1 //! Note: tests specific to this file can be found in:
2 //! - ui/pattern/usefulness
4 //! - ui/consts/const_in_pattern
5 //! - ui/rfc-2008-non-exhaustive
6 //! - ui/half-open-range-patterns
7 //! - probably many others
8 //! I (Nadrieril) prefer to put new tests in `ui/pattern/usefulness` unless there's a specific
9 //! reason not to, for example if they depend on a particular feature like or_patterns.
11 //! This file includes the logic for exhaustiveness and usefulness checking for
12 //! pattern-matching. Specifically, given a list of patterns for a type, we can
14 //! (a) the patterns cover every possible constructor for the type (exhaustiveness)
15 //! (b) each pattern is necessary (usefulness)
17 //! The algorithm implemented here is a modified version of the one described in:
18 //! <http://moscova.inria.fr/~maranget/papers/warn/index.html>
19 //! However, to save future implementors from reading the original paper, we
20 //! summarise the algorithm here to hopefully save time and be a little clearer
21 //! (without being so rigorous).
25 //! The core of the algorithm revolves about a "usefulness" check. In particular, we
26 //! are trying to compute a predicate `U(P, p)` where `P` is a list of patterns (we refer to this as
27 //! a matrix). `U(P, p)` represents whether, given an existing list of patterns
28 //! `P_1 ..= P_m`, adding a new pattern `p` will be "useful" (that is, cover previously-
29 //! uncovered values of the type).
31 //! If we have this predicate, then we can easily compute both exhaustiveness of an
32 //! entire set of patterns and the individual usefulness of each one.
33 //! (a) the set of patterns is exhaustive iff `U(P, _)` is false (i.e., adding a wildcard
34 //! match doesn't increase the number of values we're matching)
35 //! (b) a pattern `P_i` is not useful if `U(P[0..=(i-1), P_i)` is false (i.e., adding a
36 //! pattern to those that have come before it doesn't increase the number of values
41 //! The idea that powers everything that is done in this file is the following: a value is made
42 //! from a constructor applied to some fields. Examples of constructors are `Some`, `None`, `(,)`
43 //! (the 2-tuple constructor), `Foo {..}` (the constructor for a struct `Foo`), and `2` (the
44 //! constructor for the number `2`). Fields are just a (possibly empty) list of values.
46 //! Some of the constructors listed above might feel weird: `None` and `2` don't take any
47 //! arguments. This is part of what makes constructors so general: we will consider plain values
48 //! like numbers and string literals to be constructors that take no arguments, also called "0-ary
49 //! constructors"; they are the simplest case of constructors. This allows us to see any value as
50 //! made up from a tree of constructors, each having a given number of children. For example:
51 //! `(None, Ok(0))` is made from 4 different constructors.
53 //! This idea can be extended to patterns: a pattern captures a set of possible values, and we can
54 //! describe this set using constructors. For example, `Err(_)` captures all values of the type
55 //! `Result<T, E>` that start with the `Err` constructor (for some choice of `T` and `E`). The
56 //! wildcard `_` captures all values of the given type starting with any of the constructors for
59 //! We use this to compute whether different patterns might capture a same value. Do the patterns
60 //! `Ok("foo")` and `Err(_)` capture a common value? The answer is no, because the first pattern
61 //! captures only values starting with the `Ok` constructor and the second only values starting
62 //! with the `Err` constructor. Do the patterns `Some(42)` and `Some(1..10)` intersect? They might,
63 //! since they both capture values starting with `Some`. To be certain, we need to dig under the
64 //! `Some` constructor and continue asking the question. This is the main idea behind the
65 //! exhaustiveness algorithm: by looking at patterns constructor-by-constructor, we can efficiently
66 //! figure out if some new pattern might capture a value that hadn't been captured by previous
69 //! Constructors are represented by the `Constructor` enum, and its fields by the `Fields` enum.
70 //! Most of the complexity of this file resides in transforming between patterns and
71 //! (`Constructor`, `Fields`) pairs, handling all the special cases correctly.
73 //! Caveat: this constructors/fields distinction doesn't quite cover every Rust value. For example
74 //! a value of type `Rc<u64>` doesn't fit this idea very well, nor do various other things.
75 //! However, this idea covers most of the cases that are relevant to exhaustiveness checking.
80 //! Recall that `U(P, p)` represents whether, given an existing list of patterns (aka matrix) `P`,
81 //! adding a new pattern `p` will cover previously-uncovered values of the type.
82 //! During the course of the algorithm, the rows of the matrix won't just be individual patterns,
83 //! but rather partially-deconstructed patterns in the form of a list of fields. The paper
84 //! calls those pattern-vectors, and we will call them pattern-stacks. The same holds for the
87 //! For example, say we have the following:
90 //! // x: (Option<bool>, Result<()>)
92 //! (Some(true), _) => {}
93 //! (None, Err(())) => {}
94 //! (None, Err(_)) => {}
98 //! Here, the matrix `P` starts as:
102 //! [(Some(true), _)],
103 //! [(None, Err(()))],
104 //! [(None, Err(_))],
108 //! We can tell it's not exhaustive, because `U(P, _)` is true (we're not covering
109 //! `[(Some(false), _)]`, for instance). In addition, row 3 is not useful, because
110 //! all the values it covers are already covered by row 2.
112 //! A list of patterns can be thought of as a stack, because we are mainly interested in the top of
113 //! the stack at any given point, and we can pop or apply constructors to get new pattern-stacks.
114 //! To match the paper, the top of the stack is at the beginning / on the left.
116 //! There are two important operations on pattern-stacks necessary to understand the algorithm:
118 //! 1. We can pop a given constructor off the top of a stack. This operation is called
119 //! `specialize`, and is denoted `S(c, p)` where `c` is a constructor (like `Some` or
120 //! `None`) and `p` a pattern-stack.
121 //! If the pattern on top of the stack can cover `c`, this removes the constructor and
122 //! pushes its arguments onto the stack. It also expands OR-patterns into distinct patterns.
123 //! Otherwise the pattern-stack is discarded.
124 //! This essentially filters those pattern-stacks whose top covers the constructor `c` and
125 //! discards the others.
127 //! For example, the first pattern above initially gives a stack `[(Some(true), _)]`. If we
128 //! pop the tuple constructor, we are left with `[Some(true), _]`, and if we then pop the
129 //! `Some` constructor we get `[true, _]`. If we had popped `None` instead, we would get
132 //! This returns zero or more new pattern-stacks, as follows. We look at the pattern `p_1`
133 //! on top of the stack, and we have four cases:
134 //! 1.1. `p_1 = c(r_1, .., r_a)`, i.e. the top of the stack has constructor `c`. We
135 //! push onto the stack the arguments of this constructor, and return the result:
136 //! r_1, .., r_a, p_2, .., p_n
137 //! 1.2. `p_1 = c'(r_1, .., r_a')` where `c ≠ c'`. We discard the current stack and
139 //! 1.3. `p_1 = _`. We push onto the stack as many wildcards as the constructor `c` has
140 //! arguments (its arity), and return the resulting stack:
141 //! _, .., _, p_2, .., p_n
142 //! 1.4. `p_1 = r_1 | r_2`. We expand the OR-pattern and then recurse on each resulting
144 //! S(c, (r_1, p_2, .., p_n))
145 //! S(c, (r_2, p_2, .., p_n))
147 //! 2. We can pop a wildcard off the top of the stack. This is called `S(_, p)`, where `p` is
148 //! a pattern-stack. Note: the paper calls this `D(p)`.
149 //! This is used when we know there are missing constructor cases, but there might be
150 //! existing wildcard patterns, so to check the usefulness of the matrix, we have to check
151 //! all its *other* components.
153 //! It is computed as follows. We look at the pattern `p_1` on top of the stack,
154 //! and we have three cases:
155 //! 2.1. `p_1 = c(r_1, .., r_a)`. We discard the current stack and return nothing.
156 //! 2.2. `p_1 = _`. We return the rest of the stack:
158 //! 2.3. `p_1 = r_1 | r_2`. We expand the OR-pattern and then recurse on each resulting
160 //! S(_, (r_1, p_2, .., p_n))
161 //! S(_, (r_2, p_2, .., p_n))
163 //! Note that the OR-patterns are not always used directly in Rust, but are used to derive the
164 //! exhaustive integer matching rules, so they're written here for posterity.
166 //! Both those operations extend straightforwardly to a list or pattern-stacks, i.e. a matrix, by
167 //! working row-by-row. Popping a constructor ends up keeping only the matrix rows that start with
168 //! the given constructor, and popping a wildcard keeps those rows that start with a wildcard.
171 //! The algorithm for computing `U`
172 //! -------------------------------
173 //! The algorithm is inductive (on the number of columns: i.e., components of tuple patterns).
174 //! That means we're going to check the components from left-to-right, so the algorithm
175 //! operates principally on the first component of the matrix and new pattern-stack `p`.
176 //! This algorithm is realised in the `is_useful` function.
178 //! Base case. (`n = 0`, i.e., an empty tuple pattern)
179 //! - If `P` already contains an empty pattern (i.e., if the number of patterns `m > 0`),
180 //! then `U(P, p)` is false.
181 //! - Otherwise, `P` must be empty, so `U(P, p)` is true.
183 //! Inductive step. (`n > 0`, i.e., whether there's at least one column
184 //! [which may then be expanded into further columns later])
185 //! We're going to match on the top of the new pattern-stack, `p_1`.
186 //! - If `p_1 == c(r_1, .., r_a)`, i.e. we have a constructor pattern.
187 //! Then, the usefulness of `p_1` can be reduced to whether it is useful when
188 //! we ignore all the patterns in the first column of `P` that involve other constructors.
189 //! This is where `S(c, P)` comes in:
190 //! `U(P, p) := U(S(c, P), S(c, p))`
192 //! For example, if `P` is:
201 //! and `p` is [Some(false), 0], then we don't care about row 2 since we know `p` only
202 //! matches values that row 2 doesn't. For row 1 however, we need to dig into the
203 //! arguments of `Some` to know whether some new value is covered. So we compute
204 //! `U([[true, _]], [false, 0])`.
206 //! - If `p_1 == _`, then we look at the list of constructors that appear in the first
207 //! component of the rows of `P`:
208 //! + If there are some constructors that aren't present, then we might think that the
209 //! wildcard `_` is useful, since it covers those constructors that weren't covered
211 //! That's almost correct, but only works if there were no wildcards in those first
212 //! components. So we need to check that `p` is useful with respect to the rows that
213 //! start with a wildcard, if there are any. This is where `S(_, x)` comes in:
214 //! `U(P, p) := U(S(_, P), S(_, p))`
216 //! For example, if `P` is:
221 //! [None, false, 1],
225 //! and `p` is [_, false, _], the `Some` constructor doesn't appear in `P`. So if we
226 //! only had row 2, we'd know that `p` is useful. However row 1 starts with a
227 //! wildcard, so we need to check whether `U([[true, _]], [false, 1])`.
229 //! + Otherwise, all possible constructors (for the relevant type) are present. In this
230 //! case we must check whether the wildcard pattern covers any unmatched value. For
231 //! that, we can think of the `_` pattern as a big OR-pattern that covers all
232 //! possible constructors. For `Option`, that would mean `_ = None | Some(_)` for
233 //! example. The wildcard pattern is useful in this case if it is useful when
234 //! specialized to one of the possible constructors. So we compute:
235 //! `U(P, p) := ∃(k ϵ constructors) U(S(k, P), S(k, p))`
237 //! For example, if `P` is:
246 //! and `p` is [_, false], both `None` and `Some` constructors appear in the first
247 //! components of `P`. We will therefore try popping both constructors in turn: we
248 //! compute `U([[true, _]], [_, false])` for the `Some` constructor, and `U([[false]],
249 //! [false])` for the `None` constructor. The first case returns true, so we know that
250 //! `p` is useful for `P`. Indeed, it matches `[Some(false), _]` that wasn't matched
253 //! - If `p_1 == r_1 | r_2`, then the usefulness depends on each `r_i` separately:
254 //! `U(P, p) := U(P, (r_1, p_2, .., p_n))
255 //! || U(P, (r_2, p_2, .., p_n))`
257 //! Modifications to the algorithm
258 //! ------------------------------
259 //! The algorithm in the paper doesn't cover some of the special cases that arise in Rust, for
260 //! example uninhabited types and variable-length slice patterns. These are drawn attention to
261 //! throughout the code below. I'll make a quick note here about how exhaustive integer matching is
262 //! accounted for, though.
264 //! Exhaustive integer matching
265 //! ---------------------------
266 //! An integer type can be thought of as a (huge) sum type: 1 | 2 | 3 | ...
267 //! So to support exhaustive integer matching, we can make use of the logic in the paper for
268 //! OR-patterns. However, we obviously can't just treat ranges x..=y as individual sums, because
269 //! they are likely gigantic. So we instead treat ranges as constructors of the integers. This means
270 //! that we have a constructor *of* constructors (the integers themselves). We then need to work
271 //! through all the inductive step rules above, deriving how the ranges would be treated as
272 //! OR-patterns, and making sure that they're treated in the same way even when they're ranges.
273 //! There are really only four special cases here:
274 //! - When we match on a constructor that's actually a range, we have to treat it as if we would
276 //! + It turns out that we can simply extend the case for single-value patterns in
277 //! `specialize` to either be *equal* to a value constructor, or *contained within* a range
279 //! + When the pattern itself is a range, you just want to tell whether any of the values in
280 //! the pattern range coincide with values in the constructor range, which is precisely
282 //! Since when encountering a range pattern for a value constructor, we also use inclusion, it
283 //! means that whenever the constructor is a value/range and the pattern is also a value/range,
284 //! we can simply use intersection to test usefulness.
285 //! - When we're testing for usefulness of a pattern and the pattern's first component is a
287 //! + If all the constructors appear in the matrix, we have a slight complication. By default,
288 //! the behaviour (i.e., a disjunction over specialised matrices for each constructor) is
289 //! invalid, because we want a disjunction over every *integer* in each range, not just a
290 //! disjunction over every range. This is a bit more tricky to deal with: essentially we need
291 //! to form equivalence classes of subranges of the constructor range for which the behaviour
292 //! of the matrix `P` and new pattern `p` are the same. This is described in more
293 //! detail in `Constructor::split`.
294 //! + If some constructors are missing from the matrix, it turns out we don't need to do
295 //! anything special (because we know none of the integers are actually wildcards: i.e., we
296 //! can't span wildcards using ranges).
297 use self::Constructor::*;
298 use self::SliceKind::*;
299 use self::Usefulness::*;
300 use self::WitnessPreference::*;
302 use rustc_data_structures::captures::Captures;
303 use rustc_data_structures::fx::FxHashSet;
304 use rustc_data_structures::sync::OnceCell;
305 use rustc_index::vec::Idx;
307 use super::{compare_const_vals, PatternFoldable, PatternFolder};
308 use super::{FieldPat, Pat, PatKind, PatRange};
310 use rustc_arena::TypedArena;
311 use rustc_attr::{SignedInt, UnsignedInt};
312 use rustc_hir::def_id::DefId;
313 use rustc_hir::{HirId, RangeEnd};
314 use rustc_middle::mir::interpret::ConstValue;
315 use rustc_middle::mir::Field;
316 use rustc_middle::ty::layout::IntegerExt;
317 use rustc_middle::ty::{self, Const, Ty, TyCtxt};
318 use rustc_session::lint;
319 use rustc_span::{Span, DUMMY_SP};
320 use rustc_target::abi::{Integer, Size, VariantIdx};
322 use smallvec::{smallvec, SmallVec};
323 use std::cmp::{self, max, min, Ordering};
325 use std::iter::{FromIterator, IntoIterator};
326 use std::ops::RangeInclusive;
328 crate fn expand_pattern<'tcx>(pat: Pat<'tcx>) -> Pat<'tcx> {
329 LiteralExpander.fold_pattern(&pat)
332 struct LiteralExpander;
334 impl<'tcx> PatternFolder<'tcx> for LiteralExpander {
335 fn fold_pattern(&mut self, pat: &Pat<'tcx>) -> Pat<'tcx> {
336 debug!("fold_pattern {:?} {:?} {:?}", pat, pat.ty.kind(), pat.kind);
337 match (pat.ty.kind(), pat.kind.as_ref()) {
338 (_, PatKind::Binding { subpattern: Some(s), .. }) => s.fold_with(self),
339 (_, PatKind::AscribeUserType { subpattern: s, .. }) => s.fold_with(self),
340 (ty::Ref(_, t, _), PatKind::Constant { .. }) if t.is_str() => {
341 // Treat string literal patterns as deref patterns to a `str` constant, i.e.
342 // `&CONST`. This expands them like other const patterns. This could have been done
343 // in `const_to_pat`, but that causes issues with the rest of the matching code.
344 let mut new_pat = pat.super_fold_with(self);
345 // Make a fake const pattern of type `str` (instead of `&str`). That the carried
346 // constant value still knows it is of type `&str`.
349 kind: Box::new(PatKind::Deref { subpattern: new_pat }),
354 _ => pat.super_fold_with(self),
359 impl<'tcx> Pat<'tcx> {
360 pub(super) fn is_wildcard(&self) -> bool {
361 matches!(*self.kind, PatKind::Binding { subpattern: None, .. } | PatKind::Wild)
365 /// A row of a matrix. Rows of len 1 are very common, which is why `SmallVec[_; 2]`
367 #[derive(Debug, Clone)]
368 struct PatStack<'p, 'tcx> {
369 pats: SmallVec<[&'p Pat<'tcx>; 2]>,
370 /// Cache for the constructor of the head
371 head_ctor: OnceCell<Constructor<'tcx>>,
374 impl<'p, 'tcx> PatStack<'p, 'tcx> {
375 fn from_pattern(pat: &'p Pat<'tcx>) -> Self {
376 Self::from_vec(smallvec![pat])
379 fn from_vec(vec: SmallVec<[&'p Pat<'tcx>; 2]>) -> Self {
380 PatStack { pats: vec, head_ctor: OnceCell::new() }
383 fn is_empty(&self) -> bool {
387 fn len(&self) -> usize {
391 fn head(&self) -> &'p Pat<'tcx> {
395 fn head_ctor<'a>(&'a self, cx: &MatchCheckCtxt<'p, 'tcx>) -> &'a Constructor<'tcx> {
396 self.head_ctor.get_or_init(|| pat_constructor(cx, self.head()))
399 fn iter(&self) -> impl Iterator<Item = &Pat<'tcx>> {
400 self.pats.iter().copied()
403 // If the first pattern is an or-pattern, expand this pattern. Otherwise, return `None`.
404 fn expand_or_pat(&self) -> Option<Vec<Self>> {
407 } else if let PatKind::Or { pats } = &*self.head().kind {
411 let mut new_patstack = PatStack::from_pattern(pat);
412 new_patstack.pats.extend_from_slice(&self.pats[1..]);
422 /// This computes `S(self.head_ctor(), self)`. See top of the file for explanations.
424 /// Structure patterns with a partial wild pattern (Foo { a: 42, .. }) have their missing
425 /// fields filled with wild patterns.
427 /// This is roughly the inverse of `Constructor::apply`.
428 fn pop_head_constructor(&self, ctor_wild_subpatterns: &Fields<'p, 'tcx>) -> PatStack<'p, 'tcx> {
429 // We pop the head pattern and push the new fields extracted from the arguments of
431 let new_fields = ctor_wild_subpatterns.replace_with_pattern_arguments(self.head());
432 new_fields.push_on_patstack(&self.pats[1..])
436 impl<'p, 'tcx> Default for PatStack<'p, 'tcx> {
437 fn default() -> Self {
438 Self::from_vec(smallvec![])
442 impl<'p, 'tcx> PartialEq for PatStack<'p, 'tcx> {
443 fn eq(&self, other: &Self) -> bool {
444 self.pats == other.pats
448 impl<'p, 'tcx> FromIterator<&'p Pat<'tcx>> for PatStack<'p, 'tcx> {
449 fn from_iter<T>(iter: T) -> Self
451 T: IntoIterator<Item = &'p Pat<'tcx>>,
453 Self::from_vec(iter.into_iter().collect())
458 #[derive(Clone, PartialEq)]
459 struct Matrix<'p, 'tcx> {
460 patterns: Vec<PatStack<'p, 'tcx>>,
463 impl<'p, 'tcx> Matrix<'p, 'tcx> {
465 Matrix { patterns: vec![] }
468 /// Pushes a new row to the matrix. If the row starts with an or-pattern, this expands it.
469 fn push(&mut self, row: PatStack<'p, 'tcx>) {
470 if let Some(rows) = row.expand_or_pat() {
472 // We recursively expand the or-patterns of the new rows.
473 // This is necessary as we might have `0 | (1 | 2)` or e.g., `x @ 0 | x @ (1 | 2)`.
477 self.patterns.push(row);
481 /// Iterate over the first component of each row
482 fn heads<'a>(&'a self) -> impl Iterator<Item = &'a Pat<'tcx>> + Captures<'p> {
483 self.patterns.iter().map(|r| r.head())
486 /// Iterate over the first constructor of each row
489 cx: &'a MatchCheckCtxt<'p, 'tcx>,
490 ) -> impl Iterator<Item = &'a Constructor<'tcx>> + Captures<'a> + Captures<'p> {
491 self.patterns.iter().map(move |r| r.head_ctor(cx))
494 /// This computes `S(constructor, self)`. See top of the file for explanations.
495 fn specialize_constructor(
497 pcx: PatCtxt<'_, 'p, 'tcx>,
498 ctor: &Constructor<'tcx>,
499 ctor_wild_subpatterns: &Fields<'p, 'tcx>,
500 ) -> Matrix<'p, 'tcx> {
503 .filter(|r| ctor.is_covered_by(pcx, r.head_ctor(pcx.cx)))
504 .map(|r| r.pop_head_constructor(ctor_wild_subpatterns))
509 /// Pretty-printer for matrices of patterns, example:
512 /// +++++++++++++++++++++++++++++
514 /// +++++++++++++++++++++++++++++
515 /// + true + [First] +
516 /// +++++++++++++++++++++++++++++
517 /// + true + [Second(true)] +
518 /// +++++++++++++++++++++++++++++
520 /// +++++++++++++++++++++++++++++
521 /// + _ + [_, _, tail @ ..] +
522 /// +++++++++++++++++++++++++++++
524 impl<'p, 'tcx> fmt::Debug for Matrix<'p, 'tcx> {
525 fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
528 let Matrix { patterns: m, .. } = self;
529 let pretty_printed_matrix: Vec<Vec<String>> =
530 m.iter().map(|row| row.iter().map(|pat| format!("{:?}", pat)).collect()).collect();
532 let column_count = m.iter().map(|row| row.len()).max().unwrap_or(0);
533 assert!(m.iter().all(|row| row.len() == column_count));
534 let column_widths: Vec<usize> = (0..column_count)
535 .map(|col| pretty_printed_matrix.iter().map(|row| row[col].len()).max().unwrap_or(0))
538 let total_width = column_widths.iter().cloned().sum::<usize>() + column_count * 3 + 1;
539 let br = "+".repeat(total_width);
540 write!(f, "{}\n", br)?;
541 for row in pretty_printed_matrix {
543 for (column, pat_str) in row.into_iter().enumerate() {
545 write!(f, "{:1$}", pat_str, column_widths[column])?;
549 write!(f, "{}\n", br)?;
555 impl<'p, 'tcx> FromIterator<PatStack<'p, 'tcx>> for Matrix<'p, 'tcx> {
556 fn from_iter<T>(iter: T) -> Self
558 T: IntoIterator<Item = PatStack<'p, 'tcx>>,
560 let mut matrix = Matrix::empty();
562 // Using `push` ensures we correctly expand or-patterns.
569 crate struct MatchCheckCtxt<'a, 'tcx> {
570 crate tcx: TyCtxt<'tcx>,
571 /// The module in which the match occurs. This is necessary for
572 /// checking inhabited-ness of types because whether a type is (visibly)
573 /// inhabited can depend on whether it was defined in the current module or
574 /// not. E.g., `struct Foo { _private: ! }` cannot be seen to be empty
575 /// outside it's module and should not be matchable with an empty match
578 crate param_env: ty::ParamEnv<'tcx>,
579 crate pattern_arena: &'a TypedArena<Pat<'tcx>>,
582 impl<'a, 'tcx> MatchCheckCtxt<'a, 'tcx> {
583 fn is_uninhabited(&self, ty: Ty<'tcx>) -> bool {
584 if self.tcx.features().exhaustive_patterns {
585 self.tcx.is_ty_uninhabited_from(self.module, ty, self.param_env)
591 /// Returns whether the given type is an enum from another crate declared `#[non_exhaustive]`.
592 fn is_foreign_non_exhaustive_enum(&self, ty: Ty<'tcx>) -> bool {
594 ty::Adt(def, ..) => {
595 def.is_enum() && def.is_variant_list_non_exhaustive() && !def.did.is_local()
602 #[derive(Copy, Clone, Debug, PartialEq, Eq)]
604 /// Patterns of length `n` (`[x, y]`).
606 /// Patterns using the `..` notation (`[x, .., y]`).
607 /// Captures any array constructor of `length >= i + j`.
608 /// In the case where `array_len` is `Some(_)`,
609 /// this indicates that we only care about the first `i` and the last `j` values of the array,
610 /// and everything in between is a wildcard `_`.
615 fn arity(self) -> u64 {
617 FixedLen(length) => length,
618 VarLen(prefix, suffix) => prefix + suffix,
622 /// Whether this pattern includes patterns of length `other_len`.
623 fn covers_length(self, other_len: u64) -> bool {
625 FixedLen(len) => len == other_len,
626 VarLen(prefix, suffix) => prefix + suffix <= other_len,
631 /// A constructor for array and slice patterns.
632 #[derive(Copy, Clone, Debug, PartialEq, Eq)]
634 /// `None` if the matched value is a slice, `Some(n)` if it is an array of size `n`.
635 array_len: Option<u64>,
636 /// The kind of pattern it is: fixed-length `[x, y]` or variable length `[x, .., y]`.
641 fn new(array_len: Option<u64>, kind: SliceKind) -> Self {
642 let kind = match (array_len, kind) {
643 // If the middle `..` is empty, we effectively have a fixed-length pattern.
644 (Some(len), VarLen(prefix, suffix)) if prefix + suffix >= len => FixedLen(len),
647 Slice { array_len, kind }
650 fn arity(self) -> u64 {
654 /// The exhaustiveness-checking paper does not include any details on
655 /// checking variable-length slice patterns. However, they may be
656 /// matched by an infinite collection of fixed-length array patterns.
658 /// Checking the infinite set directly would take an infinite amount
659 /// of time. However, it turns out that for each finite set of
660 /// patterns `P`, all sufficiently large array lengths are equivalent:
662 /// Each slice `s` with a "sufficiently-large" length `l ≥ L` that applies
663 /// to exactly the subset `Pₜ` of `P` can be transformed to a slice
664 /// `sₘ` for each sufficiently-large length `m` that applies to exactly
665 /// the same subset of `P`.
667 /// Because of that, each witness for reachability-checking of one
668 /// of the sufficiently-large lengths can be transformed to an
669 /// equally-valid witness of any other length, so we only have
670 /// to check slices of the "minimal sufficiently-large length"
673 /// Note that the fact that there is a *single* `sₘ` for each `m`
674 /// not depending on the specific pattern in `P` is important: if
675 /// you look at the pair of patterns
678 /// Then any slice of length ≥1 that matches one of these two
679 /// patterns can be trivially turned to a slice of any
680 /// other length ≥1 that matches them and vice-versa,
681 /// but the slice of length 2 `[false, true]` that matches neither
682 /// of these patterns can't be turned to a slice from length 1 that
683 /// matches neither of these patterns, so we have to consider
684 /// slices from length 2 there.
686 /// Now, to see that that length exists and find it, observe that slice
687 /// patterns are either "fixed-length" patterns (`[_, _, _]`) or
688 /// "variable-length" patterns (`[_, .., _]`).
690 /// For fixed-length patterns, all slices with lengths *longer* than
691 /// the pattern's length have the same outcome (of not matching), so
692 /// as long as `L` is greater than the pattern's length we can pick
693 /// any `sₘ` from that length and get the same result.
695 /// For variable-length patterns, the situation is more complicated,
696 /// because as seen above the precise value of `sₘ` matters.
698 /// However, for each variable-length pattern `p` with a prefix of length
699 /// `plₚ` and suffix of length `slₚ`, only the first `plₚ` and the last
700 /// `slₚ` elements are examined.
702 /// Therefore, as long as `L` is positive (to avoid concerns about empty
703 /// types), all elements after the maximum prefix length and before
704 /// the maximum suffix length are not examined by any variable-length
705 /// pattern, and therefore can be added/removed without affecting
706 /// them - creating equivalent patterns from any sufficiently-large
709 /// Of course, if fixed-length patterns exist, we must be sure
710 /// that our length is large enough to miss them all, so
711 /// we can pick `L = max(max(FIXED_LEN)+1, max(PREFIX_LEN) + max(SUFFIX_LEN))`
713 /// for example, with the above pair of patterns, all elements
714 /// but the first and last can be added/removed, so any
715 /// witness of length ≥2 (say, `[false, false, true]`) can be
716 /// turned to a witness from any other length ≥2.
717 fn split<'p, 'tcx>(self, pcx: PatCtxt<'_, 'p, 'tcx>) -> SmallVec<[Constructor<'tcx>; 1]> {
718 let (self_prefix, self_suffix) = match self.kind {
719 VarLen(self_prefix, self_suffix) => (self_prefix, self_suffix),
720 _ => return smallvec![Slice(self)],
723 let head_ctors = pcx.matrix.head_ctors(pcx.cx).filter(|c| !c.is_wildcard());
725 let mut max_prefix_len = self_prefix;
726 let mut max_suffix_len = self_suffix;
727 let mut max_fixed_len = 0;
729 for ctor in head_ctors {
730 if let Slice(slice) = ctor {
733 max_fixed_len = cmp::max(max_fixed_len, len);
735 VarLen(prefix, suffix) => {
736 max_prefix_len = cmp::max(max_prefix_len, prefix);
737 max_suffix_len = cmp::max(max_suffix_len, suffix);
741 bug!("unexpected ctor for slice type: {:?}", ctor);
745 // For diagnostics, we keep the prefix and suffix lengths separate, so in the case
746 // where `max_fixed_len + 1` is the largest, we adapt `max_prefix_len` accordingly,
747 // so that `L = max_prefix_len + max_suffix_len`.
748 if max_fixed_len + 1 >= max_prefix_len + max_suffix_len {
749 // The subtraction can't overflow thanks to the above check.
750 // The new `max_prefix_len` is also guaranteed to be larger than its previous
752 max_prefix_len = max_fixed_len + 1 - max_suffix_len;
755 let final_slice = VarLen(max_prefix_len, max_suffix_len);
756 let final_slice = Slice::new(self.array_len, final_slice);
757 match self.array_len {
758 Some(_) => smallvec![Slice(final_slice)],
760 // `self` originally covered the range `(self.arity()..infinity)`. We split that
761 // range into two: lengths smaller than `final_slice.arity()` are treated
762 // independently as fixed-lengths slices, and lengths above are captured by
764 let smaller_lengths = (self.arity()..final_slice.arity()).map(FixedLen);
766 .map(|kind| Slice::new(self.array_len, kind))
767 .chain(Some(final_slice))
774 /// See `Constructor::is_covered_by`
775 fn is_covered_by(self, other: Self) -> bool {
776 other.kind.covers_length(self.arity())
780 /// A value can be decomposed into a constructor applied to some fields. This struct represents
781 /// the constructor. See also `Fields`.
783 /// `pat_constructor` retrieves the constructor corresponding to a pattern.
784 /// `specialize_constructor` returns the list of fields corresponding to a pattern, given a
785 /// constructor. `Constructor::apply` reconstructs the pattern from a pair of `Constructor` and
787 #[derive(Clone, Debug, PartialEq)]
788 enum Constructor<'tcx> {
789 /// The constructor for patterns that have a single constructor, like tuples, struct patterns
790 /// and fixed-length arrays.
794 /// Ranges of integer literal values (`2`, `2..=5` or `2..5`).
795 IntRange(IntRange<'tcx>),
796 /// Ranges of floating-point literal values (`2.0..=5.2`).
797 FloatRange(&'tcx ty::Const<'tcx>, &'tcx ty::Const<'tcx>, RangeEnd),
798 /// String literals. Strings are not quite the same as `&[u8]` so we treat them separately.
799 Str(&'tcx ty::Const<'tcx>),
800 /// Array and slice patterns.
802 /// Constants that must not be matched structurally. They are treated as black
803 /// boxes for the purposes of exhaustiveness: we must not inspect them, and they
804 /// don't count towards making a match exhaustive.
806 /// Fake extra constructor for enums that aren't allowed to be matched exhaustively. Also used
807 /// for those types for which we cannot list constructors explicitly, like `f64` and `str`.
809 /// Wildcard pattern.
813 impl<'tcx> Constructor<'tcx> {
814 fn is_wildcard(&self) -> bool {
815 matches!(self, Wildcard)
818 fn as_int_range(&self) -> Option<&IntRange<'tcx>> {
820 IntRange(range) => Some(range),
825 fn as_slice(&self) -> Option<Slice> {
827 Slice(slice) => Some(*slice),
832 fn variant_index_for_adt(&self, adt: &'tcx ty::AdtDef) -> VariantIdx {
834 Variant(id) => adt.variant_index_with_id(id),
836 assert!(!adt.is_enum());
839 _ => bug!("bad constructor {:?} for adt {:?}", self, adt),
843 /// Some constructors (namely `Wildcard`, `IntRange` and `Slice`) actually stand for a set of actual
844 /// constructors (like variants, integers or fixed-sized slices). When specializing for these
845 /// constructors, we want to be specialising for the actual underlying constructors.
846 /// Naively, we would simply return the list of constructors they correspond to. We instead are
847 /// more clever: if there are constructors that we know will behave the same wrt the current
848 /// matrix, we keep them grouped. For example, all slices of a sufficiently large length
849 /// will either be all useful or all non-useful with a given matrix.
851 /// See the branches for details on how the splitting is done.
853 /// This function may discard some irrelevant constructors if this preserves behavior and
854 /// diagnostics. Eg. for the `_` case, we ignore the constructors already present in the
855 /// matrix, unless all of them are.
857 /// `hir_id` is `None` when we're evaluating the wildcard pattern. In that case we do not want
858 /// to lint for overlapping ranges.
859 fn split<'p>(&self, pcx: PatCtxt<'_, 'p, 'tcx>, hir_id: Option<HirId>) -> SmallVec<[Self; 1]> {
860 debug!("Constructor::split({:#?}, {:#?})", self, pcx.matrix);
863 Wildcard => Constructor::split_wildcard(pcx),
864 // Fast-track if the range is trivial. In particular, we don't do the overlapping
867 if ctor_range.treat_exhaustively(pcx.cx.tcx) && !ctor_range.is_singleton() =>
869 ctor_range.split(pcx, hir_id)
871 Slice(slice @ Slice { kind: VarLen(..), .. }) => slice.split(pcx),
872 // Any other constructor can be used unchanged.
873 _ => smallvec![self.clone()],
877 /// For wildcards, there are two groups of constructors: there are the constructors actually
878 /// present in the matrix (`head_ctors`), and the constructors not present (`missing_ctors`).
879 /// Two constructors that are not in the matrix will either both be caught (by a wildcard), or
880 /// both not be caught. Therefore we can keep the missing constructors grouped together.
881 fn split_wildcard<'p>(pcx: PatCtxt<'_, 'p, 'tcx>) -> SmallVec<[Self; 1]> {
882 // Missing constructors are those that are not matched by any non-wildcard patterns in the
883 // current column. We only fully construct them on-demand, because they're rarely used and
885 let missing_ctors = MissingConstructors::new(pcx);
886 if missing_ctors.is_empty(pcx) {
887 // All the constructors are present in the matrix, so we just go through them all.
888 // We must also split them first.
889 missing_ctors.all_ctors
891 // Some constructors are missing, thus we can specialize with the wildcard constructor,
892 // which will stand for those constructors that are missing, and behaves like any of
898 /// Returns whether `self` is covered by `other`, i.e. whether `self` is a subset of `other`.
899 /// For the simple cases, this is simply checking for equality. For the "grouped" constructors,
900 /// this checks for inclusion.
901 fn is_covered_by<'p>(&self, pcx: PatCtxt<'_, 'p, 'tcx>, other: &Self) -> bool {
902 // This must be kept in sync with `is_covered_by_any`.
903 match (self, other) {
904 // Wildcards cover anything
905 (_, Wildcard) => true,
906 // Wildcards are only covered by wildcards
907 (Wildcard, _) => false,
909 (Single, Single) => true,
910 (Variant(self_id), Variant(other_id)) => self_id == other_id,
912 (IntRange(self_range), IntRange(other_range)) => {
913 self_range.is_covered_by(pcx, other_range)
916 FloatRange(self_from, self_to, self_end),
917 FloatRange(other_from, other_to, other_end),
920 compare_const_vals(pcx.cx.tcx, self_to, other_to, pcx.cx.param_env, pcx.ty),
921 compare_const_vals(pcx.cx.tcx, self_from, other_from, pcx.cx.param_env, pcx.ty),
923 (Some(to), Some(from)) => {
924 (from == Ordering::Greater || from == Ordering::Equal)
925 && (to == Ordering::Less
926 || (other_end == self_end && to == Ordering::Equal))
931 (Str(self_val), Str(other_val)) => {
932 // FIXME: there's probably a more direct way of comparing for equality
933 match compare_const_vals(pcx.cx.tcx, self_val, other_val, pcx.cx.param_env, pcx.ty)
935 Some(comparison) => comparison == Ordering::Equal,
939 (Slice(self_slice), Slice(other_slice)) => self_slice.is_covered_by(*other_slice),
941 // We are trying to inspect an opaque constant. Thus we skip the row.
942 (Opaque, _) | (_, Opaque) => false,
943 // Only a wildcard pattern can match the special extra constructor.
944 (NonExhaustive, _) => false,
948 "trying to compare incompatible constructors {:?} and {:?}",
955 /// Faster version of `is_covered_by` when applied to many constructors. `used_ctors` is
956 /// assumed to be built from `matrix.head_ctors()` with wildcards filtered out, and `self` is
957 /// assumed to have been split from a wildcard.
958 fn is_covered_by_any<'p>(
960 pcx: PatCtxt<'_, 'p, 'tcx>,
961 used_ctors: &[Constructor<'tcx>],
963 if used_ctors.is_empty() {
967 // This must be kept in sync with `is_covered_by`.
969 // If `self` is `Single`, `used_ctors` cannot contain anything else than `Single`s.
970 Single => !used_ctors.is_empty(),
971 Variant(_) => used_ctors.iter().any(|c| c == self),
972 IntRange(range) => used_ctors
974 .filter_map(|c| c.as_int_range())
975 .any(|other| range.is_covered_by(pcx, other)),
976 Slice(slice) => used_ctors
978 .filter_map(|c| c.as_slice())
979 .any(|other| slice.is_covered_by(other)),
980 // This constructor is never covered by anything else
981 NonExhaustive => false,
982 Str(..) | FloatRange(..) | Opaque | Wildcard => {
983 bug!("found unexpected ctor in all_ctors: {:?}", self)
988 /// Apply a constructor to a list of patterns, yielding a new pattern. `pats`
989 /// must have as many elements as this constructor's arity.
991 /// This is roughly the inverse of `specialize_constructor`.
994 /// `self`: `Constructor::Single`
995 /// `ty`: `(u32, u32, u32)`
996 /// `pats`: `[10, 20, _]`
997 /// returns `(10, 20, _)`
999 /// `self`: `Constructor::Variant(Option::Some)`
1000 /// `ty`: `Option<bool>`
1001 /// `pats`: `[false]`
1002 /// returns `Some(false)`
1003 fn apply<'p>(&self, pcx: PatCtxt<'_, 'p, 'tcx>, fields: Fields<'p, 'tcx>) -> Pat<'tcx> {
1004 let mut subpatterns = fields.all_patterns();
1006 let pat = match self {
1007 Single | Variant(_) => match pcx.ty.kind() {
1008 ty::Adt(..) | ty::Tuple(..) => {
1009 let subpatterns = subpatterns
1011 .map(|(i, p)| FieldPat { field: Field::new(i), pattern: p })
1014 if let ty::Adt(adt, substs) = pcx.ty.kind() {
1019 variant_index: self.variant_index_for_adt(adt),
1023 PatKind::Leaf { subpatterns }
1026 PatKind::Leaf { subpatterns }
1029 // Note: given the expansion of `&str` patterns done in `expand_pattern`, we should
1030 // be careful to reconstruct the correct constant pattern here. However a string
1031 // literal pattern will never be reported as a non-exhaustiveness witness, so we
1032 // can ignore this issue.
1033 ty::Ref(..) => PatKind::Deref { subpattern: subpatterns.next().unwrap() },
1034 ty::Slice(_) | ty::Array(..) => bug!("bad slice pattern {:?} {:?}", self, pcx.ty),
1037 Slice(slice) => match slice.kind {
1039 PatKind::Slice { prefix: subpatterns.collect(), slice: None, suffix: vec![] }
1041 VarLen(prefix, _) => {
1042 let mut prefix: Vec<_> = subpatterns.by_ref().take(prefix as usize).collect();
1043 if slice.array_len.is_some() {
1044 // Improves diagnostics a bit: if the type is a known-size array, instead
1045 // of reporting `[x, _, .., _, y]`, we prefer to report `[x, .., y]`.
1046 // This is incorrect if the size is not known, since `[_, ..]` captures
1047 // arrays of lengths `>= 1` whereas `[..]` captures any length.
1048 while !prefix.is_empty() && prefix.last().unwrap().is_wildcard() {
1052 let suffix: Vec<_> = if slice.array_len.is_some() {
1054 subpatterns.skip_while(Pat::is_wildcard).collect()
1056 subpatterns.collect()
1058 let wild = Pat::wildcard_from_ty(pcx.ty);
1059 PatKind::Slice { prefix, slice: Some(wild), suffix }
1062 &Str(value) => PatKind::Constant { value },
1063 &FloatRange(lo, hi, end) => PatKind::Range(PatRange { lo, hi, end }),
1064 IntRange(range) => return range.to_pat(pcx.cx.tcx),
1065 NonExhaustive => PatKind::Wild,
1066 Opaque => bug!("we should not try to apply an opaque constructor"),
1068 "trying to apply a wildcard constructor; this should have been done in `apply_constructors`"
1072 Pat { ty: pcx.ty, span: DUMMY_SP, kind: Box::new(pat) }
1076 /// Some fields need to be explicitly hidden away in certain cases; see the comment above the
1077 /// `Fields` struct. This struct represents such a potentially-hidden field. When a field is hidden
1078 /// we still keep its type around.
1079 #[derive(Debug, Copy, Clone)]
1080 enum FilteredField<'p, 'tcx> {
1081 Kept(&'p Pat<'tcx>),
1085 impl<'p, 'tcx> FilteredField<'p, 'tcx> {
1086 fn kept(self) -> Option<&'p Pat<'tcx>> {
1088 FilteredField::Kept(p) => Some(p),
1089 FilteredField::Hidden(_) => None,
1093 fn to_pattern(self) -> Pat<'tcx> {
1095 FilteredField::Kept(p) => p.clone(),
1096 FilteredField::Hidden(ty) => Pat::wildcard_from_ty(ty),
1101 /// A value can be decomposed into a constructor applied to some fields. This struct represents
1102 /// those fields, generalized to allow patterns in each field. See also `Constructor`.
1104 /// If a private or `non_exhaustive` field is uninhabited, the code mustn't observe that it is
1105 /// uninhabited. For that, we filter these fields out of the matrix. This is subtle because we
1106 /// still need to have those fields back when going to/from a `Pat`. Most of this is handled
1107 /// automatically in `Fields`, but when constructing or deconstructing `Fields` you need to be
1108 /// careful. As a rule, when going to/from the matrix, use the filtered field list; when going
1109 /// to/from `Pat`, use the full field list.
1110 /// This filtering is uncommon in practice, because uninhabited fields are rarely used, so we avoid
1111 /// it when possible to preserve performance.
1112 #[derive(Debug, Clone)]
1113 enum Fields<'p, 'tcx> {
1114 /// Lists of patterns that don't contain any filtered fields.
1115 /// `Slice` and `Vec` behave the same; the difference is only to avoid allocating and
1116 /// triple-dereferences when possible. Frankly this is premature optimization, I (Nadrieril)
1117 /// have not measured if it really made a difference.
1118 Slice(&'p [Pat<'tcx>]),
1119 Vec(SmallVec<[&'p Pat<'tcx>; 2]>),
1120 /// Patterns where some of the fields need to be hidden. `kept_count` caches the number of
1121 /// non-hidden fields.
1123 fields: SmallVec<[FilteredField<'p, 'tcx>; 2]>,
1128 impl<'p, 'tcx> Fields<'p, 'tcx> {
1129 fn empty() -> Self {
1133 /// Construct a new `Fields` from the given pattern. Must not be used if the pattern is a field
1134 /// of a struct/tuple/variant.
1135 fn from_single_pattern(pat: &'p Pat<'tcx>) -> Self {
1136 Fields::Slice(std::slice::from_ref(pat))
1139 /// Convenience; internal use.
1140 fn wildcards_from_tys(
1141 cx: &MatchCheckCtxt<'p, 'tcx>,
1142 tys: impl IntoIterator<Item = Ty<'tcx>>,
1144 let wilds = tys.into_iter().map(Pat::wildcard_from_ty);
1145 let pats = cx.pattern_arena.alloc_from_iter(wilds);
1149 /// Creates a new list of wildcard fields for a given constructor.
1150 fn wildcards(pcx: PatCtxt<'_, 'p, 'tcx>, constructor: &Constructor<'tcx>) -> Self {
1153 let wildcard_from_ty = |ty| &*cx.pattern_arena.alloc(Pat::wildcard_from_ty(ty));
1155 let ret = match constructor {
1156 Single | Variant(_) => match ty.kind() {
1157 ty::Tuple(ref fs) => {
1158 Fields::wildcards_from_tys(cx, fs.into_iter().map(|ty| ty.expect_ty()))
1160 ty::Ref(_, rty, _) => Fields::from_single_pattern(wildcard_from_ty(rty)),
1161 ty::Adt(adt, substs) => {
1163 // Use T as the sub pattern type of Box<T>.
1164 Fields::from_single_pattern(wildcard_from_ty(substs.type_at(0)))
1166 let variant = &adt.variants[constructor.variant_index_for_adt(adt)];
1167 // Whether we must not match the fields of this variant exhaustively.
1168 let is_non_exhaustive =
1169 variant.is_field_list_non_exhaustive() && !adt.did.is_local();
1170 let field_tys = variant.fields.iter().map(|field| field.ty(cx.tcx, substs));
1171 // In the following cases, we don't need to filter out any fields. This is
1172 // the vast majority of real cases, since uninhabited fields are uncommon.
1173 let has_no_hidden_fields = (adt.is_enum() && !is_non_exhaustive)
1174 || !field_tys.clone().any(|ty| cx.is_uninhabited(ty));
1176 if has_no_hidden_fields {
1177 Fields::wildcards_from_tys(cx, field_tys)
1179 let mut kept_count = 0;
1180 let fields = variant
1184 let ty = field.ty(cx.tcx, substs);
1185 let is_visible = adt.is_enum()
1186 || field.vis.is_accessible_from(cx.module, cx.tcx);
1187 let is_uninhabited = cx.is_uninhabited(ty);
1189 // In the cases of either a `#[non_exhaustive]` field list
1190 // or a non-public field, we hide uninhabited fields in
1191 // order not to reveal the uninhabitedness of the whole
1193 if is_uninhabited && (!is_visible || is_non_exhaustive) {
1194 FilteredField::Hidden(ty)
1197 FilteredField::Kept(wildcard_from_ty(ty))
1201 Fields::Filtered { fields, kept_count }
1205 _ => bug!("Unexpected type for `Single` constructor: {:?}", ty),
1207 Slice(slice) => match *ty.kind() {
1208 ty::Slice(ty) | ty::Array(ty, _) => {
1209 let arity = slice.arity();
1210 Fields::wildcards_from_tys(cx, (0..arity).map(|_| ty))
1212 _ => bug!("bad slice pattern {:?} {:?}", constructor, ty),
1214 Str(..) | FloatRange(..) | IntRange(..) | NonExhaustive | Opaque | Wildcard => {
1218 debug!("Fields::wildcards({:?}, {:?}) = {:#?}", constructor, ty, ret);
1222 /// Returns the number of patterns from the viewpoint of match-checking, i.e. excluding hidden
1223 /// fields. This is what we want in most cases in this file, the only exception being
1224 /// conversion to/from `Pat`.
1225 fn len(&self) -> usize {
1227 Fields::Slice(pats) => pats.len(),
1228 Fields::Vec(pats) => pats.len(),
1229 Fields::Filtered { kept_count, .. } => *kept_count,
1233 /// Returns the complete list of patterns, including hidden fields.
1234 fn all_patterns(self) -> impl Iterator<Item = Pat<'tcx>> {
1235 let pats: SmallVec<[_; 2]> = match self {
1236 Fields::Slice(pats) => pats.iter().cloned().collect(),
1237 Fields::Vec(pats) => pats.into_iter().cloned().collect(),
1238 Fields::Filtered { fields, .. } => {
1239 // We don't skip any fields here.
1240 fields.into_iter().map(|p| p.to_pattern()).collect()
1246 /// Overrides some of the fields with the provided patterns. Exactly like
1247 /// `replace_fields_indexed`, except that it takes `FieldPat`s as input.
1248 fn replace_with_fieldpats(
1250 new_pats: impl IntoIterator<Item = &'p FieldPat<'tcx>>,
1252 self.replace_fields_indexed(
1253 new_pats.into_iter().map(|pat| (pat.field.index(), &pat.pattern)),
1257 /// Overrides some of the fields with the provided patterns. This is used when a pattern
1258 /// defines some fields but not all, for example `Foo { field1: Some(_), .. }`: here we start with a
1259 /// `Fields` that is just one wildcard per field of the `Foo` struct, and override the entry
1260 /// corresponding to `field1` with the pattern `Some(_)`. This is also used for slice patterns
1261 /// for the same reason.
1262 fn replace_fields_indexed(
1264 new_pats: impl IntoIterator<Item = (usize, &'p Pat<'tcx>)>,
1266 let mut fields = self.clone();
1267 if let Fields::Slice(pats) = fields {
1268 fields = Fields::Vec(pats.iter().collect());
1272 Fields::Vec(pats) => {
1273 for (i, pat) in new_pats {
1277 Fields::Filtered { fields, .. } => {
1278 for (i, pat) in new_pats {
1279 if let FilteredField::Kept(p) = &mut fields[i] {
1284 Fields::Slice(_) => unreachable!(),
1289 /// Replaces contained fields with the given filtered list of patterns, e.g. taken from the
1290 /// matrix. There must be `len()` patterns in `pats`.
1293 cx: &MatchCheckCtxt<'p, 'tcx>,
1294 pats: impl IntoIterator<Item = Pat<'tcx>>,
1296 let pats: &[_] = cx.pattern_arena.alloc_from_iter(pats);
1299 Fields::Filtered { fields, kept_count } => {
1300 let mut pats = pats.iter();
1301 let mut fields = fields.clone();
1302 for f in &mut fields {
1303 if let FilteredField::Kept(p) = f {
1304 // We take one input pattern for each `Kept` field, in order.
1305 *p = pats.next().unwrap();
1308 Fields::Filtered { fields, kept_count: *kept_count }
1310 _ => Fields::Slice(pats),
1314 /// Replaces contained fields with the arguments of the given pattern. Only use on a pattern
1315 /// that is compatible with the constructor used to build `self`.
1316 /// This is meant to be used on the result of `Fields::wildcards()`. The idea is that
1317 /// `wildcards` constructs a list of fields where all entries are wildcards, and the pattern
1318 /// provided to this function fills some of the fields with non-wildcards.
1319 /// In the following example `Fields::wildcards` would return `[_, _, _, _]`. If we call
1320 /// `replace_with_pattern_arguments` on it with the pattern, the result will be `[Some(0), _,
1323 /// let x: [Option<u8>; 4] = foo();
1325 /// [Some(0), ..] => {}
1328 /// This is guaranteed to preserve the number of patterns in `self`.
1329 fn replace_with_pattern_arguments(&self, pat: &'p Pat<'tcx>) -> Self {
1330 match pat.kind.as_ref() {
1331 PatKind::Deref { subpattern } => {
1332 assert_eq!(self.len(), 1);
1333 Fields::from_single_pattern(subpattern)
1335 PatKind::Leaf { subpatterns } | PatKind::Variant { subpatterns, .. } => {
1336 self.replace_with_fieldpats(subpatterns)
1338 PatKind::Array { prefix, suffix, .. } | PatKind::Slice { prefix, suffix, .. } => {
1339 // Number of subpatterns for the constructor
1340 let ctor_arity = self.len();
1342 // Replace the prefix and the suffix with the given patterns, leaving wildcards in
1343 // the middle if there was a subslice pattern `..`.
1344 let prefix = prefix.iter().enumerate();
1346 suffix.iter().enumerate().map(|(i, p)| (ctor_arity - suffix.len() + i, p));
1347 self.replace_fields_indexed(prefix.chain(suffix))
1353 fn push_on_patstack(self, stack: &[&'p Pat<'tcx>]) -> PatStack<'p, 'tcx> {
1354 let pats: SmallVec<_> = match self {
1355 Fields::Slice(pats) => pats.iter().chain(stack.iter().copied()).collect(),
1356 Fields::Vec(mut pats) => {
1357 pats.extend_from_slice(stack);
1360 Fields::Filtered { fields, .. } => {
1361 // We skip hidden fields here
1362 fields.into_iter().filter_map(|p| p.kept()).chain(stack.iter().copied()).collect()
1365 PatStack::from_vec(pats)
1369 #[derive(Clone, Debug)]
1370 crate enum Usefulness<'tcx> {
1371 /// Carries, for each column in the matrix, a set of sub-branches that have been found to be
1372 /// unreachable. Used only in the presence of or-patterns, otherwise it stays empty.
1373 Useful(Vec<FxHashSet<Span>>),
1374 /// Carries a list of witnesses of non-exhaustiveness.
1375 UsefulWithWitness(Vec<Witness<'tcx>>),
1379 impl<'tcx> Usefulness<'tcx> {
1380 fn new_useful(preference: WitnessPreference) -> Self {
1382 ConstructWitness => UsefulWithWitness(vec![Witness(vec![])]),
1383 LeaveOutWitness => Useful(vec![]),
1387 fn is_useful(&self) -> bool {
1388 !matches!(*self, NotUseful)
1391 fn apply_constructor<'p>(
1393 pcx: PatCtxt<'_, 'p, 'tcx>,
1394 ctor: &Constructor<'tcx>,
1395 ctor_wild_subpatterns: &Fields<'p, 'tcx>,
1398 UsefulWithWitness(witnesses) => {
1399 let new_witnesses = if ctor.is_wildcard() {
1400 let missing_ctors = MissingConstructors::new(pcx);
1401 let new_patterns = missing_ctors.report_patterns(pcx);
1404 .flat_map(|witness| {
1405 new_patterns.iter().map(move |pat| {
1406 let mut witness = witness.clone();
1407 witness.0.push(pat.clone());
1415 .map(|witness| witness.apply_constructor(pcx, &ctor, ctor_wild_subpatterns))
1418 UsefulWithWitness(new_witnesses)
1420 Useful(mut unreachables) => {
1421 if !unreachables.is_empty() {
1422 // When we apply a constructor, there are `arity` columns of the matrix that
1423 // corresponded to its arguments. All the unreachables found in these columns
1424 // will, after `apply`, come from the first column. So we take the union of all
1425 // the corresponding sets and put them in the first column.
1426 // Note that `arity` may be 0, in which case we just push a new empty set.
1427 let len = unreachables.len();
1428 let arity = ctor_wild_subpatterns.len();
1429 let mut unioned = FxHashSet::default();
1430 for set in unreachables.drain((len - arity)..) {
1433 unreachables.push(unioned);
1435 Useful(unreachables)
1442 #[derive(Copy, Clone, Debug)]
1443 enum WitnessPreference {
1448 #[derive(Copy, Clone)]
1449 struct PatCtxt<'a, 'p, 'tcx> {
1450 cx: &'a MatchCheckCtxt<'p, 'tcx>,
1451 /// Current state of the matrix.
1452 matrix: &'a Matrix<'p, 'tcx>,
1453 /// Type of the current column under investigation.
1455 /// Span of the current pattern under investigation.
1457 /// Whether the current pattern is the whole pattern as found in a match arm, or if it's a
1462 /// A witness of non-exhaustiveness for error reporting, represented
1463 /// as a list of patterns (in reverse order of construction) with
1464 /// wildcards inside to represent elements that can take any inhabitant
1465 /// of the type as a value.
1467 /// A witness against a list of patterns should have the same types
1468 /// and length as the pattern matched against. Because Rust `match`
1469 /// is always against a single pattern, at the end the witness will
1470 /// have length 1, but in the middle of the algorithm, it can contain
1471 /// multiple patterns.
1473 /// For example, if we are constructing a witness for the match against
1476 /// struct Pair(Option<(u32, u32)>, bool);
1478 /// match (p: Pair) {
1479 /// Pair(None, _) => {}
1480 /// Pair(_, false) => {}
1484 /// We'll perform the following steps:
1485 /// 1. Start with an empty witness
1486 /// `Witness(vec![])`
1487 /// 2. Push a witness `Some(_)` against the `None`
1488 /// `Witness(vec![Some(_)])`
1489 /// 3. Push a witness `true` against the `false`
1490 /// `Witness(vec![Some(_), true])`
1491 /// 4. Apply the `Pair` constructor to the witnesses
1492 /// `Witness(vec![Pair(Some(_), true)])`
1494 /// The final `Pair(Some(_), true)` is then the resulting witness.
1495 #[derive(Clone, Debug)]
1496 crate struct Witness<'tcx>(Vec<Pat<'tcx>>);
1498 impl<'tcx> Witness<'tcx> {
1499 /// Asserts that the witness contains a single pattern, and returns it.
1500 fn single_pattern(self) -> Pat<'tcx> {
1501 assert_eq!(self.0.len(), 1);
1502 self.0.into_iter().next().unwrap()
1505 /// Constructs a partial witness for a pattern given a list of
1506 /// patterns expanded by the specialization step.
1508 /// When a pattern P is discovered to be useful, this function is used bottom-up
1509 /// to reconstruct a complete witness, e.g., a pattern P' that covers a subset
1510 /// of values, V, where each value in that set is not covered by any previously
1511 /// used patterns and is covered by the pattern P'. Examples:
1513 /// left_ty: tuple of 3 elements
1514 /// pats: [10, 20, _] => (10, 20, _)
1516 /// left_ty: struct X { a: (bool, &'static str), b: usize}
1517 /// pats: [(false, "foo"), 42] => X { a: (false, "foo"), b: 42 }
1518 fn apply_constructor<'p>(
1520 pcx: PatCtxt<'_, 'p, 'tcx>,
1521 ctor: &Constructor<'tcx>,
1522 ctor_wild_subpatterns: &Fields<'p, 'tcx>,
1525 let len = self.0.len();
1526 let arity = ctor_wild_subpatterns.len();
1527 let pats = self.0.drain((len - arity)..).rev();
1528 let fields = ctor_wild_subpatterns.replace_fields(pcx.cx, pats);
1529 ctor.apply(pcx, fields)
1538 /// This determines the set of all possible constructors of a pattern matching
1539 /// values of type `left_ty`. For vectors, this would normally be an infinite set
1540 /// but is instead bounded by the maximum fixed length of slice patterns in
1541 /// the column of patterns being analyzed.
1543 /// We make sure to omit constructors that are statically impossible. E.g., for
1544 /// `Option<!>`, we do not include `Some(_)` in the returned list of constructors.
1545 /// Invariant: this returns an empty `Vec` if and only if the type is uninhabited (as determined by
1546 /// `cx.is_uninhabited()`).
1547 fn all_constructors<'p, 'tcx>(pcx: PatCtxt<'_, 'p, 'tcx>) -> Vec<Constructor<'tcx>> {
1548 debug!("all_constructors({:?})", pcx.ty);
1550 let make_range = |start, end| {
1552 // `unwrap()` is ok because we know the type is an integer.
1553 IntRange::from_range(cx.tcx, start, end, pcx.ty, &RangeEnd::Included, pcx.span)
1557 match pcx.ty.kind() {
1558 ty::Bool => vec![make_range(0, 1)],
1559 ty::Array(sub_ty, len) if len.try_eval_usize(cx.tcx, cx.param_env).is_some() => {
1560 let len = len.eval_usize(cx.tcx, cx.param_env);
1561 if len != 0 && cx.is_uninhabited(sub_ty) {
1564 vec![Slice(Slice::new(Some(len), VarLen(0, 0)))]
1567 // Treat arrays of a constant but unknown length like slices.
1568 ty::Array(sub_ty, _) | ty::Slice(sub_ty) => {
1569 let kind = if cx.is_uninhabited(sub_ty) { FixedLen(0) } else { VarLen(0, 0) };
1570 vec![Slice(Slice::new(None, kind))]
1572 ty::Adt(def, substs) if def.is_enum() => {
1573 // If the enum is declared as `#[non_exhaustive]`, we treat it as if it had an
1574 // additional "unknown" constructor.
1575 // There is no point in enumerating all possible variants, because the user can't
1576 // actually match against them all themselves. So we always return only the fictitious
1578 // E.g., in an example like:
1581 // let err: io::ErrorKind = ...;
1583 // io::ErrorKind::NotFound => {},
1587 // we don't want to show every possible IO error, but instead have only `_` as the
1589 let is_declared_nonexhaustive = cx.is_foreign_non_exhaustive_enum(pcx.ty);
1591 // If `exhaustive_patterns` is disabled and our scrutinee is an empty enum, we treat it
1592 // as though it had an "unknown" constructor to avoid exposing its emptiness. The
1593 // exception is if the pattern is at the top level, because we want empty matches to be
1594 // considered exhaustive.
1595 let is_secretly_empty = def.variants.is_empty()
1596 && !cx.tcx.features().exhaustive_patterns
1597 && !pcx.is_top_level;
1599 if is_secretly_empty || is_declared_nonexhaustive {
1601 } else if cx.tcx.features().exhaustive_patterns {
1602 // If `exhaustive_patterns` is enabled, we exclude variants known to be
1607 !v.uninhabited_from(cx.tcx, substs, def.adt_kind(), cx.param_env)
1608 .contains(cx.tcx, cx.module)
1610 .map(|v| Variant(v.def_id))
1613 def.variants.iter().map(|v| Variant(v.def_id)).collect()
1618 // The valid Unicode Scalar Value ranges.
1619 make_range('\u{0000}' as u128, '\u{D7FF}' as u128),
1620 make_range('\u{E000}' as u128, '\u{10FFFF}' as u128),
1623 ty::Int(_) | ty::Uint(_)
1624 if pcx.ty.is_ptr_sized_integral()
1625 && !cx.tcx.features().precise_pointer_size_matching =>
1627 // `usize`/`isize` are not allowed to be matched exhaustively unless the
1628 // `precise_pointer_size_matching` feature is enabled. So we treat those types like
1629 // `#[non_exhaustive]` enums by returning a special unmatcheable constructor.
1633 let bits = Integer::from_attr(&cx.tcx, SignedInt(ity)).size().bits() as u128;
1634 let min = 1u128 << (bits - 1);
1636 vec![make_range(min, max)]
1639 let size = Integer::from_attr(&cx.tcx, UnsignedInt(uty)).size();
1640 let max = size.truncate(u128::MAX);
1641 vec![make_range(0, max)]
1643 // If `exhaustive_patterns` is disabled and our scrutinee is the never type, we cannot
1644 // expose its emptiness. The exception is if the pattern is at the top level, because we
1645 // want empty matches to be considered exhaustive.
1646 ty::Never if !cx.tcx.features().exhaustive_patterns && !pcx.is_top_level => {
1649 ty::Never => vec![],
1650 _ if cx.is_uninhabited(pcx.ty) => vec![],
1651 ty::Adt(..) | ty::Tuple(..) | ty::Ref(..) => vec![Single],
1652 // This type is one for which we cannot list constructors, like `str` or `f64`.
1653 _ => vec![NonExhaustive],
1657 /// An inclusive interval, used for precise integer exhaustiveness checking.
1658 /// `IntRange`s always store a contiguous range. This means that values are
1659 /// encoded such that `0` encodes the minimum value for the integer,
1660 /// regardless of the signedness.
1661 /// For example, the pattern `-128..=127i8` is encoded as `0..=255`.
1662 /// This makes comparisons and arithmetic on interval endpoints much more
1663 /// straightforward. See `signed_bias` for details.
1665 /// `IntRange` is never used to encode an empty range or a "range" that wraps
1666 /// around the (offset) space: i.e., `range.lo <= range.hi`.
1667 #[derive(Clone, Debug)]
1668 struct IntRange<'tcx> {
1669 range: RangeInclusive<u128>,
1674 impl<'tcx> IntRange<'tcx> {
1676 fn is_integral(ty: Ty<'_>) -> bool {
1677 matches!(ty.kind(), ty::Char | ty::Int(_) | ty::Uint(_) | ty::Bool)
1680 fn is_singleton(&self) -> bool {
1681 self.range.start() == self.range.end()
1684 fn boundaries(&self) -> (u128, u128) {
1685 (*self.range.start(), *self.range.end())
1688 /// Don't treat `usize`/`isize` exhaustively unless the `precise_pointer_size_matching` feature
1690 fn treat_exhaustively(&self, tcx: TyCtxt<'tcx>) -> bool {
1691 !self.ty.is_ptr_sized_integral() || tcx.features().precise_pointer_size_matching
1695 fn integral_size_and_signed_bias(tcx: TyCtxt<'tcx>, ty: Ty<'_>) -> Option<(Size, u128)> {
1697 ty::Bool => Some((Size::from_bytes(1), 0)),
1698 ty::Char => Some((Size::from_bytes(4), 0)),
1700 let size = Integer::from_attr(&tcx, SignedInt(ity)).size();
1701 Some((size, 1u128 << (size.bits() as u128 - 1)))
1703 ty::Uint(uty) => Some((Integer::from_attr(&tcx, UnsignedInt(uty)).size(), 0)),
1711 param_env: ty::ParamEnv<'tcx>,
1712 value: &Const<'tcx>,
1714 ) -> Option<IntRange<'tcx>> {
1715 if let Some((target_size, bias)) = Self::integral_size_and_signed_bias(tcx, value.ty) {
1718 if let ty::ConstKind::Value(ConstValue::Scalar(scalar)) = value.val {
1719 // For this specific pattern we can skip a lot of effort and go
1720 // straight to the result, after doing a bit of checking. (We
1721 // could remove this branch and just fall through, which
1722 // is more general but much slower.)
1723 if let Ok(bits) = scalar.to_bits_or_ptr(target_size, &tcx) {
1727 // This is a more general form of the previous case.
1728 value.try_eval_bits(tcx, param_env, ty)
1730 let val = val ^ bias;
1731 Some(IntRange { range: val..=val, ty, span })
1745 ) -> Option<IntRange<'tcx>> {
1746 if Self::is_integral(ty) {
1747 // Perform a shift if the underlying types are signed,
1748 // which makes the interval arithmetic simpler.
1749 let bias = IntRange::signed_bias(tcx, ty);
1750 let (lo, hi) = (lo ^ bias, hi ^ bias);
1751 let offset = (*end == RangeEnd::Excluded) as u128;
1752 if lo > hi || (lo == hi && *end == RangeEnd::Excluded) {
1753 // This should have been caught earlier by E0030.
1754 bug!("malformed range pattern: {}..={}", lo, (hi - offset));
1756 Some(IntRange { range: lo..=(hi - offset), ty, span })
1762 // The return value of `signed_bias` should be XORed with an endpoint to encode/decode it.
1763 fn signed_bias(tcx: TyCtxt<'tcx>, ty: Ty<'tcx>) -> u128 {
1766 let bits = Integer::from_attr(&tcx, SignedInt(ity)).size().bits() as u128;
1773 fn is_subrange(&self, other: &Self) -> bool {
1774 other.range.start() <= self.range.start() && self.range.end() <= other.range.end()
1777 fn intersection(&self, tcx: TyCtxt<'tcx>, other: &Self) -> Option<Self> {
1779 let (lo, hi) = self.boundaries();
1780 let (other_lo, other_hi) = other.boundaries();
1781 if self.treat_exhaustively(tcx) {
1782 if lo <= other_hi && other_lo <= hi {
1783 let span = other.span;
1784 Some(IntRange { range: max(lo, other_lo)..=min(hi, other_hi), ty, span })
1789 // If the range should not be treated exhaustively, fallback to checking for inclusion.
1790 if self.is_subrange(other) { Some(self.clone()) } else { None }
1794 fn suspicious_intersection(&self, other: &Self) -> bool {
1795 // `false` in the following cases:
1796 // 1 ---- // 1 ---------- // 1 ---- // 1 ----
1797 // 2 ---------- // 2 ---- // 2 ---- // 2 ----
1799 // The following are currently `false`, but could be `true` in the future (#64007):
1800 // 1 --------- // 1 ---------
1801 // 2 ---------- // 2 ----------
1803 // `true` in the following cases:
1804 // 1 ------- // 1 -------
1805 // 2 -------- // 2 -------
1806 let (lo, hi) = self.boundaries();
1807 let (other_lo, other_hi) = other.boundaries();
1808 lo == other_hi || hi == other_lo
1811 fn to_pat(&self, tcx: TyCtxt<'tcx>) -> Pat<'tcx> {
1812 let (lo, hi) = self.boundaries();
1814 let bias = IntRange::signed_bias(tcx, self.ty);
1815 let (lo, hi) = (lo ^ bias, hi ^ bias);
1817 let ty = ty::ParamEnv::empty().and(self.ty);
1818 let lo_const = ty::Const::from_bits(tcx, lo, ty);
1819 let hi_const = ty::Const::from_bits(tcx, hi, ty);
1821 let kind = if lo == hi {
1822 PatKind::Constant { value: lo_const }
1824 PatKind::Range(PatRange { lo: lo_const, hi: hi_const, end: RangeEnd::Included })
1827 // This is a brand new pattern, so we don't reuse `self.span`.
1828 Pat { ty: self.ty, span: DUMMY_SP, kind: Box::new(kind) }
1831 /// For exhaustive integer matching, some constructors are grouped within other constructors
1832 /// (namely integer typed values are grouped within ranges). However, when specialising these
1833 /// constructors, we want to be specialising for the underlying constructors (the integers), not
1834 /// the groups (the ranges). Thus we need to split the groups up. Splitting them up naïvely would
1835 /// mean creating a separate constructor for every single value in the range, which is clearly
1836 /// impractical. However, observe that for some ranges of integers, the specialisation will be
1837 /// identical across all values in that range (i.e., there are equivalence classes of ranges of
1838 /// constructors based on their `U(S(c, P), S(c, p))` outcome). These classes are grouped by
1839 /// the patterns that apply to them (in the matrix `P`). We can split the range whenever the
1840 /// patterns that apply to that range (specifically: the patterns that *intersect* with that range)
1842 /// Our solution, therefore, is to split the range constructor into subranges at every single point
1843 /// the group of intersecting patterns changes (using the method described below).
1844 /// And voilà! We're testing precisely those ranges that we need to, without any exhaustive matching
1845 /// on actual integers. The nice thing about this is that the number of subranges is linear in the
1846 /// number of rows in the matrix (i.e., the number of cases in the `match` statement), so we don't
1847 /// need to be worried about matching over gargantuan ranges.
1849 /// Essentially, given the first column of a matrix representing ranges, looking like the following:
1851 /// |------| |----------| |-------| ||
1852 /// |-------| |-------| |----| ||
1855 /// We split the ranges up into equivalence classes so the ranges are no longer overlapping:
1857 /// |--|--|||-||||--||---|||-------| |-|||| ||
1859 /// The logic for determining how to split the ranges is fairly straightforward: we calculate
1860 /// boundaries for each interval range, sort them, then create constructors for each new interval
1861 /// between every pair of boundary points. (This essentially sums up to performing the intuitive
1862 /// merging operation depicted above.)
1865 pcx: PatCtxt<'_, 'p, 'tcx>,
1866 hir_id: Option<HirId>,
1867 ) -> SmallVec<[Constructor<'tcx>; 1]> {
1870 /// Represents a border between 2 integers. Because the intervals spanning borders
1871 /// must be able to cover every integer, we need to be able to represent
1872 /// 2^128 + 1 such borders.
1873 #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Debug)]
1879 // A function for extracting the borders of an integer interval.
1880 fn range_borders(r: IntRange<'_>) -> impl Iterator<Item = Border> {
1881 let (lo, hi) = r.range.into_inner();
1882 let from = Border::JustBefore(lo);
1883 let to = match hi.checked_add(1) {
1884 Some(m) => Border::JustBefore(m),
1885 None => Border::AfterMax,
1887 vec![from, to].into_iter()
1890 // Collect the span and range of all the intersecting ranges to lint on likely
1891 // incorrect range patterns. (#63987)
1892 let mut overlaps = vec![];
1893 let row_len = pcx.matrix.patterns.get(0).map(|r| r.len()).unwrap_or(0);
1894 // `borders` is the set of borders between equivalence classes: each equivalence
1895 // class lies between 2 borders.
1896 let row_borders = pcx
1899 .filter_map(|ctor| ctor.as_int_range())
1900 .filter_map(|range| {
1901 let intersection = self.intersection(pcx.cx.tcx, &range);
1902 let should_lint = self.suspicious_intersection(&range);
1903 if let (Some(range), 1, true) = (&intersection, row_len, should_lint) {
1904 // FIXME: for now, only check for overlapping ranges on simple range
1905 // patterns. Otherwise with the current logic the following is detected
1907 // match (10u8, true) {
1908 // (0 ..= 125, false) => {}
1909 // (126 ..= 255, false) => {}
1910 // (0 ..= 255, true) => {}
1912 overlaps.push(range.clone());
1916 .flat_map(range_borders);
1917 let self_borders = range_borders(self.clone());
1918 let mut borders: Vec<_> = row_borders.chain(self_borders).collect();
1919 borders.sort_unstable();
1921 self.lint_overlapping_patterns(pcx.cx.tcx, hir_id, ty, overlaps);
1923 // We're going to iterate through every adjacent pair of borders, making sure that
1924 // each represents an interval of nonnegative length, and convert each such
1925 // interval into a constructor.
1928 .filter_map(|&pair| match pair {
1929 [Border::JustBefore(n), Border::JustBefore(m)] => {
1936 [Border::JustBefore(n), Border::AfterMax] => Some(n..=u128::MAX),
1937 [Border::AfterMax, _] => None,
1939 .map(|range| IntRange { range, ty, span: pcx.span })
1944 fn lint_overlapping_patterns(
1947 hir_id: Option<HirId>,
1949 overlaps: Vec<IntRange<'tcx>>,
1951 if let (true, Some(hir_id)) = (!overlaps.is_empty(), hir_id) {
1952 tcx.struct_span_lint_hir(
1953 lint::builtin::OVERLAPPING_PATTERNS,
1957 let mut err = lint.build("multiple patterns covering the same range");
1958 err.span_label(self.span, "overlapping patterns");
1959 for int_range in overlaps {
1960 // Use the real type for user display of the ranges:
1964 "this range overlaps on `{}`",
1965 IntRange { range: int_range.range, ty, span: DUMMY_SP }.to_pat(tcx),
1975 /// See `Constructor::is_covered_by`
1976 fn is_covered_by<'p>(&self, pcx: PatCtxt<'_, 'p, 'tcx>, other: &Self) -> bool {
1977 if self.intersection(pcx.cx.tcx, other).is_some() {
1978 // Constructor splitting should ensure that all intersections we encounter are actually
1980 assert!(self.is_subrange(other));
1988 /// Ignore spans when comparing, they don't carry semantic information as they are only for lints.
1989 impl<'tcx> std::cmp::PartialEq for IntRange<'tcx> {
1990 fn eq(&self, other: &Self) -> bool {
1991 self.range == other.range && self.ty == other.ty
1995 // A struct to compute a set of constructors equivalent to `all_ctors \ used_ctors`.
1997 struct MissingConstructors<'tcx> {
1998 all_ctors: SmallVec<[Constructor<'tcx>; 1]>,
1999 used_ctors: Vec<Constructor<'tcx>>,
2002 impl<'tcx> MissingConstructors<'tcx> {
2003 fn new<'p>(pcx: PatCtxt<'_, 'p, 'tcx>) -> Self {
2004 let used_ctors: Vec<Constructor<'_>> =
2005 pcx.matrix.head_ctors(pcx.cx).cloned().filter(|c| !c.is_wildcard()).collect();
2006 // Since `all_ctors` never contains wildcards, this won't recurse further.
2008 all_constructors(pcx).into_iter().flat_map(|ctor| ctor.split(pcx, None)).collect();
2010 MissingConstructors { all_ctors, used_ctors }
2013 fn is_empty<'p>(&self, pcx: PatCtxt<'_, 'p, 'tcx>) -> bool {
2014 self.iter(pcx).next().is_none()
2017 /// Iterate over all_ctors \ used_ctors
2020 pcx: PatCtxt<'a, 'p, 'tcx>,
2021 ) -> impl Iterator<Item = &'a Constructor<'tcx>> + Captures<'p> {
2022 self.all_ctors.iter().filter(move |ctor| !ctor.is_covered_by_any(pcx, &self.used_ctors))
2025 /// List the patterns corresponding to the missing constructors. In some cases, instead of
2026 /// listing all constructors of a given type, we prefer to simply report a wildcard.
2027 fn report_patterns<'p>(&self, pcx: PatCtxt<'_, 'p, 'tcx>) -> SmallVec<[Pat<'tcx>; 1]> {
2028 // There are 2 ways we can report a witness here.
2029 // Commonly, we can report all the "free"
2030 // constructors as witnesses, e.g., if we have:
2033 // enum Direction { N, S, E, W }
2034 // let Direction::N = ...;
2037 // we can report 3 witnesses: `S`, `E`, and `W`.
2039 // However, there is a case where we don't want
2040 // to do this and instead report a single `_` witness:
2041 // if the user didn't actually specify a constructor
2042 // in this arm, e.g., in
2045 // let x: (Direction, Direction, bool) = ...;
2046 // let (_, _, false) = x;
2049 // we don't want to show all 16 possible witnesses
2050 // `(<direction-1>, <direction-2>, true)` - we are
2051 // satisfied with `(_, _, true)`. In this case,
2052 // `used_ctors` is empty.
2053 // The exception is: if we are at the top-level, for example in an empty match, we
2054 // sometimes prefer reporting the list of constructors instead of just `_`.
2055 let report_when_all_missing = pcx.is_top_level && !IntRange::is_integral(pcx.ty);
2056 if self.used_ctors.is_empty() && !report_when_all_missing {
2057 // All constructors are unused. Report only a wildcard
2058 // rather than each individual constructor.
2059 smallvec![Pat::wildcard_from_ty(pcx.ty)]
2061 // Construct for each missing constructor a "wild" version of this
2062 // constructor, that matches everything that can be built with
2063 // it. For example, if `ctor` is a `Constructor::Variant` for
2064 // `Option::Some`, we get the pattern `Some(_)`.
2066 .map(|missing_ctor| {
2067 let fields = Fields::wildcards(pcx, &missing_ctor);
2068 missing_ctor.apply(pcx, fields)
2075 /// Algorithm from <http://moscova.inria.fr/~maranget/papers/warn/index.html>.
2076 /// The algorithm from the paper has been modified to correctly handle empty
2077 /// types. The changes are:
2078 /// (0) We don't exit early if the pattern matrix has zero rows. We just
2079 /// continue to recurse over columns.
2080 /// (1) all_constructors will only return constructors that are statically
2081 /// possible. E.g., it will only return `Ok` for `Result<T, !>`.
2083 /// This finds whether a (row) vector `v` of patterns is 'useful' in relation
2084 /// to a set of such vectors `m` - this is defined as there being a set of
2085 /// inputs that will match `v` but not any of the sets in `m`.
2087 /// All the patterns at each column of the `matrix ++ v` matrix must have the same type.
2089 /// This is used both for reachability checking (if a pattern isn't useful in
2090 /// relation to preceding patterns, it is not reachable) and exhaustiveness
2091 /// checking (if a wildcard pattern is useful in relation to a matrix, the
2092 /// matrix isn't exhaustive).
2094 /// `is_under_guard` is used to inform if the pattern has a guard. If it
2095 /// has one it must not be inserted into the matrix. This shouldn't be
2096 /// relied on for soundness.
2097 fn is_useful<'p, 'tcx>(
2098 cx: &MatchCheckCtxt<'p, 'tcx>,
2099 matrix: &Matrix<'p, 'tcx>,
2100 v: &PatStack<'p, 'tcx>,
2101 witness_preference: WitnessPreference,
2103 is_under_guard: bool,
2105 ) -> Usefulness<'tcx> {
2106 let Matrix { patterns: rows, .. } = matrix;
2107 debug!("is_useful({:#?}, {:#?})", matrix, v);
2109 // The base case. We are pattern-matching on () and the return value is
2110 // based on whether our matrix has a row or not.
2111 // NOTE: This could potentially be optimized by checking rows.is_empty()
2112 // first and then, if v is non-empty, the return value is based on whether
2113 // the type of the tuple we're checking is inhabited or not.
2115 return if rows.is_empty() {
2116 Usefulness::new_useful(witness_preference)
2122 assert!(rows.iter().all(|r| r.len() == v.len()));
2124 // If the first pattern is an or-pattern, expand it.
2125 if let Some(vs) = v.expand_or_pat() {
2126 // We expand the or pattern, trying each of its branches in turn and keeping careful track
2127 // of possible unreachable sub-branches.
2129 // If two branches have detected some unreachable sub-branches, we need to be careful. If
2130 // they were detected in columns that are not the current one, we want to keep only the
2131 // sub-branches that were unreachable in _all_ branches. Eg. in the following, the last
2132 // `true` is unreachable in the second branch of the first or-pattern, but not otherwise.
2133 // Therefore we don't want to lint that it is unreachable.
2136 // match (true, true) {
2137 // (true, true) => {}
2138 // (false | true, false | true) => {}
2141 // If however the sub-branches come from the current column, they come from the inside of
2142 // the current or-pattern, and we want to keep them all. Eg. in the following, we _do_ want
2143 // to lint that the last `false` is unreachable.
2146 // Some(false) => {}
2147 // None | Some(true | false) => {}
2151 let mut matrix = matrix.clone();
2152 // We keep track of sub-branches separately depending on whether they come from this column
2154 let mut unreachables_this_column: FxHashSet<Span> = FxHashSet::default();
2155 let mut unreachables_other_columns: Vec<FxHashSet<Span>> = Vec::default();
2156 // Whether at least one branch is reachable.
2157 let mut any_is_useful = false;
2160 let res = is_useful(cx, &matrix, &v, witness_preference, hir_id, is_under_guard, false);
2162 Useful(unreachables) => {
2163 if let Some((this_column, other_columns)) = unreachables.split_last() {
2164 // We keep the union of unreachables found in the first column.
2165 unreachables_this_column.extend(this_column);
2166 // We keep the intersection of unreachables found in other columns.
2167 if unreachables_other_columns.is_empty() {
2168 unreachables_other_columns = other_columns.to_vec();
2170 unreachables_other_columns = unreachables_other_columns
2173 .map(|(x, y)| x.intersection(&y).copied().collect())
2177 any_is_useful = true;
2180 unreachables_this_column.insert(v.head().span);
2182 UsefulWithWitness(_) => bug!(
2183 "encountered or-pat in the expansion of `_` during exhaustiveness checking"
2187 // If pattern has a guard don't add it to the matrix.
2188 if !is_under_guard {
2189 // We push the already-seen patterns into the matrix in order to detect redundant
2190 // branches like `Some(_) | Some(0)`.
2195 return if any_is_useful {
2196 let mut unreachables = if unreachables_other_columns.is_empty() {
2197 let n_columns = v.len();
2198 (0..n_columns - 1).map(|_| FxHashSet::default()).collect()
2200 unreachables_other_columns
2202 unreachables.push(unreachables_this_column);
2203 Useful(unreachables)
2209 // FIXME(Nadrieril): Hack to work around type normalization issues (see #72476).
2210 let ty = matrix.heads().next().map(|r| r.ty).unwrap_or(v.head().ty);
2211 let pcx = PatCtxt { cx, matrix, ty, span: v.head().span, is_top_level };
2213 debug!("is_useful_expand_first_col: ty={:#?}, expanding {:#?}", pcx.ty, v.head());
2217 .split(pcx, Some(hir_id))
2220 // We cache the result of `Fields::wildcards` because it is used a lot.
2221 let ctor_wild_subpatterns = Fields::wildcards(pcx, &ctor);
2222 let matrix = pcx.matrix.specialize_constructor(pcx, &ctor, &ctor_wild_subpatterns);
2223 let v = v.pop_head_constructor(&ctor_wild_subpatterns);
2225 is_useful(pcx.cx, &matrix, &v, witness_preference, hir_id, is_under_guard, false);
2226 usefulness.apply_constructor(pcx, &ctor, &ctor_wild_subpatterns)
2228 .find(|result| result.is_useful())
2229 .unwrap_or(NotUseful);
2230 debug!("is_useful::returns({:#?}, {:#?}) = {:?}", matrix, v, ret);
2234 /// Determines the constructor that the given pattern can be specialized to.
2235 /// Returns `None` in case of a catch-all, which can't be specialized.
2236 fn pat_constructor<'p, 'tcx>(
2237 cx: &MatchCheckCtxt<'p, 'tcx>,
2239 ) -> Constructor<'tcx> {
2240 match pat.kind.as_ref() {
2241 PatKind::AscribeUserType { .. } => bug!(), // Handled by `expand_pattern`
2242 PatKind::Binding { .. } | PatKind::Wild => Wildcard,
2243 PatKind::Leaf { .. } | PatKind::Deref { .. } => Single,
2244 &PatKind::Variant { adt_def, variant_index, .. } => {
2245 Variant(adt_def.variants[variant_index].def_id)
2247 PatKind::Constant { value } => {
2248 if let Some(int_range) = IntRange::from_const(cx.tcx, cx.param_env, value, pat.span) {
2251 match pat.ty.kind() {
2252 ty::Float(_) => FloatRange(value, value, RangeEnd::Included),
2253 // In `expand_pattern`, we convert string literals to `&CONST` patterns with
2254 // `CONST` a pattern of type `str`. In truth this contains a constant of type
2256 ty::Str => Str(value),
2257 // All constants that can be structurally matched have already been expanded
2258 // into the corresponding `Pat`s by `const_to_pat`. Constants that remain are
2264 &PatKind::Range(PatRange { lo, hi, end }) => {
2266 if let Some(int_range) = IntRange::from_range(
2268 lo.eval_bits(cx.tcx, cx.param_env, lo.ty),
2269 hi.eval_bits(cx.tcx, cx.param_env, hi.ty),
2276 FloatRange(lo, hi, end)
2279 PatKind::Array { prefix, slice, suffix } | PatKind::Slice { prefix, slice, suffix } => {
2280 let array_len = match pat.ty.kind() {
2281 ty::Array(_, length) => Some(length.eval_usize(cx.tcx, cx.param_env)),
2282 ty::Slice(_) => None,
2283 _ => span_bug!(pat.span, "bad ty {:?} for slice pattern", pat.ty),
2285 let prefix = prefix.len() as u64;
2286 let suffix = suffix.len() as u64;
2288 if slice.is_some() { VarLen(prefix, suffix) } else { FixedLen(prefix + suffix) };
2289 Slice(Slice::new(array_len, kind))
2291 PatKind::Or { .. } => bug!("Or-pattern should have been expanded earlier on."),
2295 /// The arm of a match expression.
2296 #[derive(Clone, Copy)]
2297 crate struct MatchArm<'p, 'tcx> {
2298 /// The pattern must have been lowered through `MatchVisitor::lower_pattern`.
2299 crate pat: &'p super::Pat<'tcx>,
2300 crate hir_id: HirId,
2301 crate has_guard: bool,
2304 /// The output of checking a match for exhaustiveness and arm reachability.
2305 crate struct UsefulnessReport<'p, 'tcx> {
2306 /// For each arm of the input, whether that arm is reachable after the arms above it.
2307 crate arm_usefulness: Vec<(MatchArm<'p, 'tcx>, Usefulness<'tcx>)>,
2308 /// If the match is exhaustive, this is empty. If not, this contains witnesses for the lack of
2310 crate non_exhaustiveness_witnesses: Vec<super::Pat<'tcx>>,
2313 /// The entrypoint for the usefulness algorithm. Computes whether a match is exhaustive and which
2314 /// of its arms are reachable.
2316 /// Note: the input patterns must have been lowered through `MatchVisitor::lower_pattern`.
2317 crate fn compute_match_usefulness<'p, 'tcx>(
2318 cx: &MatchCheckCtxt<'p, 'tcx>,
2319 arms: &[MatchArm<'p, 'tcx>],
2320 scrut_hir_id: HirId,
2322 ) -> UsefulnessReport<'p, 'tcx> {
2323 let mut matrix = Matrix::empty();
2324 let arm_usefulness: Vec<_> = arms
2328 let v = PatStack::from_pattern(arm.pat);
2330 is_useful(cx, &matrix, &v, LeaveOutWitness, arm.hir_id, arm.has_guard, true);
2338 let wild_pattern = cx.pattern_arena.alloc(super::Pat::wildcard_from_ty(scrut_ty));
2339 let v = PatStack::from_pattern(wild_pattern);
2340 let usefulness = is_useful(cx, &matrix, &v, ConstructWitness, scrut_hir_id, false, true);
2341 let non_exhaustiveness_witnesses = match usefulness {
2342 NotUseful => vec![], // Wildcard pattern isn't useful, so the match is exhaustive.
2343 UsefulWithWitness(pats) => {
2344 if pats.is_empty() {
2345 bug!("Exhaustiveness check returned no witnesses")
2347 pats.into_iter().map(|w| w.single_pattern()).collect()
2350 Useful(_) => bug!(),
2352 UsefulnessReport { arm_usefulness, non_exhaustiveness_witnesses }