)`. See top of the file for explanations.
582 fn specialize_constructor(
584 cx: &MatchCheckCtxt<'p, 'tcx>,
585 constructor: &Constructor<'tcx>,
586 ctor_wild_subpatterns: &Fields<'p, 'tcx>,
587 ) -> Matrix<'p, 'tcx> {
589 SpecializationCache::Variants { lookup, wilds } => {
590 let result: Self = if let Constructor::Variant(id) = constructor {
593 // Default to `wilds` for absent keys. See `update_cache` for an explanation.
597 self.patterns[i].specialize_constructor(
600 ctor_wild_subpatterns,
608 // When debug assertions are enabled, check the results against the "slow path"
613 patterns: self.patterns.clone(),
614 cache: SpecializationCache::Incompatible
616 .specialize_constructor(
619 ctor_wild_subpatterns
624 SpecializationCache::Incompatible => self
628 r.specialize_constructor(cx, constructor, ctor_wild_subpatterns, false)
635 /// Pretty-printer for matrices of patterns, example:
638 /// +++++++++++++++++++++++++++++
640 /// +++++++++++++++++++++++++++++
641 /// + true + [First] +
642 /// +++++++++++++++++++++++++++++
643 /// + true + [Second(true)] +
644 /// +++++++++++++++++++++++++++++
646 /// +++++++++++++++++++++++++++++
647 /// + _ + [_, _, tail @ ..] +
648 /// +++++++++++++++++++++++++++++
650 impl<'p, 'tcx> fmt::Debug for Matrix<'p, 'tcx> {
651 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
654 let Matrix { patterns: m, .. } = self;
655 let pretty_printed_matrix: Vec<Vec<String>> =
656 m.iter().map(|row| row.iter().map(|pat| format!("{:?}", pat)).collect()).collect();
658 let column_count = m.iter().map(|row| row.len()).max().unwrap_or(0);
659 assert!(m.iter().all(|row| row.len() == column_count));
660 let column_widths: Vec<usize> = (0..column_count)
661 .map(|col| pretty_printed_matrix.iter().map(|row| row[col].len()).max().unwrap_or(0))
664 let total_width = column_widths.iter().cloned().sum::<usize>() + column_count * 3 + 1;
665 let br = "+".repeat(total_width);
666 write!(f, "{}\n", br)?;
667 for row in pretty_printed_matrix {
669 for (column, pat_str) in row.into_iter().enumerate() {
671 write!(f, "{:1$}", pat_str, column_widths[column])?;
675 write!(f, "{}\n", br)?;
681 impl<'p, 'tcx> FromIterator<PatStack<'p, 'tcx>> for Matrix<'p, 'tcx> {
682 fn from_iter<T>(iter: T) -> Self
684 T: IntoIterator<Item = PatStack<'p, 'tcx>>,
686 let mut matrix = Matrix::empty();
688 // Using `push` ensures we correctly expand or-patterns.
695 crate struct MatchCheckCtxt<'a, 'tcx> {
696 crate tcx: TyCtxt<'tcx>,
697 /// The module in which the match occurs. This is necessary for
698 /// checking inhabited-ness of types because whether a type is (visibly)
699 /// inhabited can depend on whether it was defined in the current module or
700 /// not. E.g., `struct Foo { _private: ! }` cannot be seen to be empty
701 /// outside it's module and should not be matchable with an empty match
704 crate param_env: ty::ParamEnv<'tcx>,
705 crate pattern_arena: &'a TypedArena<Pat<'tcx>>,
708 impl<'a, 'tcx> MatchCheckCtxt<'a, 'tcx> {
709 fn is_uninhabited(&self, ty: Ty<'tcx>) -> bool {
710 if self.tcx.features().exhaustive_patterns {
711 self.tcx.is_ty_uninhabited_from(self.module, ty, self.param_env)
717 /// Returns whether the given type is an enum from another crate declared `#[non_exhaustive]`.
718 crate fn is_foreign_non_exhaustive_enum(&self, ty: Ty<'tcx>) -> bool {
720 ty::Adt(def, ..) => {
721 def.is_enum() && def.is_variant_list_non_exhaustive() && !def.did.is_local()
728 #[derive(Copy, Clone, Debug, PartialEq, Eq)]
730 /// Patterns of length `n` (`[x, y]`).
732 /// Patterns using the `..` notation (`[x, .., y]`).
733 /// Captures any array constructor of `length >= i + j`.
734 /// In the case where `array_len` is `Some(_)`,
735 /// this indicates that we only care about the first `i` and the last `j` values of the array,
736 /// and everything in between is a wildcard `_`.
741 fn arity(self) -> u64 {
743 FixedLen(length) => length,
744 VarLen(prefix, suffix) => prefix + suffix,
748 /// Whether this pattern includes patterns of length `other_len`.
749 fn covers_length(self, other_len: u64) -> bool {
751 FixedLen(len) => len == other_len,
752 VarLen(prefix, suffix) => prefix + suffix <= other_len,
756 /// Returns a collection of slices that spans the values covered by `self`, subtracted by the
757 /// values covered by `other`: i.e., `self \ other` (in set notation).
758 fn subtract(self, other: Self) -> SmallVec<[Self; 1]> {
759 // Remember, `VarLen(i, j)` covers the union of `FixedLen` from `i + j` to infinity.
760 // Naming: we remove the "neg" constructors from the "pos" ones.
762 FixedLen(pos_len) => {
763 if other.covers_length(pos_len) {
769 VarLen(pos_prefix, pos_suffix) => {
770 let pos_len = pos_prefix + pos_suffix;
772 FixedLen(neg_len) => {
773 if neg_len < pos_len {
778 // We know that `neg_len + 1 >= pos_len >= pos_suffix`.
779 .chain(Some(VarLen(neg_len + 1 - pos_suffix, pos_suffix)))
783 VarLen(neg_prefix, neg_suffix) => {
784 let neg_len = neg_prefix + neg_suffix;
785 if neg_len <= pos_len {
788 (pos_len..neg_len).map(FixedLen).collect()
797 /// A constructor for array and slice patterns.
798 #[derive(Copy, Clone, Debug, PartialEq, Eq)]
800 /// `None` if the matched value is a slice, `Some(n)` if it is an array of size `n`.
801 array_len: Option<u64>,
802 /// The kind of pattern it is: fixed-length `[x, y]` or variable length `[x, .., y]`.
807 /// Returns what patterns this constructor covers: either fixed-length patterns or
808 /// variable-length patterns.
809 fn pattern_kind(self) -> SliceKind {
811 Slice { array_len: Some(len), kind: VarLen(prefix, suffix) }
812 if prefix + suffix == len =>
820 /// Returns what values this constructor covers: either values of only one given length, or
821 /// values of length above a given length.
822 /// This is different from `pattern_kind()` because in some cases the pattern only takes into
823 /// account a subset of the entries of the array, but still only captures values of a given
825 fn value_kind(self) -> SliceKind {
827 Slice { array_len: Some(len), kind: VarLen(_, _) } => FixedLen(len),
832 fn arity(self) -> u64 {
833 self.pattern_kind().arity()
836 /// The exhaustiveness-checking paper does not include any details on
837 /// checking variable-length slice patterns. However, they are matched
838 /// by an infinite collection of fixed-length array patterns.
840 /// Checking the infinite set directly would take an infinite amount
841 /// of time. However, it turns out that for each finite set of
842 /// patterns `P`, all sufficiently large array lengths are equivalent:
844 /// Each slice `s` with a "sufficiently-large" length `l ≥ L` that applies
845 /// to exactly the subset `Pₜ` of `P` can be transformed to a slice
846 /// `sₘ` for each sufficiently-large length `m` that applies to exactly
847 /// the same subset of `P`.
849 /// Because of that, each witness for reachability-checking from one
850 /// of the sufficiently-large lengths can be transformed to an
851 /// equally-valid witness from any other length, so we only have
852 /// to check slice lengths from the "minimal sufficiently-large length"
855 /// Note that the fact that there is a *single* `sₘ` for each `m`
856 /// not depending on the specific pattern in `P` is important: if
857 /// you look at the pair of patterns
860 /// Then any slice of length ≥1 that matches one of these two
861 /// patterns can be trivially turned to a slice of any
862 /// other length ≥1 that matches them and vice-versa - for
863 /// but the slice from length 2 `[false, true]` that matches neither
864 /// of these patterns can't be turned to a slice from length 1 that
865 /// matches neither of these patterns, so we have to consider
866 /// slices from length 2 there.
868 /// Now, to see that that length exists and find it, observe that slice
869 /// patterns are either "fixed-length" patterns (`[_, _, _]`) or
870 /// "variable-length" patterns (`[_, .., _]`).
872 /// For fixed-length patterns, all slices with lengths *longer* than
873 /// the pattern's length have the same outcome (of not matching), so
874 /// as long as `L` is greater than the pattern's length we can pick
875 /// any `sₘ` from that length and get the same result.
877 /// For variable-length patterns, the situation is more complicated,
878 /// because as seen above the precise value of `sₘ` matters.
880 /// However, for each variable-length pattern `p` with a prefix of length
881 /// `plₚ` and suffix of length `slₚ`, only the first `plₚ` and the last
882 /// `slₚ` elements are examined.
884 /// Therefore, as long as `L` is positive (to avoid concerns about empty
885 /// types), all elements after the maximum prefix length and before
886 /// the maximum suffix length are not examined by any variable-length
887 /// pattern, and therefore can be added/removed without affecting
888 /// them - creating equivalent patterns from any sufficiently-large
891 /// Of course, if fixed-length patterns exist, we must be sure
892 /// that our length is large enough to miss them all, so
893 /// we can pick `L = max(max(FIXED_LEN)+1, max(PREFIX_LEN) + max(SUFFIX_LEN))`
895 /// for example, with the above pair of patterns, all elements
896 /// but the first and last can be added/removed, so any
897 /// witness of length ≥2 (say, `[false, false, true]`) can be
898 /// turned to a witness from any other length ≥2.
901 cx: &MatchCheckCtxt<'p, 'tcx>,
902 matrix: &Matrix<'p, 'tcx>,
903 ) -> SmallVec<[Constructor<'tcx>; 1]> {
904 let (array_len, self_prefix, self_suffix) = match self {
905 Slice { array_len, kind: VarLen(self_prefix, self_suffix) } => {
906 (array_len, self_prefix, self_suffix)
908 _ => return smallvec![Slice(self)],
912 matrix.heads().filter_map(|pat| pat_constructor(cx.tcx, cx.param_env, pat));
914 let mut max_prefix_len = self_prefix;
915 let mut max_suffix_len = self_suffix;
916 let mut max_fixed_len = 0;
918 for ctor in head_ctors {
919 if let Slice(slice) = ctor {
920 match slice.pattern_kind() {
922 max_fixed_len = cmp::max(max_fixed_len, len);
924 VarLen(prefix, suffix) => {
925 max_prefix_len = cmp::max(max_prefix_len, prefix);
926 max_suffix_len = cmp::max(max_suffix_len, suffix);
932 // For diagnostics, we keep the prefix and suffix lengths separate, so in the case
933 // where `max_fixed_len + 1` is the largest, we adapt `max_prefix_len` accordingly,
934 // so that `L = max_prefix_len + max_suffix_len`.
935 if max_fixed_len + 1 >= max_prefix_len + max_suffix_len {
936 // The subtraction can't overflow thanks to the above check.
937 // The new `max_prefix_len` is also guaranteed to be larger than its previous
939 max_prefix_len = max_fixed_len + 1 - max_suffix_len;
944 let kind = if max_prefix_len + max_suffix_len < len {
945 VarLen(max_prefix_len, max_suffix_len)
949 smallvec![Slice(Slice { array_len, kind })]
952 // `ctor` originally covered the range `(self_prefix +
953 // self_suffix..infinity)`. We now split it into two: lengths smaller than
954 // `max_prefix_len + max_suffix_len` are treated independently as
955 // fixed-lengths slices, and lengths above are captured by a final VarLen
957 let smaller_lengths =
958 (self_prefix + self_suffix..max_prefix_len + max_suffix_len).map(FixedLen);
959 let final_slice = VarLen(max_prefix_len, max_suffix_len);
961 .chain(Some(final_slice))
962 .map(|kind| Slice { array_len, kind })
970 /// A value can be decomposed into a constructor applied to some fields. This struct represents
971 /// the constructor. See also `Fields`.
973 /// `pat_constructor` retrieves the constructor corresponding to a pattern.
974 /// `specialize_one_pattern` returns the list of fields corresponding to a pattern, given a
975 /// constructor. `Constructor::apply` reconstructs the pattern from a pair of `Constructor` and
977 #[derive(Clone, Debug, PartialEq)]
978 enum Constructor<'tcx> {
979 /// The constructor for patterns that have a single constructor, like tuples, struct patterns
980 /// and fixed-length arrays.
984 /// Ranges of integer literal values (`2`, `2..=5` or `2..5`).
985 IntRange(IntRange<'tcx>),
986 /// Ranges of floating-point literal values (`2.0..=5.2`).
987 FloatRange(&'tcx ty::Const<'tcx>, &'tcx ty::Const<'tcx>, RangeEnd),
988 /// String literals. Strings are not quite the same as `&[u8]` so we treat them separately.
989 Str(&'tcx ty::Const<'tcx>),
990 /// Array and slice patterns.
992 /// Constants that must not be matched structurally. They are treated as black
993 /// boxes for the purposes of exhaustiveness: we must not inspect them, and they
994 /// don't count towards making a match exhaustive.
996 /// Fake extra constructor for enums that aren't allowed to be matched exhaustively.
1000 impl<'tcx> Constructor<'tcx> {
1001 fn variant_index_for_adt(&self, adt: &'tcx ty::AdtDef) -> VariantIdx {
1003 Variant(id) => adt.variant_index_with_id(id),
1005 assert!(!adt.is_enum());
1008 _ => bug!("bad constructor {:?} for adt {:?}", self, adt),
1012 // Returns the set of constructors covered by `self` but not by
1013 // anything in `other_ctors`.
1014 fn subtract_ctors(&self, other_ctors: &Vec<Constructor<'tcx>>) -> Vec<Constructor<'tcx>> {
1015 if other_ctors.is_empty() {
1016 return vec![self.clone()];
1020 // Those constructors can only match themselves.
1021 Single | Variant(_) | Str(..) | FloatRange(..) => {
1022 if other_ctors.iter().any(|c| c == self) { vec![] } else { vec![self.clone()] }
1025 let mut other_slices = other_ctors
1027 .filter_map(|c: &Constructor<'_>| match c {
1028 Slice(slice) => Some(*slice),
1029 _ => bug!("bad slice pattern constructor {:?}", c),
1031 .map(Slice::value_kind);
1033 match slice.value_kind() {
1034 FixedLen(self_len) => {
1035 if other_slices.any(|other_slice| other_slice.covers_length(self_len)) {
1041 kind @ VarLen(..) => {
1042 let mut remaining_slices = vec![kind];
1044 // For each used slice, subtract from the current set of slices.
1045 for other_slice in other_slices {
1046 remaining_slices = remaining_slices
1048 .flat_map(|remaining_slice| remaining_slice.subtract(other_slice))
1051 // If the constructors that have been considered so far already cover
1052 // the entire range of `self`, no need to look at more constructors.
1053 if remaining_slices.is_empty() {
1060 .map(|kind| Slice { array_len: slice.array_len, kind })
1066 IntRange(self_range) => {
1067 let mut remaining_ranges = vec![self_range.clone()];
1068 for other_ctor in other_ctors {
1069 if let IntRange(other_range) = other_ctor {
1070 if other_range == self_range {
1071 // If the `self` range appears directly in a `match` arm, we can
1072 // eliminate it straight away.
1073 remaining_ranges = vec![];
1075 // Otherwise explicitly compute the remaining ranges.
1076 remaining_ranges = other_range.subtract_from(remaining_ranges);
1079 // If the ranges that have been considered so far already cover the entire
1080 // range of values, we can return early.
1081 if remaining_ranges.is_empty() {
1087 // Convert the ranges back into constructors.
1088 remaining_ranges.into_iter().map(IntRange).collect()
1090 // This constructor is never covered by anything else
1091 NonExhaustive => vec![NonExhaustive],
1092 Opaque => bug!("unexpected opaque ctor {:?} found in all_ctors", self),
1096 /// Some constructors (namely IntRange and Slice) actually stand for a set of actual
1097 /// constructors (integers and fixed-sized slices). When specializing for these
1098 /// constructors, we want to be specialising for the actual underlying constructors.
1099 /// Naively, we would simply return the list of constructors they correspond to. We instead are
1100 /// more clever: if there are constructors that we know will behave the same wrt the current
1101 /// matrix, we keep them grouped. For example, all slices of a sufficiently large length
1102 /// will either be all useful or all non-useful with a given matrix.
1104 /// See the branches for details on how the splitting is done.
1106 /// This function may discard some irrelevant constructors if this preserves behavior and
1107 /// diagnostics. Eg. for the `_` case, we ignore the constructors already present in the
1108 /// matrix, unless all of them are.
1110 /// `hir_id` is `None` when we're evaluating the wildcard pattern. In that case we do not want
1111 /// to lint for overlapping ranges.
1114 cx: &MatchCheckCtxt<'p, 'tcx>,
1116 matrix: &Matrix<'p, 'tcx>,
1117 hir_id: Option<HirId>,
1118 ) -> SmallVec<[Self; 1]> {
1119 debug!("Constructor::split({:#?}, {:#?})", self, matrix);
1122 // Fast-track if the range is trivial. In particular, we don't do the overlapping
1124 IntRange(ctor_range)
1125 if ctor_range.treat_exhaustively(cx.tcx) && !ctor_range.is_singleton() =>
1127 ctor_range.split(cx, pcx, matrix, hir_id)
1129 Slice(slice @ Slice { kind: VarLen(..), .. }) => slice.split(cx, matrix),
1130 // Any other constructor can be used unchanged.
1131 _ => smallvec![self],
1135 /// Apply a constructor to a list of patterns, yielding a new pattern. `pats`
1136 /// must have as many elements as this constructor's arity.
1138 /// This is roughly the inverse of `specialize_one_pattern`.
1141 /// `self`: `Constructor::Single`
1142 /// `ty`: `(u32, u32, u32)`
1143 /// `pats`: `[10, 20, _]`
1144 /// returns `(10, 20, _)`
1146 /// `self`: `Constructor::Variant(Option::Some)`
1147 /// `ty`: `Option<bool>`
1148 /// `pats`: `[false]`
1149 /// returns `Some(false)`
1152 cx: &MatchCheckCtxt<'p, 'tcx>,
1154 fields: Fields<'p, 'tcx>,
1156 let mut subpatterns = fields.all_patterns();
1158 let pat = match self {
1159 Single | Variant(_) => match ty.kind() {
1160 ty::Adt(..) | ty::Tuple(..) => {
1161 let subpatterns = subpatterns
1163 .map(|(i, p)| FieldPat { field: Field::new(i), pattern: p })
1166 if let ty::Adt(adt, substs) = ty.kind() {
1171 variant_index: self.variant_index_for_adt(adt),
1175 PatKind::Leaf { subpatterns }
1178 PatKind::Leaf { subpatterns }
1181 ty::Ref(..) => PatKind::Deref { subpattern: subpatterns.next().unwrap() },
1182 ty::Slice(_) | ty::Array(..) => bug!("bad slice pattern {:?} {:?}", self, ty),
1185 Slice(slice) => match slice.pattern_kind() {
1187 PatKind::Slice { prefix: subpatterns.collect(), slice: None, suffix: vec![] }
1189 VarLen(prefix, _) => {
1190 let mut prefix: Vec<_> = subpatterns.by_ref().take(prefix as usize).collect();
1191 if slice.array_len.is_some() {
1192 // Improves diagnostics a bit: if the type is a known-size array, instead
1193 // of reporting `[x, _, .., _, y]`, we prefer to report `[x, .., y]`.
1194 // This is incorrect if the size is not known, since `[_, ..]` captures
1195 // arrays of lengths `>= 1` whereas `[..]` captures any length.
1196 while !prefix.is_empty() && prefix.last().unwrap().is_wildcard() {
1200 let suffix: Vec<_> = if slice.array_len.is_some() {
1202 subpatterns.skip_while(Pat::is_wildcard).collect()
1204 subpatterns.collect()
1206 let wild = Pat::wildcard_from_ty(ty);
1207 PatKind::Slice { prefix, slice: Some(wild), suffix }
1210 &Str(value) => PatKind::Constant { value },
1211 &FloatRange(lo, hi, end) => PatKind::Range(PatRange { lo, hi, end }),
1212 IntRange(range) => return range.to_pat(cx.tcx),
1213 NonExhaustive => PatKind::Wild,
1214 Opaque => bug!("we should not try to apply an opaque constructor {:?}", self),
1217 Pat { ty, span: DUMMY_SP, kind: Box::new(pat) }
1220 /// Like `apply`, but where all the subpatterns are wildcards `_`.
1221 fn apply_wildcards<'a>(&self, cx: &MatchCheckCtxt<'a, 'tcx>, ty: Ty<'tcx>) -> Pat<'tcx> {
1222 self.apply(cx, ty, Fields::wildcards(cx, self, ty))
1226 /// Some fields need to be explicitly hidden away in certain cases; see the comment above the
1227 /// `Fields` struct. This struct represents such a potentially-hidden field. When a field is hidden
1228 /// we still keep its type around.
1229 #[derive(Debug, Copy, Clone)]
1230 enum FilteredField<'p, 'tcx> {
1231 Kept(&'p Pat<'tcx>),
1235 impl<'p, 'tcx> FilteredField<'p, 'tcx> {
1236 fn kept(self) -> Option<&'p Pat<'tcx>> {
1238 FilteredField::Kept(p) => Some(p),
1239 FilteredField::Hidden(_) => None,
1243 fn to_pattern(self) -> Pat<'tcx> {
1245 FilteredField::Kept(p) => p.clone(),
1246 FilteredField::Hidden(ty) => Pat::wildcard_from_ty(ty),
1251 /// A value can be decomposed into a constructor applied to some fields. This struct represents
1252 /// those fields, generalized to allow patterns in each field. See also `Constructor`.
1254 /// If a private or `non_exhaustive` field is uninhabited, the code mustn't observe that it is
1255 /// uninhabited. For that, we filter these fields out of the matrix. This is subtle because we
1256 /// still need to have those fields back when going to/from a `Pat`. Most of this is handled
1257 /// automatically in `Fields`, but when constructing or deconstructing `Fields` you need to be
1258 /// careful. As a rule, when going to/from the matrix, use the filtered field list; when going
1259 /// to/from `Pat`, use the full field list.
1260 /// This filtering is uncommon in practice, because uninhabited fields are rarely used, so we avoid
1261 /// it when possible to preserve performance.
1262 #[derive(Debug, Clone)]
1263 enum Fields<'p, 'tcx> {
1264 /// Lists of patterns that don't contain any filtered fields.
1265 /// `Slice` and `Vec` behave the same; the difference is only to avoid allocating and
1266 /// triple-dereferences when possible. Frankly this is premature optimization, I (Nadrieril)
1267 /// have not measured if it really made a difference.
1268 Slice(&'p [Pat<'tcx>]),
1269 Vec(SmallVec<[&'p Pat<'tcx>; 2]>),
1270 /// Patterns where some of the fields need to be hidden. `kept_count` caches the number of
1271 /// non-hidden fields.
1273 fields: SmallVec<[FilteredField<'p, 'tcx>; 2]>,
1278 impl<'p, 'tcx> Fields<'p, 'tcx> {
1279 fn empty() -> Self {
1283 /// Construct a new `Fields` from the given pattern. Must not be used if the pattern is a field
1284 /// of a struct/tuple/variant.
1285 fn from_single_pattern(pat: &'p Pat<'tcx>) -> Self {
1286 Fields::Slice(std::slice::from_ref(pat))
1289 /// Convenience; internal use.
1290 fn wildcards_from_tys(
1291 cx: &MatchCheckCtxt<'p, 'tcx>,
1292 tys: impl IntoIterator<Item = Ty<'tcx>>,
1294 let wilds = tys.into_iter().map(Pat::wildcard_from_ty);
1295 let pats = cx.pattern_arena.alloc_from_iter(wilds);
1299 /// Creates a new list of wildcard fields for a given constructor.
1301 cx: &MatchCheckCtxt<'p, 'tcx>,
1302 constructor: &Constructor<'tcx>,
1305 let wildcard_from_ty = |ty| &*cx.pattern_arena.alloc(Pat::wildcard_from_ty(ty));
1307 let ret = match constructor {
1308 Single | Variant(_) => match ty.kind() {
1309 ty::Tuple(ref fs) => {
1310 Fields::wildcards_from_tys(cx, fs.into_iter().map(|ty| ty.expect_ty()))
1312 ty::Ref(_, rty, _) => Fields::from_single_pattern(wildcard_from_ty(rty)),
1313 ty::Adt(adt, substs) => {
1315 // Use T as the sub pattern type of Box<T>.
1316 Fields::from_single_pattern(wildcard_from_ty(substs.type_at(0)))
1318 let variant = &adt.variants[constructor.variant_index_for_adt(adt)];
1319 // Whether we must not match the fields of this variant exhaustively.
1320 let is_non_exhaustive =
1321 variant.is_field_list_non_exhaustive() && !adt.did.is_local();
1322 let field_tys = variant.fields.iter().map(|field| field.ty(cx.tcx, substs));
1323 // In the following cases, we don't need to filter out any fields. This is
1324 // the vast majority of real cases, since uninhabited fields are uncommon.
1325 let has_no_hidden_fields = (adt.is_enum() && !is_non_exhaustive)
1326 || !field_tys.clone().any(|ty| cx.is_uninhabited(ty));
1328 if has_no_hidden_fields {
1329 Fields::wildcards_from_tys(cx, field_tys)
1331 let mut kept_count = 0;
1332 let fields = variant
1336 let ty = field.ty(cx.tcx, substs);
1337 let is_visible = adt.is_enum()
1338 || field.vis.is_accessible_from(cx.module, cx.tcx);
1339 let is_uninhabited = cx.is_uninhabited(ty);
1341 // In the cases of either a `#[non_exhaustive]` field list
1342 // or a non-public field, we hide uninhabited fields in
1343 // order not to reveal the uninhabitedness of the whole
1345 if is_uninhabited && (!is_visible || is_non_exhaustive) {
1346 FilteredField::Hidden(ty)
1349 FilteredField::Kept(wildcard_from_ty(ty))
1353 Fields::Filtered { fields, kept_count }
1357 _ => Fields::empty(),
1359 Slice(slice) => match *ty.kind() {
1360 ty::Slice(ty) | ty::Array(ty, _) => {
1361 let arity = slice.arity();
1362 Fields::wildcards_from_tys(cx, (0..arity).map(|_| ty))
1364 _ => bug!("bad slice pattern {:?} {:?}", constructor, ty),
1366 Str(..) | FloatRange(..) | IntRange(..) | NonExhaustive | Opaque => Fields::empty(),
1368 debug!("Fields::wildcards({:?}, {:?}) = {:#?}", constructor, ty, ret);
1372 /// Returns the number of patterns from the viewpoint of match-checking, i.e. excluding hidden
1373 /// fields. This is what we want in most cases in this file, the only exception being
1374 /// conversion to/from `Pat`.
1375 fn len(&self) -> usize {
1377 Fields::Slice(pats) => pats.len(),
1378 Fields::Vec(pats) => pats.len(),
1379 Fields::Filtered { kept_count, .. } => *kept_count,
1383 /// Returns the complete list of patterns, including hidden fields.
1384 fn all_patterns(self) -> impl Iterator<Item = Pat<'tcx>> {
1385 let pats: SmallVec<[_; 2]> = match self {
1386 Fields::Slice(pats) => pats.iter().cloned().collect(),
1387 Fields::Vec(pats) => pats.into_iter().cloned().collect(),
1388 Fields::Filtered { fields, .. } => {
1389 // We don't skip any fields here.
1390 fields.into_iter().map(|p| p.to_pattern()).collect()
1396 /// Overrides some of the fields with the provided patterns. Exactly like
1397 /// `replace_fields_indexed`, except that it takes `FieldPat`s as input.
1398 fn replace_with_fieldpats(
1400 new_pats: impl IntoIterator<Item = &'p FieldPat<'tcx>>,
1402 self.replace_fields_indexed(
1403 new_pats.into_iter().map(|pat| (pat.field.index(), &pat.pattern)),
1407 /// Overrides some of the fields with the provided patterns. This is used when a pattern
1408 /// defines some fields but not all, for example `Foo { field1: Some(_), .. }`: here we start with a
1409 /// `Fields` that is just one wildcard per field of the `Foo` struct, and override the entry
1410 /// corresponding to `field1` with the pattern `Some(_)`. This is also used for slice patterns
1411 /// for the same reason.
1412 fn replace_fields_indexed(
1414 new_pats: impl IntoIterator<Item = (usize, &'p Pat<'tcx>)>,
1416 let mut fields = self.clone();
1417 if let Fields::Slice(pats) = fields {
1418 fields = Fields::Vec(pats.iter().collect());
1422 Fields::Vec(pats) => {
1423 for (i, pat) in new_pats {
1427 Fields::Filtered { fields, .. } => {
1428 for (i, pat) in new_pats {
1429 if let FilteredField::Kept(p) = &mut fields[i] {
1434 Fields::Slice(_) => unreachable!(),
1439 /// Replaces contained fields with the given filtered list of patterns, e.g. taken from the
1440 /// matrix. There must be `len()` patterns in `pats`.
1443 cx: &MatchCheckCtxt<'p, 'tcx>,
1444 pats: impl IntoIterator<Item = Pat<'tcx>>,
1446 let pats: &[_] = cx.pattern_arena.alloc_from_iter(pats);
1449 Fields::Filtered { fields, kept_count } => {
1450 let mut pats = pats.iter();
1451 let mut fields = fields.clone();
1452 for f in &mut fields {
1453 if let FilteredField::Kept(p) = f {
1454 // We take one input pattern for each `Kept` field, in order.
1455 *p = pats.next().unwrap();
1458 Fields::Filtered { fields, kept_count: *kept_count }
1460 _ => Fields::Slice(pats),
1464 fn push_on_patstack(self, stack: &[&'p Pat<'tcx>]) -> PatStack<'p, 'tcx> {
1465 let pats: SmallVec<_> = match self {
1466 Fields::Slice(pats) => pats.iter().chain(stack.iter().copied()).collect(),
1467 Fields::Vec(mut pats) => {
1468 pats.extend_from_slice(stack);
1471 Fields::Filtered { fields, .. } => {
1472 // We skip hidden fields here
1473 fields.into_iter().filter_map(|p| p.kept()).chain(stack.iter().copied()).collect()
1476 PatStack::from_vec(pats)
1480 #[derive(Clone, Debug)]
1481 crate enum Usefulness<'tcx> {
1482 /// Carries a list of unreachable subpatterns. Used only in the presence of or-patterns.
1484 /// Carries a list of witnesses of non-exhaustiveness.
1485 UsefulWithWitness(Vec<Witness<'tcx>>),
1489 impl<'tcx> Usefulness<'tcx> {
1490 fn new_useful(preference: WitnessPreference) -> Self {
1492 ConstructWitness => UsefulWithWitness(vec![Witness(vec![])]),
1493 LeaveOutWitness => Useful(vec![]),
1497 fn is_useful(&self) -> bool {
1504 fn apply_constructor<'p>(
1506 cx: &MatchCheckCtxt<'p, 'tcx>,
1507 ctor: &Constructor<'tcx>,
1509 ctor_wild_subpatterns: &Fields<'p, 'tcx>,
1512 UsefulWithWitness(witnesses) => UsefulWithWitness(
1515 .map(|witness| witness.apply_constructor(cx, &ctor, ty, ctor_wild_subpatterns))
1522 fn apply_wildcard(self, ty: Ty<'tcx>) -> Self {
1524 UsefulWithWitness(witnesses) => {
1525 let wild = Pat::wildcard_from_ty(ty);
1529 .map(|mut witness| {
1530 witness.0.push(wild.clone());
1540 fn apply_missing_ctors(
1542 cx: &MatchCheckCtxt<'_, 'tcx>,
1544 missing_ctors: &MissingConstructors<'tcx>,
1547 UsefulWithWitness(witnesses) => {
1548 let new_patterns: Vec<_> =
1549 missing_ctors.iter().map(|ctor| ctor.apply_wildcards(cx, ty)).collect();
1550 // Add the new patterns to each witness
1554 .flat_map(|witness| {
1555 new_patterns.iter().map(move |pat| {
1556 let mut witness = witness.clone();
1557 witness.0.push(pat.clone());
1569 #[derive(Copy, Clone, Debug)]
1570 crate enum WitnessPreference {
1575 #[derive(Copy, Clone, Debug)]
1576 struct PatCtxt<'tcx> {
1581 /// A witness of non-exhaustiveness for error reporting, represented
1582 /// as a list of patterns (in reverse order of construction) with
1583 /// wildcards inside to represent elements that can take any inhabitant
1584 /// of the type as a value.
1586 /// A witness against a list of patterns should have the same types
1587 /// and length as the pattern matched against. Because Rust `match`
1588 /// is always against a single pattern, at the end the witness will
1589 /// have length 1, but in the middle of the algorithm, it can contain
1590 /// multiple patterns.
1592 /// For example, if we are constructing a witness for the match against
1595 /// struct Pair(Option<(u32, u32)>, bool);
1597 /// match (p: Pair) {
1598 /// Pair(None, _) => {}
1599 /// Pair(_, false) => {}
1603 /// We'll perform the following steps:
1604 /// 1. Start with an empty witness
1605 /// `Witness(vec![])`
1606 /// 2. Push a witness `Some(_)` against the `None`
1607 /// `Witness(vec![Some(_)])`
1608 /// 3. Push a witness `true` against the `false`
1609 /// `Witness(vec![Some(_), true])`
1610 /// 4. Apply the `Pair` constructor to the witnesses
1611 /// `Witness(vec![Pair(Some(_), true)])`
1613 /// The final `Pair(Some(_), true)` is then the resulting witness.
1614 #[derive(Clone, Debug)]
1615 crate struct Witness<'tcx>(Vec<Pat<'tcx>>);
1617 impl<'tcx> Witness<'tcx> {
1618 crate fn single_pattern(self) -> Pat<'tcx> {
1619 assert_eq!(self.0.len(), 1);
1620 self.0.into_iter().next().unwrap()
1623 /// Constructs a partial witness for a pattern given a list of
1624 /// patterns expanded by the specialization step.
1626 /// When a pattern P is discovered to be useful, this function is used bottom-up
1627 /// to reconstruct a complete witness, e.g., a pattern P' that covers a subset
1628 /// of values, V, where each value in that set is not covered by any previously
1629 /// used patterns and is covered by the pattern P'. Examples:
1631 /// left_ty: tuple of 3 elements
1632 /// pats: [10, 20, _] => (10, 20, _)
1634 /// left_ty: struct X { a: (bool, &'static str), b: usize}
1635 /// pats: [(false, "foo"), 42] => X { a: (false, "foo"), b: 42 }
1636 fn apply_constructor<'p>(
1638 cx: &MatchCheckCtxt<'p, 'tcx>,
1639 ctor: &Constructor<'tcx>,
1641 ctor_wild_subpatterns: &Fields<'p, 'tcx>,
1644 let len = self.0.len();
1645 let arity = ctor_wild_subpatterns.len();
1646 let pats = self.0.drain((len - arity)..).rev();
1647 let fields = ctor_wild_subpatterns.replace_fields(cx, pats);
1648 ctor.apply(cx, ty, fields)
1657 /// This determines the set of all possible constructors of a pattern matching
1658 /// values of type `left_ty`. For vectors, this would normally be an infinite set
1659 /// but is instead bounded by the maximum fixed length of slice patterns in
1660 /// the column of patterns being analyzed.
1662 /// We make sure to omit constructors that are statically impossible. E.g., for
1663 /// `Option<!>`, we do not include `Some(_)` in the returned list of constructors.
1664 /// Invariant: this returns an empty `Vec` if and only if the type is uninhabited (as determined by
1665 /// `cx.is_uninhabited()`).
1666 fn all_constructors<'a, 'tcx>(
1667 cx: &MatchCheckCtxt<'a, 'tcx>,
1669 ) -> Vec<Constructor<'tcx>> {
1670 debug!("all_constructors({:?})", pcx.ty);
1671 let make_range = |start, end| {
1673 // `unwrap()` is ok because we know the type is an integer.
1674 IntRange::from_range(cx.tcx, start, end, pcx.ty, &RangeEnd::Included, pcx.span)
1678 match *pcx.ty.kind() {
1679 ty::Bool => vec![make_range(0, 1)],
1680 ty::Array(ref sub_ty, len) if len.try_eval_usize(cx.tcx, cx.param_env).is_some() => {
1681 let len = len.eval_usize(cx.tcx, cx.param_env);
1682 if len != 0 && cx.is_uninhabited(sub_ty) {
1685 vec![Slice(Slice { array_len: Some(len), kind: VarLen(0, 0) })]
1688 // Treat arrays of a constant but unknown length like slices.
1689 ty::Array(ref sub_ty, _) | ty::Slice(ref sub_ty) => {
1690 let kind = if cx.is_uninhabited(sub_ty) { FixedLen(0) } else { VarLen(0, 0) };
1691 vec![Slice(Slice { array_len: None, kind })]
1693 ty::Adt(def, substs) if def.is_enum() => {
1694 let ctors: Vec<_> = if cx.tcx.features().exhaustive_patterns {
1695 // If `exhaustive_patterns` is enabled, we exclude variants known to be
1700 !v.uninhabited_from(cx.tcx, substs, def.adt_kind(), cx.param_env)
1701 .contains(cx.tcx, cx.module)
1703 .map(|v| Variant(v.def_id))
1706 def.variants.iter().map(|v| Variant(v.def_id)).collect()
1709 // If the enum is declared as `#[non_exhaustive]`, we treat it as if it had an
1710 // additional "unknown" constructor.
1711 // There is no point in enumerating all possible variants, because the user can't
1712 // actually match against them all themselves. So we always return only the fictitious
1714 // E.g., in an example like:
1717 // let err: io::ErrorKind = ...;
1719 // io::ErrorKind::NotFound => {},
1723 // we don't want to show every possible IO error, but instead have only `_` as the
1725 let is_declared_nonexhaustive = cx.is_foreign_non_exhaustive_enum(pcx.ty);
1727 // If `exhaustive_patterns` is disabled and our scrutinee is an empty enum, we treat it
1728 // as though it had an "unknown" constructor to avoid exposing its emptyness. Note that
1729 // an empty match will still be considered exhaustive because that case is handled
1730 // separately in `check_match`.
1731 let is_secretly_empty =
1732 def.variants.is_empty() && !cx.tcx.features().exhaustive_patterns;
1734 if is_secretly_empty || is_declared_nonexhaustive { vec![NonExhaustive] } else { ctors }
1738 // The valid Unicode Scalar Value ranges.
1739 make_range('\u{0000}' as u128, '\u{D7FF}' as u128),
1740 make_range('\u{E000}' as u128, '\u{10FFFF}' as u128),
1743 ty::Int(_) | ty::Uint(_)
1744 if pcx.ty.is_ptr_sized_integral()
1745 && !cx.tcx.features().precise_pointer_size_matching =>
1747 // `usize`/`isize` are not allowed to be matched exhaustively unless the
1748 // `precise_pointer_size_matching` feature is enabled. So we treat those types like
1749 // `#[non_exhaustive]` enums by returning a special unmatcheable constructor.
1753 let bits = Integer::from_attr(&cx.tcx, SignedInt(ity)).size().bits() as u128;
1754 let min = 1u128 << (bits - 1);
1756 vec![make_range(min, max)]
1759 let size = Integer::from_attr(&cx.tcx, UnsignedInt(uty)).size();
1760 let max = truncate(u128::MAX, size);
1761 vec![make_range(0, max)]
1764 if cx.is_uninhabited(pcx.ty) {
1773 /// An inclusive interval, used for precise integer exhaustiveness checking.
1774 /// `IntRange`s always store a contiguous range. This means that values are
1775 /// encoded such that `0` encodes the minimum value for the integer,
1776 /// regardless of the signedness.
1777 /// For example, the pattern `-128..=127i8` is encoded as `0..=255`.
1778 /// This makes comparisons and arithmetic on interval endpoints much more
1779 /// straightforward. See `signed_bias` for details.
1781 /// `IntRange` is never used to encode an empty range or a "range" that wraps
1782 /// around the (offset) space: i.e., `range.lo <= range.hi`.
1783 #[derive(Clone, Debug)]
1784 struct IntRange<'tcx> {
1785 range: RangeInclusive<u128>,
1790 impl<'tcx> IntRange<'tcx> {
1792 fn is_integral(ty: Ty<'_>) -> bool {
1794 ty::Char | ty::Int(_) | ty::Uint(_) | ty::Bool => true,
1799 fn is_singleton(&self) -> bool {
1800 self.range.start() == self.range.end()
1803 fn boundaries(&self) -> (u128, u128) {
1804 (*self.range.start(), *self.range.end())
1807 /// Don't treat `usize`/`isize` exhaustively unless the `precise_pointer_size_matching` feature
1809 fn treat_exhaustively(&self, tcx: TyCtxt<'tcx>) -> bool {
1810 !self.ty.is_ptr_sized_integral() || tcx.features().precise_pointer_size_matching
1814 fn integral_size_and_signed_bias(tcx: TyCtxt<'tcx>, ty: Ty<'_>) -> Option<(Size, u128)> {
1816 ty::Bool => Some((Size::from_bytes(1), 0)),
1817 ty::Char => Some((Size::from_bytes(4), 0)),
1819 let size = Integer::from_attr(&tcx, SignedInt(ity)).size();
1820 Some((size, 1u128 << (size.bits() as u128 - 1)))
1822 ty::Uint(uty) => Some((Integer::from_attr(&tcx, UnsignedInt(uty)).size(), 0)),
1830 param_env: ty::ParamEnv<'tcx>,
1831 value: &Const<'tcx>,
1833 ) -> Option<IntRange<'tcx>> {
1834 if let Some((target_size, bias)) = Self::integral_size_and_signed_bias(tcx, value.ty) {
1837 if let ty::ConstKind::Value(ConstValue::Scalar(scalar)) = value.val {
1838 // For this specific pattern we can skip a lot of effort and go
1839 // straight to the result, after doing a bit of checking. (We
1840 // could remove this branch and just fall through, which
1841 // is more general but much slower.)
1842 if let Ok(bits) = scalar.to_bits_or_ptr(target_size, &tcx) {
1846 // This is a more general form of the previous case.
1847 value.try_eval_bits(tcx, param_env, ty)
1849 let val = val ^ bias;
1850 Some(IntRange { range: val..=val, ty, span })
1864 ) -> Option<IntRange<'tcx>> {
1865 if Self::is_integral(ty) {
1866 // Perform a shift if the underlying types are signed,
1867 // which makes the interval arithmetic simpler.
1868 let bias = IntRange::signed_bias(tcx, ty);
1869 let (lo, hi) = (lo ^ bias, hi ^ bias);
1870 let offset = (*end == RangeEnd::Excluded) as u128;
1871 if lo > hi || (lo == hi && *end == RangeEnd::Excluded) {
1872 // This should have been caught earlier by E0030.
1873 bug!("malformed range pattern: {}..={}", lo, (hi - offset));
1875 Some(IntRange { range: lo..=(hi - offset), ty, span })
1883 param_env: ty::ParamEnv<'tcx>,
1885 ) -> Option<IntRange<'tcx>> {
1886 // This MUST be kept in sync with `pat_constructor`.
1888 PatKind::AscribeUserType { .. } => bug!(), // Handled by `expand_pattern`
1889 PatKind::Or { .. } => bug!("Or-pattern should have been expanded earlier on."),
1891 PatKind::Binding { .. }
1893 | PatKind::Leaf { .. }
1894 | PatKind::Deref { .. }
1895 | PatKind::Variant { .. }
1896 | PatKind::Array { .. }
1897 | PatKind::Slice { .. } => None,
1899 PatKind::Constant { value } => Self::from_const(tcx, param_env, value, pat.span),
1901 PatKind::Range(PatRange { lo, hi, end }) => {
1905 lo.eval_bits(tcx, param_env, lo.ty),
1906 hi.eval_bits(tcx, param_env, hi.ty),
1915 // The return value of `signed_bias` should be XORed with an endpoint to encode/decode it.
1916 fn signed_bias(tcx: TyCtxt<'tcx>, ty: Ty<'tcx>) -> u128 {
1919 let bits = Integer::from_attr(&tcx, SignedInt(ity)).size().bits() as u128;
1926 /// Returns a collection of ranges that spans the values covered by `ranges`, subtracted
1927 /// by the values covered by `self`: i.e., `ranges \ self` (in set notation).
1928 fn subtract_from(&self, ranges: Vec<IntRange<'tcx>>) -> Vec<IntRange<'tcx>> {
1929 let mut remaining_ranges = vec![];
1931 let span = self.span;
1932 let (lo, hi) = self.boundaries();
1933 for subrange in ranges {
1934 let (subrange_lo, subrange_hi) = subrange.range.into_inner();
1935 if lo > subrange_hi || subrange_lo > hi {
1936 // The pattern doesn't intersect with the subrange at all,
1937 // so the subrange remains untouched.
1938 remaining_ranges.push(IntRange { range: subrange_lo..=subrange_hi, ty, span });
1940 if lo > subrange_lo {
1941 // The pattern intersects an upper section of the
1942 // subrange, so a lower section will remain.
1943 remaining_ranges.push(IntRange { range: subrange_lo..=(lo - 1), ty, span });
1945 if hi < subrange_hi {
1946 // The pattern intersects a lower section of the
1947 // subrange, so an upper section will remain.
1948 remaining_ranges.push(IntRange { range: (hi + 1)..=subrange_hi, ty, span });
1955 fn is_subrange(&self, other: &Self) -> bool {
1956 other.range.start() <= self.range.start() && self.range.end() <= other.range.end()
1959 fn intersection(&self, tcx: TyCtxt<'tcx>, other: &Self) -> Option<Self> {
1961 let (lo, hi) = self.boundaries();
1962 let (other_lo, other_hi) = other.boundaries();
1963 if self.treat_exhaustively(tcx) {
1964 if lo <= other_hi && other_lo <= hi {
1965 let span = other.span;
1966 Some(IntRange { range: max(lo, other_lo)..=min(hi, other_hi), ty, span })
1971 // If the range should not be treated exhaustively, fallback to checking for inclusion.
1972 if self.is_subrange(other) { Some(self.clone()) } else { None }
1976 fn suspicious_intersection(&self, other: &Self) -> bool {
1977 // `false` in the following cases:
1978 // 1 ---- // 1 ---------- // 1 ---- // 1 ----
1979 // 2 ---------- // 2 ---- // 2 ---- // 2 ----
1981 // The following are currently `false`, but could be `true` in the future (#64007):
1982 // 1 --------- // 1 ---------
1983 // 2 ---------- // 2 ----------
1985 // `true` in the following cases:
1986 // 1 ------- // 1 -------
1987 // 2 -------- // 2 -------
1988 let (lo, hi) = self.boundaries();
1989 let (other_lo, other_hi) = other.boundaries();
1990 lo == other_hi || hi == other_lo
1993 fn to_pat(&self, tcx: TyCtxt<'tcx>) -> Pat<'tcx> {
1994 let (lo, hi) = self.boundaries();
1996 let bias = IntRange::signed_bias(tcx, self.ty);
1997 let (lo, hi) = (lo ^ bias, hi ^ bias);
1999 let ty = ty::ParamEnv::empty().and(self.ty);
2000 let lo_const = ty::Const::from_bits(tcx, lo, ty);
2001 let hi_const = ty::Const::from_bits(tcx, hi, ty);
2003 let kind = if lo == hi {
2004 PatKind::Constant { value: lo_const }
2006 PatKind::Range(PatRange { lo: lo_const, hi: hi_const, end: RangeEnd::Included })
2009 // This is a brand new pattern, so we don't reuse `self.span`.
2010 Pat { ty: self.ty, span: DUMMY_SP, kind: Box::new(kind) }
2013 /// For exhaustive integer matching, some constructors are grouped within other constructors
2014 /// (namely integer typed values are grouped within ranges). However, when specialising these
2015 /// constructors, we want to be specialising for the underlying constructors (the integers), not
2016 /// the groups (the ranges). Thus we need to split the groups up. Splitting them up naïvely would
2017 /// mean creating a separate constructor for every single value in the range, which is clearly
2018 /// impractical. However, observe that for some ranges of integers, the specialisation will be
2019 /// identical across all values in that range (i.e., there are equivalence classes of ranges of
2020 /// constructors based on their `U(S(c, P), S(c, p))` outcome). These classes are grouped by
2021 /// the patterns that apply to them (in the matrix `P`). We can split the range whenever the
2022 /// patterns that apply to that range (specifically: the patterns that *intersect* with that range)
2024 /// Our solution, therefore, is to split the range constructor into subranges at every single point
2025 /// the group of intersecting patterns changes (using the method described below).
2026 /// And voilà! We're testing precisely those ranges that we need to, without any exhaustive matching
2027 /// on actual integers. The nice thing about this is that the number of subranges is linear in the
2028 /// number of rows in the matrix (i.e., the number of cases in the `match` statement), so we don't
2029 /// need to be worried about matching over gargantuan ranges.
2031 /// Essentially, given the first column of a matrix representing ranges, looking like the following:
2033 /// |------| |----------| |-------| ||
2034 /// |-------| |-------| |----| ||
2037 /// We split the ranges up into equivalence classes so the ranges are no longer overlapping:
2039 /// |--|--|||-||||--||---|||-------| |-|||| ||
2041 /// The logic for determining how to split the ranges is fairly straightforward: we calculate
2042 /// boundaries for each interval range, sort them, then create constructors for each new interval
2043 /// between every pair of boundary points. (This essentially sums up to performing the intuitive
2044 /// merging operation depicted above.)
2047 cx: &MatchCheckCtxt<'p, 'tcx>,
2049 matrix: &Matrix<'p, 'tcx>,
2050 hir_id: Option<HirId>,
2051 ) -> SmallVec<[Constructor<'tcx>; 1]> {
2054 /// Represents a border between 2 integers. Because the intervals spanning borders
2055 /// must be able to cover every integer, we need to be able to represent
2056 /// 2^128 + 1 such borders.
2057 #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Debug)]
2063 // A function for extracting the borders of an integer interval.
2064 fn range_borders(r: IntRange<'_>) -> impl Iterator<Item = Border> {
2065 let (lo, hi) = r.range.into_inner();
2066 let from = Border::JustBefore(lo);
2067 let to = match hi.checked_add(1) {
2068 Some(m) => Border::JustBefore(m),
2069 None => Border::AfterMax,
2071 vec![from, to].into_iter()
2074 // Collect the span and range of all the intersecting ranges to lint on likely
2075 // incorrect range patterns. (#63987)
2076 let mut overlaps = vec![];
2077 // `borders` is the set of borders between equivalence classes: each equivalence
2078 // class lies between 2 borders.
2079 let row_borders = matrix
2083 IntRange::from_pat(cx.tcx, cx.param_env, row.head()).map(|r| (r, row.len()))
2085 .flat_map(|(range, row_len)| {
2086 let intersection = self.intersection(cx.tcx, &range);
2087 let should_lint = self.suspicious_intersection(&range);
2088 if let (Some(range), 1, true) = (&intersection, row_len, should_lint) {
2089 // FIXME: for now, only check for overlapping ranges on simple range
2090 // patterns. Otherwise with the current logic the following is detected
2092 // match (10u8, true) {
2093 // (0 ..= 125, false) => {}
2094 // (126 ..= 255, false) => {}
2095 // (0 ..= 255, true) => {}
2097 overlaps.push(range.clone());
2101 .flat_map(range_borders);
2102 let self_borders = range_borders(self.clone());
2103 let mut borders: Vec<_> = row_borders.chain(self_borders).collect();
2104 borders.sort_unstable();
2106 self.lint_overlapping_patterns(cx.tcx, hir_id, ty, overlaps);
2108 // We're going to iterate through every adjacent pair of borders, making sure that
2109 // each represents an interval of nonnegative length, and convert each such
2110 // interval into a constructor.
2113 .filter_map(|&pair| match pair {
2114 [Border::JustBefore(n), Border::JustBefore(m)] => {
2121 [Border::JustBefore(n), Border::AfterMax] => Some(n..=u128::MAX),
2122 [Border::AfterMax, _] => None,
2124 .map(|range| IntRange { range, ty, span: pcx.span })
2129 fn lint_overlapping_patterns(
2132 hir_id: Option<HirId>,
2134 overlaps: Vec<IntRange<'tcx>>,
2136 if let (true, Some(hir_id)) = (!overlaps.is_empty(), hir_id) {
2137 tcx.struct_span_lint_hir(
2138 lint::builtin::OVERLAPPING_PATTERNS,
2142 let mut err = lint.build("multiple patterns covering the same range");
2143 err.span_label(self.span, "overlapping patterns");
2144 for int_range in overlaps {
2145 // Use the real type for user display of the ranges:
2149 "this range overlaps on `{}`",
2150 IntRange { range: int_range.range, ty, span: DUMMY_SP }.to_pat(tcx),
2161 /// Ignore spans when comparing, they don't carry semantic information as they are only for lints.
2162 impl<'tcx> std::cmp::PartialEq for IntRange<'tcx> {
2163 fn eq(&self, other: &Self) -> bool {
2164 self.range == other.range && self.ty == other.ty
2168 // A struct to compute a set of constructors equivalent to `all_ctors \ used_ctors`.
2169 struct MissingConstructors<'tcx> {
2170 all_ctors: Vec<Constructor<'tcx>>,
2171 used_ctors: Vec<Constructor<'tcx>>,
2174 impl<'tcx> MissingConstructors<'tcx> {
2175 fn new(all_ctors: Vec<Constructor<'tcx>>, used_ctors: Vec<Constructor<'tcx>>) -> Self {
2176 MissingConstructors { all_ctors, used_ctors }
2179 fn into_inner(self) -> (Vec<Constructor<'tcx>>, Vec<Constructor<'tcx>>) {
2180 (self.all_ctors, self.used_ctors)
2183 fn is_empty(&self) -> bool {
2184 self.iter().next().is_none()
2186 /// Whether this contains all the constructors for the given type or only a
2188 fn all_ctors_are_missing(&self) -> bool {
2189 self.used_ctors.is_empty()
2192 /// Iterate over all_ctors \ used_ctors
2193 fn iter<'a>(&'a self) -> impl Iterator<Item = Constructor<'tcx>> + Captures<'a> {
2194 self.all_ctors.iter().flat_map(move |req_ctor| req_ctor.subtract_ctors(&self.used_ctors))
2198 impl<'tcx> fmt::Debug for MissingConstructors<'tcx> {
2199 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
2200 let ctors: Vec<_> = self.iter().collect();
2201 write!(f, "{:?}", ctors)
2205 /// Algorithm from http://moscova.inria.fr/~maranget/papers/warn/index.html.
2206 /// The algorithm from the paper has been modified to correctly handle empty
2207 /// types. The changes are:
2208 /// (0) We don't exit early if the pattern matrix has zero rows. We just
2209 /// continue to recurse over columns.
2210 /// (1) all_constructors will only return constructors that are statically
2211 /// possible. E.g., it will only return `Ok` for `Result<T, !>`.
2213 /// This finds whether a (row) vector `v` of patterns is 'useful' in relation
2214 /// to a set of such vectors `m` - this is defined as there being a set of
2215 /// inputs that will match `v` but not any of the sets in `m`.
2217 /// All the patterns at each column of the `matrix ++ v` matrix must have the same type.
2219 /// This is used both for reachability checking (if a pattern isn't useful in
2220 /// relation to preceding patterns, it is not reachable) and exhaustiveness
2221 /// checking (if a wildcard pattern is useful in relation to a matrix, the
2222 /// matrix isn't exhaustive).
2224 /// `is_under_guard` is used to inform if the pattern has a guard. If it
2225 /// has one it must not be inserted into the matrix. This shouldn't be
2226 /// relied on for soundness.
2227 crate fn is_useful<'p, 'tcx>(
2228 cx: &MatchCheckCtxt<'p, 'tcx>,
2229 matrix: &Matrix<'p, 'tcx>,
2230 v: &PatStack<'p, 'tcx>,
2231 witness_preference: WitnessPreference,
2233 is_under_guard: bool,
2235 ) -> Usefulness<'tcx> {
2236 let Matrix { patterns: rows, .. } = matrix;
2237 debug!("is_useful({:#?}, {:#?})", matrix, v);
2239 // The base case. We are pattern-matching on () and the return value is
2240 // based on whether our matrix has a row or not.
2241 // NOTE: This could potentially be optimized by checking rows.is_empty()
2242 // first and then, if v is non-empty, the return value is based on whether
2243 // the type of the tuple we're checking is inhabited or not.
2245 return if rows.is_empty() {
2246 Usefulness::new_useful(witness_preference)
2252 assert!(rows.iter().all(|r| r.len() == v.len()));
2254 // If the first pattern is an or-pattern, expand it.
2255 if let Some(vs) = v.expand_or_pat() {
2256 // We need to push the already-seen patterns into the matrix in order to detect redundant
2257 // branches like `Some(_) | Some(0)`. We also keep track of the unreachable subpatterns.
2258 let mut matrix = matrix.clone();
2259 // `Vec` of all the unreachable branches of the current or-pattern.
2260 let mut unreachable_branches = Vec::new();
2261 // Subpatterns that are unreachable from all branches. E.g. in the following case, the last
2262 // `true` is unreachable only from one branch, so it is overall reachable.
2265 // match (true, true) {
2266 // (true, true) => {}
2267 // (false | true, false | true) => {}
2270 let mut unreachable_subpats = FxHashSet::default();
2271 // Whether any branch at all is useful.
2272 let mut any_is_useful = false;
2275 let res = is_useful(cx, &matrix, &v, witness_preference, hir_id, is_under_guard, false);
2279 any_is_useful = true;
2280 // Initialize with the first set of unreachable subpatterns encountered.
2281 unreachable_subpats = pats.into_iter().collect();
2283 // Keep the patterns unreachable from both this and previous branches.
2284 unreachable_subpats =
2285 pats.into_iter().filter(|p| unreachable_subpats.contains(p)).collect();
2288 NotUseful => unreachable_branches.push(v.head().span),
2289 UsefulWithWitness(_) => {
2290 bug!("Encountered or-pat in `v` during exhaustiveness checking")
2293 // If pattern has a guard don't add it to the matrix
2294 if !is_under_guard {
2299 // Collect all the unreachable patterns.
2300 unreachable_branches.extend(unreachable_subpats);
2301 return Useful(unreachable_branches);
2307 // FIXME(Nadrieril): Hack to work around type normalization issues (see #72476).
2308 let ty = matrix.heads().next().map(|r| r.ty).unwrap_or(v.head().ty);
2309 let pcx = PatCtxt { ty, span: v.head().span };
2311 debug!("is_useful_expand_first_col: pcx={:#?}, expanding {:#?}", pcx, v.head());
2313 let ret = if let Some(constructor) = pat_constructor(cx.tcx, cx.param_env, v.head()) {
2314 debug!("is_useful - expanding constructor: {:#?}", constructor);
2316 .split(cx, pcx, matrix, Some(hir_id))
2319 is_useful_specialized(
2330 .find(|result| result.is_useful())
2331 .unwrap_or(NotUseful)
2333 debug!("is_useful - expanding wildcard");
2335 let used_ctors: Vec<Constructor<'_>> =
2336 matrix.heads().filter_map(|p| pat_constructor(cx.tcx, cx.param_env, p)).collect();
2337 debug!("is_useful_used_ctors = {:#?}", used_ctors);
2338 // `all_ctors` are all the constructors for the given type, which
2339 // should all be represented (or caught with the wild pattern `_`).
2340 let all_ctors = all_constructors(cx, pcx);
2341 debug!("is_useful_all_ctors = {:#?}", all_ctors);
2343 // `missing_ctors` is the set of constructors from the same type as the
2344 // first column of `matrix` that are matched only by wildcard patterns
2345 // from the first column.
2347 // Therefore, if there is some pattern that is unmatched by `matrix`,
2348 // it will still be unmatched if the first constructor is replaced by
2349 // any of the constructors in `missing_ctors`
2351 // Missing constructors are those that are not matched by any non-wildcard patterns in the
2352 // current column. We only fully construct them on-demand, because they're rarely used and
2354 let missing_ctors = MissingConstructors::new(all_ctors, used_ctors);
2356 debug!("is_useful_missing_ctors.empty()={:#?}", missing_ctors.is_empty(),);
2358 if missing_ctors.is_empty() {
2359 let (all_ctors, _) = missing_ctors.into_inner();
2362 .flat_map(|ctor| ctor.split(cx, pcx, matrix, None))
2364 is_useful_specialized(
2375 .find(|result| result.is_useful())
2376 .unwrap_or(NotUseful)
2378 let matrix = matrix.specialize_wildcard();
2379 let v = v.to_tail();
2381 is_useful(cx, &matrix, &v, witness_preference, hir_id, is_under_guard, false);
2383 // In this case, there's at least one "free"
2384 // constructor that is only matched against by
2385 // wildcard patterns.
2387 // There are 2 ways we can report a witness here.
2388 // Commonly, we can report all the "free"
2389 // constructors as witnesses, e.g., if we have:
2392 // enum Direction { N, S, E, W }
2393 // let Direction::N = ...;
2396 // we can report 3 witnesses: `S`, `E`, and `W`.
2398 // However, there is a case where we don't want
2399 // to do this and instead report a single `_` witness:
2400 // if the user didn't actually specify a constructor
2401 // in this arm, e.g., in
2404 // let x: (Direction, Direction, bool) = ...;
2405 // let (_, _, false) = x;
2408 // we don't want to show all 16 possible witnesses
2409 // `(<direction-1>, <direction-2>, true)` - we are
2410 // satisfied with `(_, _, true)`. In this case,
2411 // `used_ctors` is empty.
2412 // The exception is: if we are at the top-level, for example in an empty match, we
2413 // sometimes prefer reporting the list of constructors instead of just `_`.
2414 let report_ctors_rather_than_wildcard = is_top_level && !IntRange::is_integral(pcx.ty);
2415 if missing_ctors.all_ctors_are_missing() && !report_ctors_rather_than_wildcard {
2416 // All constructors are unused. Add a wild pattern
2417 // rather than each individual constructor.
2418 usefulness.apply_wildcard(pcx.ty)
2420 // Construct for each missing constructor a "wild" version of this
2421 // constructor, that matches everything that can be built with
2422 // it. For example, if `ctor` is a `Constructor::Variant` for
2423 // `Option::Some`, we get the pattern `Some(_)`.
2424 usefulness.apply_missing_ctors(cx, pcx.ty, &missing_ctors)
2428 debug!("is_useful::returns({:#?}, {:#?}) = {:?}", matrix, v, ret);
2432 /// A shorthand for the `U(S(c, P), S(c, q))` operation from the paper. I.e., `is_useful` applied
2433 /// to the specialised version of both the pattern matrix `P` and the new pattern `q`.
2434 fn is_useful_specialized<'p, 'tcx>(
2435 cx: &MatchCheckCtxt<'p, 'tcx>,
2436 matrix: &Matrix<'p, 'tcx>,
2437 v: &PatStack<'p, 'tcx>,
2438 ctor: Constructor<'tcx>,
2440 witness_preference: WitnessPreference,
2442 is_under_guard: bool,
2443 ) -> Usefulness<'tcx> {
2444 debug!("is_useful_specialized({:#?}, {:#?}, {:?})", v, ctor, ty);
2446 // We cache the result of `Fields::wildcards` because it is used a lot.
2447 let ctor_wild_subpatterns = Fields::wildcards(cx, &ctor, ty);
2448 let matrix = matrix.specialize_constructor(cx, &ctor, &ctor_wild_subpatterns);
2449 v.specialize_constructor(cx, &ctor, &ctor_wild_subpatterns, true)
2450 .map(|v| is_useful(cx, &matrix, &v, witness_preference, hir_id, is_under_guard, false))
2451 .map(|u| u.apply_constructor(cx, &ctor, ty, &ctor_wild_subpatterns))
2452 .unwrap_or(NotUseful)
2455 /// Determines the constructor that the given pattern can be specialized to.
2456 /// Returns `None` in case of a catch-all, which can't be specialized.
2457 fn pat_constructor<'tcx>(
2459 param_env: ty::ParamEnv<'tcx>,
2461 ) -> Option<Constructor<'tcx>> {
2462 // This MUST be kept in sync with `IntRange::from_pat`.
2464 PatKind::AscribeUserType { .. } => bug!(), // Handled by `expand_pattern`
2465 PatKind::Binding { .. } | PatKind::Wild => None,
2466 PatKind::Leaf { .. } | PatKind::Deref { .. } => Some(Single),
2467 PatKind::Variant { adt_def, variant_index, .. } => {
2468 Some(Variant(adt_def.variants[variant_index].def_id))
2470 PatKind::Constant { value } => {
2471 if let Some(int_range) = IntRange::from_const(tcx, param_env, value, pat.span) {
2472 Some(IntRange(int_range))
2474 match value.ty.kind() {
2475 ty::Float(_) => Some(FloatRange(value, value, RangeEnd::Included)),
2476 ty::Ref(_, t, _) if t.is_str() => Some(Str(value)),
2477 // All constants that can be structurally matched have already been expanded
2478 // into the corresponding `Pat`s by `const_to_pat`. Constants that remain are
2484 PatKind::Range(PatRange { lo, hi, end }) => {
2486 if let Some(int_range) = IntRange::from_range(
2488 lo.eval_bits(tcx, param_env, lo.ty),
2489 hi.eval_bits(tcx, param_env, hi.ty),
2494 Some(IntRange(int_range))
2496 Some(FloatRange(lo, hi, end))
2499 PatKind::Array { ref prefix, ref slice, ref suffix }
2500 | PatKind::Slice { ref prefix, ref slice, ref suffix } => {
2501 let array_len = match pat.ty.kind() {
2502 ty::Array(_, length) => Some(length.eval_usize(tcx, param_env)),
2503 ty::Slice(_) => None,
2504 _ => span_bug!(pat.span, "bad ty {:?} for slice pattern", pat.ty),
2506 let prefix = prefix.len() as u64;
2507 let suffix = suffix.len() as u64;
2509 if slice.is_some() { VarLen(prefix, suffix) } else { FixedLen(prefix + suffix) };
2510 Some(Slice(Slice { array_len, kind }))
2512 PatKind::Or { .. } => bug!("Or-pattern should have been expanded earlier on."),
2516 /// This is the main specialization step. It expands the pattern
2517 /// into `arity` patterns based on the constructor. For most patterns, the step is trivial,
2518 /// for instance tuple patterns are flattened and box patterns expand into their inner pattern.
2519 /// Returns `None` if the pattern does not have the given constructor.
2521 /// OTOH, slice patterns with a subslice pattern (tail @ ..) can be expanded into multiple
2522 /// different patterns.
2523 /// Structure patterns with a partial wild pattern (Foo { a: 42, .. }) have their missing
2524 /// fields filled with wild patterns.
2526 /// This is roughly the inverse of `Constructor::apply`.
2527 fn specialize_one_pattern<'p, 'tcx>(
2528 cx: &MatchCheckCtxt<'p, 'tcx>,
2530 constructor: &Constructor<'tcx>,
2531 ctor_wild_subpatterns: &Fields<'p, 'tcx>,
2532 is_its_own_ctor: bool, // Whether `constructor` is known to be derived from `pat`
2533 ) -> Option<Fields<'p, 'tcx>> {
2534 if pat.is_wildcard() {
2535 return Some(ctor_wild_subpatterns.clone());
2539 // `unwrap` is safe because `pat` is not a wildcard.
2540 let pat_ctor = pat_constructor(cx.tcx, cx.param_env, pat).unwrap();
2542 let result = match (constructor, &pat_ctor, pat.kind.as_ref()) {
2543 (Single, Single, PatKind::Leaf { subpatterns }) => {
2544 Some(ctor_wild_subpatterns.replace_with_fieldpats(subpatterns))
2546 (Single, Single, PatKind::Deref { subpattern }) => {
2547 Some(Fields::from_single_pattern(subpattern))
2549 (Variant(_), Variant(_), _) if constructor != &pat_ctor => None,
2550 (Variant(_), Variant(_), PatKind::Variant { subpatterns, .. }) => {
2551 Some(ctor_wild_subpatterns.replace_with_fieldpats(subpatterns))
2554 (IntRange(ctor_range), IntRange(pat_range), _) => {
2555 ctor_range.intersection(cx.tcx, &pat_range)?;
2556 // Constructor splitting should ensure that all intersections we encounter
2557 // are actually inclusions.
2558 assert!(ctor_range.is_subrange(&pat_range));
2559 Some(Fields::empty())
2561 (FloatRange(ctor_from, ctor_to, ctor_end), FloatRange(pat_from, pat_to, pat_end), _) => {
2562 let to = compare_const_vals(cx.tcx, ctor_to, pat_to, cx.param_env, ty)?;
2563 let from = compare_const_vals(cx.tcx, ctor_from, pat_from, cx.param_env, ty)?;
2564 let intersects = (from == Ordering::Greater || from == Ordering::Equal)
2565 && (to == Ordering::Less || (pat_end == ctor_end && to == Ordering::Equal));
2566 if intersects { Some(Fields::empty()) } else { None }
2568 (Str(ctor_val), Str(pat_val), _) => {
2569 // FIXME: there's probably a more direct way of comparing for equality
2570 let comparison = compare_const_vals(cx.tcx, ctor_val, pat_val, cx.param_env, ty)?;
2571 if comparison == Ordering::Equal { Some(Fields::empty()) } else { None }
2574 (Slice(ctor_slice), Slice(pat_slice), _)
2575 if !pat_slice.pattern_kind().covers_length(ctor_slice.arity()) =>
2582 PatKind::Array { prefix, suffix, .. } | PatKind::Slice { prefix, suffix, .. },
2584 // Number of subpatterns for the constructor
2585 let ctor_arity = ctor_slice.arity();
2587 // Replace the prefix and the suffix with the given patterns, leaving wildcards in
2588 // the middle if there was a subslice pattern `..`.
2589 let prefix = prefix.iter().enumerate();
2591 suffix.iter().enumerate().map(|(i, p)| (ctor_arity as usize - suffix.len() + i, p));
2592 Some(ctor_wild_subpatterns.replace_fields_indexed(prefix.chain(suffix)))
2595 // Only a wildcard pattern can match an opaque constant, unless we're specializing the
2596 // value against its own constructor. That happens when we call
2597 // `v.specialize_constructor(ctor)` with `ctor` obtained from `pat_constructor(v.head())`.
2598 // For example, in the following match, when we are dealing with the third branch, we will
2599 // specialize with an `Opaque` ctor. We want to ignore the second branch because opaque
2600 // constants should not be inspected, but we don't want to ignore the current (third)
2601 // branch, as that would cause us to always conclude that such a branch is unreachable.
2603 // #[derive(PartialEq)]
2605 // impl Eq for Foo {}
2606 // const FOO: Foo = Foo(42);
2608 // match (Foo(0), true) {
2610 // (FOO, true) => {}
2611 // (FOO, false) => {}
2614 (Opaque, Opaque, _) if is_its_own_ctor => Some(Fields::empty()),
2615 // We are trying to inspect an opaque constant. Thus we skip the row.
2616 (Opaque, _, _) | (_, Opaque, _) => None,
2617 // Only a wildcard pattern can match the special extra constructor.
2618 (NonExhaustive, _, _) => None,
2620 _ => bug!("trying to specialize pattern {:?} with constructor {:?}", pat, constructor),
2624 "specialize({:#?}, {:#?}, {:#?}) = {:#?}",
2625 pat, constructor, ctor_wild_subpatterns, result
2628 if let Some(fields) = &result {
2629 debug_assert_eq!(fields.len(), ctor_wild_subpatterns.len());