diff options
| -rw-r--r-- | compiler/rustc_mir_build/src/thir/pattern/check_match.rs | 4 | ||||
| -rw-r--r-- | compiler/rustc_mir_build/src/thir/pattern/deconstruct_pat.rs (renamed from compiler/rustc_mir_build/src/thir/pattern/_match.rs) | 2298 | ||||
| -rw-r--r-- | compiler/rustc_mir_build/src/thir/pattern/mod.rs | 3 | ||||
| -rw-r--r-- | compiler/rustc_mir_build/src/thir/pattern/usefulness.rs | 992 |
4 files changed, 1661 insertions, 1636 deletions
diff --git a/compiler/rustc_mir_build/src/thir/pattern/check_match.rs b/compiler/rustc_mir_build/src/thir/pattern/check_match.rs index f9fe261bcee..97edbd83b89 100644 --- a/compiler/rustc_mir_build/src/thir/pattern/check_match.rs +++ b/compiler/rustc_mir_build/src/thir/pattern/check_match.rs @@ -1,5 +1,5 @@ -use super::_match::Usefulness::*; -use super::_match::{ +use super::usefulness::Usefulness::*; +use super::usefulness::{ compute_match_usefulness, expand_pattern, MatchArm, MatchCheckCtxt, UsefulnessReport, }; use super::{PatCtxt, PatKind, PatternError}; diff --git a/compiler/rustc_mir_build/src/thir/pattern/_match.rs b/compiler/rustc_mir_build/src/thir/pattern/deconstruct_pat.rs index f299663f679..62b4468eeb3 100644 --- a/compiler/rustc_mir_build/src/thir/pattern/_match.rs +++ b/compiler/rustc_mir_build/src/thir/pattern/deconstruct_pat.rs @@ -1,322 +1,16 @@ -//! Note: tests specific to this file can be found in: -//! -//! - `ui/pattern/usefulness` -//! - `ui/or-patterns` -//! - `ui/consts/const_in_pattern` -//! - `ui/rfc-2008-non-exhaustive` -//! - `ui/half-open-range-patterns` -//! - probably many others -//! -//! I (Nadrieril) prefer to put new tests in `ui/pattern/usefulness` unless there's a specific -//! reason not to, for example if they depend on a particular feature like `or_patterns`. -//! -//! ----- -//! -//! This file includes the logic for exhaustiveness and usefulness checking for -//! pattern-matching. Specifically, given a list of patterns for a type, we can -//! tell whether: -//! (a) the patterns cover every possible constructor for the type (exhaustiveness) -//! (b) each pattern is necessary (usefulness) -//! -//! The algorithm implemented here is a modified version of the one described in -//! [this paper](http://moscova.inria.fr/~maranget/papers/warn/index.html). -//! However, to save future implementors from reading the original paper, we -//! summarise the algorithm here to hopefully save time and be a little clearer -//! (without being so rigorous). -//! -//! # Premise -//! -//! The core of the algorithm revolves about a "usefulness" check. In particular, we -//! are trying to compute a predicate `U(P, p)` where `P` is a list of patterns (we refer to this as -//! a matrix). `U(P, p)` represents whether, given an existing list of patterns -//! `P_1 ..= P_m`, adding a new pattern `p` will be "useful" (that is, cover previously- -//! uncovered values of the type). -//! -//! If we have this predicate, then we can easily compute both exhaustiveness of an -//! entire set of patterns and the individual usefulness of each one. -//! (a) the set of patterns is exhaustive iff `U(P, _)` is false (i.e., adding a wildcard -//! match doesn't increase the number of values we're matching) -//! (b) a pattern `P_i` is not useful if `U(P[0..=(i-1), P_i)` is false (i.e., adding a -//! pattern to those that have come before it doesn't increase the number of values -//! we're matching). -//! -//! # Core concept -//! -//! The idea that powers everything that is done in this file is the following: a value is made -//! from a constructor applied to some fields. Examples of constructors are `Some`, `None`, `(,)` -//! (the 2-tuple constructor), `Foo {..}` (the constructor for a struct `Foo`), and `2` (the -//! constructor for the number `2`). Fields are just a (possibly empty) list of values. -//! -//! Some of the constructors listed above might feel weird: `None` and `2` don't take any -//! arguments. This is part of what makes constructors so general: we will consider plain values -//! like numbers and string literals to be constructors that take no arguments, also called "0-ary -//! constructors"; they are the simplest case of constructors. This allows us to see any value as -//! made up from a tree of constructors, each having a given number of children. For example: -//! `(None, Ok(0))` is made from 4 different constructors. -//! -//! This idea can be extended to patterns: a pattern captures a set of possible values, and we can -//! describe this set using constructors. For example, `Err(_)` captures all values of the type -//! `Result<T, E>` that start with the `Err` constructor (for some choice of `T` and `E`). The -//! wildcard `_` captures all values of the given type starting with any of the constructors for -//! that type. -//! -//! We use this to compute whether different patterns might capture a same value. Do the patterns -//! `Ok("foo")` and `Err(_)` capture a common value? The answer is no, because the first pattern -//! captures only values starting with the `Ok` constructor and the second only values starting -//! with the `Err` constructor. Do the patterns `Some(42)` and `Some(1..10)` intersect? They might, -//! since they both capture values starting with `Some`. To be certain, we need to dig under the -//! `Some` constructor and continue asking the question. This is the main idea behind the -//! exhaustiveness algorithm: by looking at patterns constructor-by-constructor, we can efficiently -//! figure out if some new pattern might capture a value that hadn't been captured by previous -//! patterns. -//! -//! Constructors are represented by the `Constructor` enum, and its fields by the `Fields` enum. -//! Most of the complexity of this file resides in transforming between patterns and -//! (`Constructor`, `Fields`) pairs, handling all the special cases correctly. -//! -//! Caveat: this constructors/fields distinction doesn't quite cover every Rust value. For example -//! a value of type `Rc<u64>` doesn't fit this idea very well, nor do various other things. -//! However, this idea covers most of the cases that are relevant to exhaustiveness checking. -//! -//! -//! # Algorithm -//! -//! Recall that `U(P, p)` represents whether, given an existing list of patterns (aka matrix) `P`, -//! adding a new pattern `p` will cover previously-uncovered values of the type. -//! During the course of the algorithm, the rows of the matrix won't just be individual patterns, -//! but rather partially-deconstructed patterns in the form of a list of fields. The paper -//! calls those pattern-vectors, and we will call them pattern-stacks. The same holds for the -//! new pattern `p`. -//! -//! For example, say we have the following: -//! -//! ``` -//! // x: (Option<bool>, Result<()>) -//! match x { -//! (Some(true), _) => {} -//! (None, Err(())) => {} -//! (None, Err(_)) => {} -//! } -//! ``` -//! -//! Here, the matrix `P` starts as: -//! -//! ``` -//! [ -//! [(Some(true), _)], -//! [(None, Err(()))], -//! [(None, Err(_))], -//! ] -//! ``` -//! -//! We can tell it's not exhaustive, because `U(P, _)` is true (we're not covering -//! `[(Some(false), _)]`, for instance). In addition, row 3 is not useful, because -//! all the values it covers are already covered by row 2. -//! -//! A list of patterns can be thought of as a stack, because we are mainly interested in the top of -//! the stack at any given point, and we can pop or apply constructors to get new pattern-stacks. -//! To match the paper, the top of the stack is at the beginning / on the left. -//! -//! There are two important operations on pattern-stacks necessary to understand the algorithm: -//! -//! 1. We can pop a given constructor off the top of a stack. This operation is called -//! `specialize`, and is denoted `S(c, p)` where `c` is a constructor (like `Some` or -//! `None`) and `p` a pattern-stack. -//! If the pattern on top of the stack can cover `c`, this removes the constructor and -//! pushes its arguments onto the stack. It also expands OR-patterns into distinct patterns. -//! Otherwise the pattern-stack is discarded. -//! This essentially filters those pattern-stacks whose top covers the constructor `c` and -//! discards the others. -//! -//! For example, the first pattern above initially gives a stack `[(Some(true), _)]`. If we -//! pop the tuple constructor, we are left with `[Some(true), _]`, and if we then pop the -//! `Some` constructor we get `[true, _]`. If we had popped `None` instead, we would get -//! nothing back. -//! -//! This returns zero or more new pattern-stacks, as follows. We look at the pattern `p_1` -//! on top of the stack, and we have four cases: -//! -//! 1.1. `p_1 = c(r_1, .., r_a)`, i.e. the top of the stack has constructor `c`. We -//! push onto the stack the arguments of this constructor, and return the result: -//! `r_1, .., r_a, p_2, .., p_n` -//! -//! 1.2. `p_1 = c'(r_1, .., r_a')` where `c ≠ c'`. We discard the current stack and -//! return nothing. -//! -//! 1.3. `p_1 = _`. We push onto the stack as many wildcards as the constructor `c` has -//! arguments (its arity), and return the resulting stack: -//! `_, .., _, p_2, .., p_n` -//! -//! 1.4. `p_1 = r_1 | r_2`. We expand the OR-pattern and then recurse on each resulting -//! stack: -//! - `S(c, (r_1, p_2, .., p_n))` -//! - `S(c, (r_2, p_2, .., p_n))` -//! -//! 2. We can pop a wildcard off the top of the stack. This is called `S(_, p)`, where `p` is -//! a pattern-stack. Note: the paper calls this `D(p)`. -//! This is used when we know there are missing constructor cases, but there might be -//! existing wildcard patterns, so to check the usefulness of the matrix, we have to check -//! all its *other* components. -//! -//! It is computed as follows. We look at the pattern `p_1` on top of the stack, -//! and we have three cases: -//! 2.1. `p_1 = c(r_1, .., r_a)`. We discard the current stack and return nothing. -//! 2.2. `p_1 = _`. We return the rest of the stack: -//! p_2, .., p_n -//! 2.3. `p_1 = r_1 | r_2`. We expand the OR-pattern and then recurse on each resulting -//! stack. -//! - `S(_, (r_1, p_2, .., p_n))` -//! - `S(_, (r_2, p_2, .., p_n))` -//! -//! Note that the OR-patterns are not always used directly in Rust, but are used to derive the -//! exhaustive integer matching rules, so they're written here for posterity. -//! -//! Both those operations extend straightforwardly to a list or pattern-stacks, i.e. a matrix, by -//! working row-by-row. Popping a constructor ends up keeping only the matrix rows that start with -//! the given constructor, and popping a wildcard keeps those rows that start with a wildcard. -//! -//! -//! The algorithm for computing `U` -//! ------------------------------- -//! The algorithm is inductive (on the number of columns: i.e., components of tuple patterns). -//! That means we're going to check the components from left-to-right, so the algorithm -//! operates principally on the first component of the matrix and new pattern-stack `p`. -//! This algorithm is realised in the `is_useful` function. -//! -//! Base case. (`n = 0`, i.e., an empty tuple pattern) -//! - If `P` already contains an empty pattern (i.e., if the number of patterns `m > 0`), -//! then `U(P, p)` is false. -//! - Otherwise, `P` must be empty, so `U(P, p)` is true. -//! -//! Inductive step. (`n > 0`, i.e., whether there's at least one column -//! [which may then be expanded into further columns later]) -//! We're going to match on the top of the new pattern-stack, `p_1`. -//! - If `p_1 == c(r_1, .., r_a)`, i.e. we have a constructor pattern. -//! Then, the usefulness of `p_1` can be reduced to whether it is useful when -//! we ignore all the patterns in the first column of `P` that involve other constructors. -//! This is where `S(c, P)` comes in: -//! `U(P, p) := U(S(c, P), S(c, p))` -//! -//! For example, if `P` is: -//! -//! ``` -//! [ -//! [Some(true), _], -//! [None, 0], -//! ] -//! ``` -//! -//! and `p` is `[Some(false), 0]`, then we don't care about row 2 since we know `p` only -//! matches values that row 2 doesn't. For row 1 however, we need to dig into the -//! arguments of `Some` to know whether some new value is covered. So we compute -//! `U([[true, _]], [false, 0])`. -//! -//! - If `p_1 == _`, then we look at the list of constructors that appear in the first -//! component of the rows of `P`: -//! + If there are some constructors that aren't present, then we might think that the -//! wildcard `_` is useful, since it covers those constructors that weren't covered -//! before. -//! That's almost correct, but only works if there were no wildcards in those first -//! components. So we need to check that `p` is useful with respect to the rows that -//! start with a wildcard, if there are any. This is where `S(_, x)` comes in: -//! `U(P, p) := U(S(_, P), S(_, p))` -//! -//! For example, if `P` is: -//! -//! ``` -//! [ -//! [_, true, _], -//! [None, false, 1], -//! ] -//! ``` -//! -//! and `p` is `[_, false, _]`, the `Some` constructor doesn't appear in `P`. So if we -//! only had row 2, we'd know that `p` is useful. However row 1 starts with a -//! wildcard, so we need to check whether `U([[true, _]], [false, 1])`. -//! -//! + Otherwise, all possible constructors (for the relevant type) are present. In this -//! case we must check whether the wildcard pattern covers any unmatched value. For -//! that, we can think of the `_` pattern as a big OR-pattern that covers all -//! possible constructors. For `Option`, that would mean `_ = None | Some(_)` for -//! example. The wildcard pattern is useful in this case if it is useful when -//! specialized to one of the possible constructors. So we compute: -//! `U(P, p) := ∃(k ϵ constructors) U(S(k, P), S(k, p))` -//! -//! For example, if `P` is: -//! -//! ``` -//! [ -//! [Some(true), _], -//! [None, false], -//! ] -//! ``` -//! -//! and `p` is `[_, false]`, both `None` and `Some` constructors appear in the first -//! components of `P`. We will therefore try popping both constructors in turn: we -//! compute `U([[true, _]], [_, false])` for the `Some` constructor, and `U([[false]], -//! [false])` for the `None` constructor. The first case returns true, so we know that -//! `p` is useful for `P`. Indeed, it matches `[Some(false), _]` that wasn't matched -//! before. -//! -//! - If `p_1 == r_1 | r_2`, then the usefulness depends on each `r_i` separately: -//! `U(P, p) := U(P, (r_1, p_2, .., p_n)) -//! || U(P, (r_2, p_2, .., p_n))` -//! -//! Modifications to the algorithm -//! ------------------------------ -//! The algorithm in the paper doesn't cover some of the special cases that arise in Rust, for -//! example uninhabited types and variable-length slice patterns. These are drawn attention to -//! throughout the code below. I'll make a quick note here about how exhaustive integer matching is -//! accounted for, though. -//! -//! Exhaustive integer matching -//! --------------------------- -//! An integer type can be thought of as a (huge) sum type: 1 | 2 | 3 | ... -//! So to support exhaustive integer matching, we can make use of the logic in the paper for -//! OR-patterns. However, we obviously can't just treat ranges x..=y as individual sums, because -//! they are likely gigantic. So we instead treat ranges as constructors of the integers. This means -//! that we have a constructor *of* constructors (the integers themselves). We then need to work -//! through all the inductive step rules above, deriving how the ranges would be treated as -//! OR-patterns, and making sure that they're treated in the same way even when they're ranges. -//! There are really only four special cases here: -//! - When we match on a constructor that's actually a range, we have to treat it as if we would -//! an OR-pattern. -//! + It turns out that we can simply extend the case for single-value patterns in -//! `specialize` to either be *equal* to a value constructor, or *contained within* a range -//! constructor. -//! + When the pattern itself is a range, you just want to tell whether any of the values in -//! the pattern range coincide with values in the constructor range, which is precisely -//! intersection. -//! Since when encountering a range pattern for a value constructor, we also use inclusion, it -//! means that whenever the constructor is a value/range and the pattern is also a value/range, -//! we can simply use intersection to test usefulness. -//! - When we're testing for usefulness of a pattern and the pattern's first component is a -//! wildcard. -//! + If all the constructors appear in the matrix, we have a slight complication. By default, -//! the behaviour (i.e., a disjunction over specialised matrices for each constructor) is -//! invalid, because we want a disjunction over every *integer* in each range, not just a -//! disjunction over every range. This is a bit more tricky to deal with: essentially we need -//! to form equivalence classes of subranges of the constructor range for which the behaviour -//! of the matrix `P` and new pattern `p` are the same. This is described in more -//! detail in `Constructor::split`. -//! + If some constructors are missing from the matrix, it turns out we don't need to do -//! anything special (because we know none of the integers are actually wildcards: i.e., we -//! can't span wildcards using ranges). - +//! This module provides functions to deconstruct and reconstruct patterns into a constructor +//! applied to some fields. This is used by the `_match` module to compute pattern +//! usefulness/exhaustiveness. use self::Constructor::*; use self::SliceKind::*; -use self::Usefulness::*; -use self::WitnessPreference::*; + +use super::compare_const_vals; +use super::usefulness::{MatchCheckCtxt, PatCtxt}; +use super::{FieldPat, Pat, PatKind, PatRange}; use rustc_data_structures::captures::Captures; -use rustc_data_structures::fx::FxHashSet; -use rustc_data_structures::sync::OnceCell; use rustc_index::vec::Idx; -use super::{compare_const_vals, PatternFoldable, PatternFolder}; -use super::{FieldPat, Pat, PatKind, PatRange}; - -use rustc_arena::TypedArena; use rustc_attr::{SignedInt, UnsignedInt}; use rustc_hir::def_id::DefId; use rustc_hir::{HirId, RangeEnd}; @@ -330,281 +24,344 @@ use rustc_target::abi::{Integer, Size, VariantIdx}; use smallvec::{smallvec, SmallVec}; use std::cmp::{self, max, min, Ordering}; -use std::fmt; -use std::iter::{FromIterator, IntoIterator}; +use std::iter::IntoIterator; use std::ops::RangeInclusive; -crate fn expand_pattern<'tcx>(pat: Pat<'tcx>) -> Pat<'tcx> { - LiteralExpander.fold_pattern(&pat) -} - -struct LiteralExpander; - -impl<'tcx> PatternFolder<'tcx> for LiteralExpander { - fn fold_pattern(&mut self, pat: &Pat<'tcx>) -> Pat<'tcx> { - debug!("fold_pattern {:?} {:?} {:?}", pat, pat.ty.kind(), pat.kind); - match (pat.ty.kind(), pat.kind.as_ref()) { - (_, PatKind::Binding { subpattern: Some(s), .. }) => s.fold_with(self), - (_, PatKind::AscribeUserType { subpattern: s, .. }) => s.fold_with(self), - (ty::Ref(_, t, _), PatKind::Constant { .. }) if t.is_str() => { - // Treat string literal patterns as deref patterns to a `str` constant, i.e. - // `&CONST`. This expands them like other const patterns. This could have been done - // in `const_to_pat`, but that causes issues with the rest of the matching code. - let mut new_pat = pat.super_fold_with(self); - // Make a fake const pattern of type `str` (instead of `&str`). That the carried - // constant value still knows it is of type `&str`. - new_pat.ty = t; - Pat { - kind: Box::new(PatKind::Deref { subpattern: new_pat }), - span: pat.span, - ty: pat.ty, - } - } - _ => pat.super_fold_with(self), - } - } -} - -impl<'tcx> Pat<'tcx> { - pub(super) fn is_wildcard(&self) -> bool { - matches!(*self.kind, PatKind::Binding { subpattern: None, .. } | PatKind::Wild) - } -} - -/// A row of a matrix. Rows of len 1 are very common, which is why `SmallVec[_; 2]` -/// works well. -#[derive(Debug, Clone)] -struct PatStack<'p, 'tcx> { - pats: SmallVec<[&'p Pat<'tcx>; 2]>, - /// Cache for the constructor of the head - head_ctor: OnceCell<Constructor<'tcx>>, +/// An inclusive interval, used for precise integer exhaustiveness checking. +/// `IntRange`s always store a contiguous range. This means that values are +/// encoded such that `0` encodes the minimum value for the integer, +/// regardless of the signedness. +/// For example, the pattern `-128..=127i8` is encoded as `0..=255`. +/// This makes comparisons and arithmetic on interval endpoints much more +/// straightforward. See `signed_bias` for details. +/// +/// `IntRange` is never used to encode an empty range or a "range" that wraps +/// around the (offset) space: i.e., `range.lo <= range.hi`. +#[derive(Clone, Debug)] +pub(super) struct IntRange<'tcx> { + range: RangeInclusive<u128>, + ty: Ty<'tcx>, + span: Span, } -impl<'p, 'tcx> PatStack<'p, 'tcx> { - fn from_pattern(pat: &'p Pat<'tcx>) -> Self { - Self::from_vec(smallvec![pat]) - } - - fn from_vec(vec: SmallVec<[&'p Pat<'tcx>; 2]>) -> Self { - PatStack { pats: vec, head_ctor: OnceCell::new() } +impl<'tcx> IntRange<'tcx> { + #[inline] + fn is_integral(ty: Ty<'_>) -> bool { + matches!(ty.kind(), ty::Char | ty::Int(_) | ty::Uint(_) | ty::Bool) } - fn is_empty(&self) -> bool { - self.pats.is_empty() + fn is_singleton(&self) -> bool { + self.range.start() == self.range.end() } - fn len(&self) -> usize { - self.pats.len() + fn boundaries(&self) -> (u128, u128) { + (*self.range.start(), *self.range.end()) } - fn head(&self) -> &'p Pat<'tcx> { - self.pats[0] + /// Don't treat `usize`/`isize` exhaustively unless the `precise_pointer_size_matching` feature + /// is enabled. + fn treat_exhaustively(&self, tcx: TyCtxt<'tcx>) -> bool { + !self.ty.is_ptr_sized_integral() || tcx.features().precise_pointer_size_matching } - fn head_ctor<'a>(&'a self, cx: &MatchCheckCtxt<'p, 'tcx>) -> &'a Constructor<'tcx> { - self.head_ctor.get_or_init(|| pat_constructor(cx, self.head())) + #[inline] + fn integral_size_and_signed_bias(tcx: TyCtxt<'tcx>, ty: Ty<'_>) -> Option<(Size, u128)> { + match *ty.kind() { + ty::Bool => Some((Size::from_bytes(1), 0)), + ty::Char => Some((Size::from_bytes(4), 0)), + ty::Int(ity) => { + let size = Integer::from_attr(&tcx, SignedInt(ity)).size(); + Some((size, 1u128 << (size.bits() as u128 - 1))) + } + ty::Uint(uty) => Some((Integer::from_attr(&tcx, UnsignedInt(uty)).size(), 0)), + _ => None, + } } - fn iter(&self) -> impl Iterator<Item = &Pat<'tcx>> { - self.pats.iter().copied() + #[inline] + fn from_const( + tcx: TyCtxt<'tcx>, + param_env: ty::ParamEnv<'tcx>, + value: &Const<'tcx>, + span: Span, + ) -> Option<IntRange<'tcx>> { + if let Some((target_size, bias)) = Self::integral_size_and_signed_bias(tcx, value.ty) { + let ty = value.ty; + let val = (|| { + if let ty::ConstKind::Value(ConstValue::Scalar(scalar)) = value.val { + // For this specific pattern we can skip a lot of effort and go + // straight to the result, after doing a bit of checking. (We + // could remove this branch and just fall through, which + // is more general but much slower.) + if let Ok(bits) = scalar.to_bits_or_ptr(target_size, &tcx) { + return Some(bits); + } + } + // This is a more general form of the previous case. + value.try_eval_bits(tcx, param_env, ty) + })()?; + let val = val ^ bias; + Some(IntRange { range: val..=val, ty, span }) + } else { + None + } } - // If the first pattern is an or-pattern, expand this pattern. Otherwise, return `None`. - fn expand_or_pat(&self) -> Option<Vec<Self>> { - if self.is_empty() { - None - } else if let PatKind::Or { pats } = &*self.head().kind { - Some( - pats.iter() - .map(|pat| { - let mut new_patstack = PatStack::from_pattern(pat); - new_patstack.pats.extend_from_slice(&self.pats[1..]); - new_patstack - }) - .collect(), - ) + #[inline] + fn from_range( + tcx: TyCtxt<'tcx>, + lo: u128, + hi: u128, + ty: Ty<'tcx>, + end: &RangeEnd, + span: Span, + ) -> Option<IntRange<'tcx>> { + if Self::is_integral(ty) { + // Perform a shift if the underlying types are signed, + // which makes the interval arithmetic simpler. + let bias = IntRange::signed_bias(tcx, ty); + let (lo, hi) = (lo ^ bias, hi ^ bias); + let offset = (*end == RangeEnd::Excluded) as u128; + if lo > hi || (lo == hi && *end == RangeEnd::Excluded) { + // This should have been caught earlier by E0030. + bug!("malformed range pattern: {}..={}", lo, (hi - offset)); + } + Some(IntRange { range: lo..=(hi - offset), ty, span }) } else { None } } - /// This computes `S(self.head_ctor(), self)`. See top of the file for explanations. - /// - /// Structure patterns with a partial wild pattern (Foo { a: 42, .. }) have their missing - /// fields filled with wild patterns. - /// - /// This is roughly the inverse of `Constructor::apply`. - fn pop_head_constructor(&self, ctor_wild_subpatterns: &Fields<'p, 'tcx>) -> PatStack<'p, 'tcx> { - // We pop the head pattern and push the new fields extracted from the arguments of - // `self.head()`. - let new_fields = ctor_wild_subpatterns.replace_with_pattern_arguments(self.head()); - new_fields.push_on_patstack(&self.pats[1..]) + // The return value of `signed_bias` should be XORed with an endpoint to encode/decode it. + fn signed_bias(tcx: TyCtxt<'tcx>, ty: Ty<'tcx>) -> u128 { + match *ty.kind() { + ty::Int(ity) => { + let bits = Integer::from_attr(&tcx, SignedInt(ity)).size().bits() as u128; + 1u128 << (bits - 1) + } + _ => 0, + } } -} -impl<'p, 'tcx> Default for PatStack<'p, 'tcx> { - fn default() -> Self { - Self::from_vec(smallvec![]) + fn is_subrange(&self, other: &Self) -> bool { + other.range.start() <= self.range.start() && self.range.end() <= other.range.end() } -} -impl<'p, 'tcx> PartialEq for PatStack<'p, 'tcx> { - fn eq(&self, other: &Self) -> bool { - self.pats == other.pats + fn intersection(&self, tcx: TyCtxt<'tcx>, other: &Self) -> Option<Self> { + let ty = self.ty; + let (lo, hi) = self.boundaries(); + let (other_lo, other_hi) = other.boundaries(); + if self.treat_exhaustively(tcx) { + if lo <= other_hi && other_lo <= hi { + let span = other.span; + Some(IntRange { range: max(lo, other_lo)..=min(hi, other_hi), ty, span }) + } else { + None + } + } else { + // If the range should not be treated exhaustively, fallback to checking for inclusion. + if self.is_subrange(other) { Some(self.clone()) } else { None } + } } -} -impl<'p, 'tcx> FromIterator<&'p Pat<'tcx>> for PatStack<'p, 'tcx> { - fn from_iter<T>(iter: T) -> Self - where - T: IntoIterator<Item = &'p Pat<'tcx>>, - { - Self::from_vec(iter.into_iter().collect()) + fn suspicious_intersection(&self, other: &Self) -> bool { + // `false` in the following cases: + // 1 ---- // 1 ---------- // 1 ---- // 1 ---- + // 2 ---------- // 2 ---- // 2 ---- // 2 ---- + // + // The following are currently `false`, but could be `true` in the future (#64007): + // 1 --------- // 1 --------- + // 2 ---------- // 2 ---------- + // + // `true` in the following cases: + // 1 ------- // 1 ------- + // 2 -------- // 2 ------- + let (lo, hi) = self.boundaries(); + let (other_lo, other_hi) = other.boundaries(); + lo == other_hi || hi == other_lo } -} -/// A 2D matrix. -#[derive(Clone, PartialEq)] -struct Matrix<'p, 'tcx> { - patterns: Vec<PatStack<'p, 'tcx>>, -} + fn to_pat(&self, tcx: TyCtxt<'tcx>) -> Pat<'tcx> { + let (lo, hi) = self.boundaries(); -impl<'p, 'tcx> Matrix<'p, 'tcx> { - fn empty() -> Self { - Matrix { patterns: vec![] } - } + let bias = IntRange::signed_bias(tcx, self.ty); + let (lo, hi) = (lo ^ bias, hi ^ bias); - /// Pushes a new row to the matrix. If the row starts with an or-pattern, this expands it. - fn push(&mut self, row: PatStack<'p, 'tcx>) { - if let Some(rows) = row.expand_or_pat() { - for row in rows { - // We recursively expand the or-patterns of the new rows. - // This is necessary as we might have `0 | (1 | 2)` or e.g., `x @ 0 | x @ (1 | 2)`. - self.push(row) - } - } else { - self.patterns.push(row); - } - } + let ty = ty::ParamEnv::empty().and(self.ty); + let lo_const = ty::Const::from_bits(tcx, lo, ty); + let hi_const = ty::Const::from_bits(tcx, hi, ty); - /// Iterate over the first component of each row - fn heads<'a>(&'a self) -> impl Iterator<Item = &'a Pat<'tcx>> + Captures<'p> { - self.patterns.iter().map(|r| r.head()) - } + let kind = if lo == hi { + PatKind::Constant { value: lo_const } + } else { + PatKind::Range(PatRange { lo: lo_const, hi: hi_const, end: RangeEnd::Included }) + }; - /// Iterate over the first constructor of each row - fn head_ctors<'a>( - &'a self, - cx: &'a MatchCheckCtxt<'p, 'tcx>, - ) -> impl Iterator<Item = &'a Constructor<'tcx>> + Captures<'a> + Captures<'p> { - self.patterns.iter().map(move |r| r.head_ctor(cx)) + // This is a brand new pattern, so we don't reuse `self.span`. + Pat { ty: self.ty, span: DUMMY_SP, kind: Box::new(kind) } } - /// This computes `S(constructor, self)`. See top of the file for explanations. - fn specialize_constructor( + /// For exhaustive integer matching, some constructors are grouped within other constructors + /// (namely integer typed values are grouped within ranges). However, when specialising these + /// constructors, we want to be specialising for the underlying constructors (the integers), not + /// the groups (the ranges). Thus we need to split the groups up. Splitting them up naïvely would + /// mean creating a separate constructor for every single value in the range, which is clearly + /// impractical. However, observe that for some ranges of integers, the specialisation will be + /// identical across all values in that range (i.e., there are equivalence classes of ranges of + /// constructors based on their `U(S(c, P), S(c, p))` outcome). These classes are grouped by + /// the patterns that apply to them (in the matrix `P`). We can split the range whenever the + /// patterns that apply to that range (specifically: the patterns that *intersect* with that range) + /// change. + /// Our solution, therefore, is to split the range constructor into subranges at every single point + /// the group of intersecting patterns changes (using the method described below). + /// And voilà! We're testing precisely those ranges that we need to, without any exhaustive matching + /// on actual integers. The nice thing about this is that the number of subranges is linear in the + /// number of rows in the matrix (i.e., the number of cases in the `match` statement), so we don't + /// need to be worried about matching over gargantuan ranges. + /// + /// Essentially, given the first column of a matrix representing ranges, looking like the following: + /// + /// |------| |----------| |-------| || + /// |-------| |-------| |----| || + /// |---------| + /// + /// We split the ranges up into equivalence classes so the ranges are no longer overlapping: + /// + /// |--|--|||-||||--||---|||-------| |-|||| || + /// + /// The logic for determining how to split the ranges is fairly straightforward: we calculate + /// boundaries for each interval range, sort them, then create constructors for each new interval + /// between every pair of boundary points. (This essentially sums up to performing the intuitive + /// merging operation depicted above.) + fn split<'p>( &self, pcx: PatCtxt<'_, 'p, 'tcx>, - ctor: &Constructor<'tcx>, - ctor_wild_subpatterns: &Fields<'p, 'tcx>, - ) -> Matrix<'p, 'tcx> { - self.patterns - .iter() - .filter(|r| ctor.is_covered_by(pcx, r.head_ctor(pcx.cx))) - .map(|r| r.pop_head_constructor(ctor_wild_subpatterns)) - .collect() - } -} + hir_id: Option<HirId>, + ) -> SmallVec<[Constructor<'tcx>; 1]> { + let ty = pcx.ty; -/// Pretty-printer for matrices of patterns, example: -/// -/// ```text -/// +++++++++++++++++++++++++++++ -/// + _ + [] + -/// +++++++++++++++++++++++++++++ -/// + true + [First] + -/// +++++++++++++++++++++++++++++ -/// + true + [Second(true)] + -/// +++++++++++++++++++++++++++++ -/// + false + [_] + -/// +++++++++++++++++++++++++++++ -/// + _ + [_, _, tail @ ..] + -/// +++++++++++++++++++++++++++++ -/// ``` -impl<'p, 'tcx> fmt::Debug for Matrix<'p, 'tcx> { - fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { - write!(f, "\n")?; - - let Matrix { patterns: m, .. } = self; - let pretty_printed_matrix: Vec<Vec<String>> = - m.iter().map(|row| row.iter().map(|pat| format!("{:?}", pat)).collect()).collect(); - - let column_count = m.iter().map(|row| row.len()).max().unwrap_or(0); - assert!(m.iter().all(|row| row.len() == column_count)); - let column_widths: Vec<usize> = (0..column_count) - .map(|col| pretty_printed_matrix.iter().map(|row| row[col].len()).max().unwrap_or(0)) - .collect(); - - let total_width = column_widths.iter().cloned().sum::<usize>() + column_count * 3 + 1; - let br = "+".repeat(total_width); - write!(f, "{}\n", br)?; - for row in pretty_printed_matrix { - write!(f, "+")?; - for (column, pat_str) in row.into_iter().enumerate() { - write!(f, " ")?; - write!(f, "{:1$}", pat_str, column_widths[column])?; - write!(f, " +")?; - } - write!(f, "\n")?; - write!(f, "{}\n", br)?; + /// Represents a border between 2 integers. Because the intervals spanning borders + /// must be able to cover every integer, we need to be able to represent + /// 2^128 + 1 such borders. + #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Debug)] + enum Border { + JustBefore(u128), + AfterMax, } - Ok(()) - } -} -impl<'p, 'tcx> FromIterator<PatStack<'p, 'tcx>> for Matrix<'p, 'tcx> { - fn from_iter<T>(iter: T) -> Self - where - T: IntoIterator<Item = PatStack<'p, 'tcx>>, - { - let mut matrix = Matrix::empty(); - for x in iter { - // Using `push` ensures we correctly expand or-patterns. - matrix.push(x); + // A function for extracting the borders of an integer interval. + fn range_borders(r: IntRange<'_>) -> impl Iterator<Item = Border> { + let (lo, hi) = r.range.into_inner(); + let from = Border::JustBefore(lo); + let to = match hi.checked_add(1) { + Some(m) => Border::JustBefore(m), + None => Border::AfterMax, + }; + vec![from, to].into_iter() } - matrix + + // Collect the span and range of all the intersecting ranges to lint on likely + // incorrect range patterns. (#63987) + let mut overlaps = vec![]; + let row_len = pcx.matrix.column_count().unwrap_or(0); + // `borders` is the set of borders between equivalence classes: each equivalence + // class lies between 2 borders. + let row_borders = pcx + .matrix + .head_ctors(pcx.cx) + .filter_map(|ctor| ctor.as_int_range()) + .filter_map(|range| { + let intersection = self.intersection(pcx.cx.tcx, &range); + let should_lint = self.suspicious_intersection(&range); + if let (Some(range), 1, true) = (&intersection, row_len, should_lint) { + // FIXME: for now, only check for overlapping ranges on simple range + // patterns. Otherwise with the current logic the following is detected + // as overlapping: + // match (10u8, true) { + // (0 ..= 125, false) => {} + // (126 ..= 255, false) => {} + // (0 ..= 255, true) => {} + // } + overlaps.push(range.clone()); + } + intersection + }) + .flat_map(range_borders); + let self_borders = range_borders(self.clone()); + let mut borders: Vec<_> = row_borders.chain(self_borders).collect(); + borders.sort_unstable(); + + self.lint_overlapping_patterns(pcx.cx.tcx, hir_id, ty, overlaps); + + // We're going to iterate through every adjacent pair of borders, making sure that + // each represents an interval of nonnegative length, and convert each such + // interval into a constructor. + borders + .array_windows() + .filter_map(|&pair| match pair { + [Border::JustBefore(n), Border::JustBefore(m)] => { + if n < m { + Some(n..=(m - 1)) + } else { + None + } + } + [Border::JustBefore(n), Border::AfterMax] => Some(n..=u128::MAX), + [Border::AfterMax, _] => None, + }) + .map(|range| IntRange { range, ty, span: pcx.span }) + .map(IntRange) + .collect() } -} -crate struct MatchCheckCtxt<'a, 'tcx> { - crate tcx: TyCtxt<'tcx>, - /// The module in which the match occurs. This is necessary for - /// checking inhabited-ness of types because whether a type is (visibly) - /// inhabited can depend on whether it was defined in the current module or - /// not. E.g., `struct Foo { _private: ! }` cannot be seen to be empty - /// outside it's module and should not be matchable with an empty match - /// statement. - crate module: DefId, - crate param_env: ty::ParamEnv<'tcx>, - crate pattern_arena: &'a TypedArena<Pat<'tcx>>, -} + fn lint_overlapping_patterns( + &self, + tcx: TyCtxt<'tcx>, + hir_id: Option<HirId>, + ty: Ty<'tcx>, + overlaps: Vec<IntRange<'tcx>>, + ) { + if let (true, Some(hir_id)) = (!overlaps.is_empty(), hir_id) { + tcx.struct_span_lint_hir( + lint::builtin::OVERLAPPING_PATTERNS, + hir_id, + self.span, + |lint| { + let mut err = lint.build("multiple patterns covering the same range"); + err.span_label(self.span, "overlapping patterns"); + for int_range in overlaps { + // Use the real type for user display of the ranges: + err.span_label( + int_range.span, + &format!( + "this range overlaps on `{}`", + IntRange { range: int_range.range, ty, span: DUMMY_SP }.to_pat(tcx), + ), + ); + } + err.emit(); + }, + ); + } + } -impl<'a, 'tcx> MatchCheckCtxt<'a, 'tcx> { - fn is_uninhabited(&self, ty: Ty<'tcx>) -> bool { - if self.tcx.features().exhaustive_patterns { - self.tcx.is_ty_uninhabited_from(self.module, ty, self.param_env) + /// See `Constructor::is_covered_by` + fn is_covered_by<'p>(&self, pcx: PatCtxt<'_, 'p, 'tcx>, other: &Self) -> bool { + if self.intersection(pcx.cx.tcx, other).is_some() { + // Constructor splitting should ensure that all intersections we encounter are actually + // inclusions. + assert!(self.is_subrange(other)); + true } else { false } } +} - /// Returns whether the given type is an enum from another crate declared `#[non_exhaustive]`. - fn is_foreign_non_exhaustive_enum(&self, ty: Ty<'tcx>) -> bool { - match ty.kind() { - ty::Adt(def, ..) => { - def.is_enum() && def.is_variant_list_non_exhaustive() && !def.did.is_local() - } - _ => false, - } +/// Ignore spans when comparing, they don't carry semantic information as they are only for lints. +impl<'tcx> std::cmp::PartialEq for IntRange<'tcx> { + fn eq(&self, other: &Self) -> bool { + self.range == other.range && self.ty == other.ty } } @@ -639,7 +396,7 @@ impl SliceKind { /// A constructor for array and slice patterns. #[derive(Copy, Clone, Debug, PartialEq, Eq)] -struct Slice { +pub(super) struct Slice { /// `None` if the matched value is a slice, `Some(n)` if it is an array of size `n`. array_len: Option<u64>, /// The kind of pattern it is: fixed-length `[x, y]` or variable length `[x, .., y]`. @@ -794,7 +551,7 @@ impl Slice { /// constructor. `Constructor::apply` reconstructs the pattern from a pair of `Constructor` and /// `Fields`. #[derive(Clone, Debug, PartialEq)] -enum Constructor<'tcx> { +pub(super) enum Constructor<'tcx> { /// The constructor for patterns that have a single constructor, like tuples, struct patterns /// and fixed-length arrays. Single, @@ -820,7 +577,7 @@ enum Constructor<'tcx> { } impl<'tcx> Constructor<'tcx> { - fn is_wildcard(&self) -> bool { + pub(super) fn is_wildcard(&self) -> bool { matches!(self, Wildcard) } @@ -849,6 +606,67 @@ impl<'tcx> Constructor<'tcx> { } } + /// Determines the constructor that the given pattern can be specialized to. + pub(super) fn from_pat<'p>(cx: &MatchCheckCtxt<'p, 'tcx>, pat: &'p Pat<'tcx>) -> Self { + match pat.kind.as_ref() { + PatKind::AscribeUserType { .. } => bug!(), // Handled by `expand_pattern` + PatKind::Binding { .. } | PatKind::Wild => Wildcard, + PatKind::Leaf { .. } | PatKind::Deref { .. } => Single, + &PatKind::Variant { adt_def, variant_index, .. } => { + Variant(adt_def.variants[variant_index].def_id) + } + PatKind::Constant { value } => { + if let Some(int_range) = IntRange::from_const(cx.tcx, cx.param_env, value, pat.span) + { + IntRange(int_range) + } else { + match pat.ty.kind() { + ty::Float(_) => FloatRange(value, value, RangeEnd::Included), + // In `expand_pattern`, we convert string literals to `&CONST` patterns with + // `CONST` a pattern of type `str`. In truth this contains a constant of type + // `&str`. + ty::Str => Str(value), + // All constants that can be structurally matched have already been expanded + // into the corresponding `Pat`s by `const_to_pat`. Constants that remain are + // opaque. + _ => Opaque, + } + } + } + &PatKind::Range(PatRange { lo, hi, end }) => { + let ty = lo.ty; + if let Some(int_range) = IntRange::from_range( + cx.tcx, + lo.eval_bits(cx.tcx, cx.param_env, lo.ty), + hi.eval_bits(cx.tcx, cx.param_env, hi.ty), + ty, + &end, + pat.span, + ) { + IntRange(int_range) + } else { + FloatRange(lo, hi, end) + } + } + PatKind::Array { prefix, slice, suffix } | PatKind::Slice { prefix, slice, suffix } => { + let array_len = match pat.ty.kind() { + ty::Array(_, length) => Some(length.eval_usize(cx.tcx, cx.param_env)), + ty::Slice(_) => None, + _ => span_bug!(pat.span, "bad ty {:?} for slice pattern", pat.ty), + }; + let prefix = prefix.len() as u64; + let suffix = suffix.len() as u64; + let kind = if slice.is_some() { + VarLen(prefix, suffix) + } else { + FixedLen(prefix + suffix) + }; + Slice(Slice::new(array_len, kind)) + } + PatKind::Or { .. } => bug!("Or-pattern should have been expanded earlier on."), + } + } + /// Some constructors (namely `Wildcard`, `IntRange` and `Slice`) actually stand for a set of actual /// constructors (like variants, integers or fixed-sized slices). When specializing for these /// constructors, we want to be specialising for the actual underlying constructors. @@ -865,7 +683,11 @@ impl<'tcx> Constructor<'tcx> { /// /// `hir_id` is `None` when we're evaluating the wildcard pattern. In that case we do not want /// to lint for overlapping ranges. - fn split<'p>(&self, pcx: PatCtxt<'_, 'p, 'tcx>, hir_id: Option<HirId>) -> SmallVec<[Self; 1]> { + pub(super) fn split<'p>( + &self, + pcx: PatCtxt<'_, 'p, 'tcx>, + hir_id: Option<HirId>, + ) -> SmallVec<[Self; 1]> { debug!("Constructor::split({:#?}, {:#?})", self, pcx.matrix); match self { @@ -907,7 +729,7 @@ impl<'tcx> Constructor<'tcx> { /// Returns whether `self` is covered by `other`, i.e. whether `self` is a subset of `other`. /// For the simple cases, this is simply checking for equality. For the "grouped" constructors, /// this checks for inclusion. - fn is_covered_by<'p>(&self, pcx: PatCtxt<'_, 'p, 'tcx>, other: &Self) -> bool { + pub(super) fn is_covered_by<'p>(&self, pcx: PatCtxt<'_, 'p, 'tcx>, other: &Self) -> bool { // This must be kept in sync with `is_covered_by_any`. match (self, other) { // Wildcards cover anything @@ -993,92 +815,204 @@ impl<'tcx> Constructor<'tcx> { } } } +} - /// Apply a constructor to a list of patterns, yielding a new pattern. `pats` - /// must have as many elements as this constructor's arity. - /// - /// This is roughly the inverse of `specialize_constructor`. - /// - /// Examples: - /// `self`: `Constructor::Single` - /// `ty`: `(u32, u32, u32)` - /// `pats`: `[10, 20, _]` - /// returns `(10, 20, _)` - /// - /// `self`: `Constructor::Variant(Option::Some)` - /// `ty`: `Option<bool>` - /// `pats`: `[false]` - /// returns `Some(false)` - fn apply<'p>(&self, pcx: PatCtxt<'_, 'p, 'tcx>, fields: Fields<'p, 'tcx>) -> Pat<'tcx> { - let mut subpatterns = fields.all_patterns(); +/// This determines the set of all possible constructors of a pattern matching +/// values of type `left_ty`. For vectors, this would normally be an infinite set +/// but is instead bounded by the maximum fixed length of slice patterns in +/// the column of patterns being analyzed. +/// +/// We make sure to omit constructors that are statically impossible. E.g., for +/// `Option<!>`, we do not include `Some(_)` in the returned list of constructors. +/// Invariant: this returns an empty `Vec` if and only if the type is uninhabited (as determined by +/// `cx.is_uninhabited()`). +fn all_constructors<'p, 'tcx>(pcx: PatCtxt<'_, 'p, 'tcx>) -> Vec<Constructor<'tcx>> { + debug!("all_constructors({:?})", pcx.ty); + let cx = pcx.cx; + let make_range = |start, end| { + IntRange( + // `unwrap()` is ok because we know the type is an integer. + IntRange::from_range(cx.tcx, start, end, pcx.ty, &RangeEnd::Included, pcx.span) + .unwrap(), + ) + }; + match pcx.ty.kind() { + ty::Bool => vec![make_range(0, 1)], + ty::Array(sub_ty, len) if len.try_eval_usize(cx.tcx, cx.param_env).is_some() => { + let len = len.eval_usize(cx.tcx, cx.param_env); + if len != 0 && cx.is_uninhabited(sub_ty) { + vec![] + } else { + vec![Slice(Slice::new(Some(len), VarLen(0, 0)))] + } + } + // Treat arrays of a constant but unknown length like slices. + ty::Array(sub_ty, _) | ty::Slice(sub_ty) => { + let kind = if cx.is_uninhabited(sub_ty) { FixedLen(0) } else { VarLen(0, 0) }; + vec![Slice(Slice::new(None, kind))] + } + ty::Adt(def, substs) if def.is_enum() => { + // If the enum is declared as `#[non_exhaustive]`, we treat it as if it had an + // additional "unknown" constructor. + // There is no point in enumerating all possible variants, because the user can't + // actually match against them all themselves. So we always return only the fictitious + // constructor. + // E.g., in an example like: + // + // ``` + // let err: io::ErrorKind = ...; + // match err { + // io::ErrorKind::NotFound => {}, + // } + // ``` + // + // we don't want to show every possible IO error, but instead have only `_` as the + // witness. + let is_declared_nonexhaustive = cx.is_foreign_non_exhaustive_enum(pcx.ty); - let pat = match self { - Single | Variant(_) => match pcx.ty.kind() { - ty::Adt(..) | ty::Tuple(..) => { - let subpatterns = subpatterns - .enumerate() - .map(|(i, p)| FieldPat { field: Field::new(i), pattern: p }) - .collect(); + // If `exhaustive_patterns` is disabled and our scrutinee is an empty enum, we treat it + // as though it had an "unknown" constructor to avoid exposing its emptiness. The + // exception is if the pattern is at the top level, because we want empty matches to be + // considered exhaustive. + let is_secretly_empty = def.variants.is_empty() + && !cx.tcx.features().exhaustive_patterns + && !pcx.is_top_level; - if let ty::Adt(adt, substs) = pcx.ty.kind() { - if adt.is_enum() { - PatKind::Variant { - adt_def: adt, - substs, - variant_index: self.variant_index_for_adt(adt), - subpatterns, - } - } else { - PatKind::Leaf { subpatterns } - } - } else { - PatKind::Leaf { subpatterns } - } - } - // Note: given the expansion of `&str` patterns done in `expand_pattern`, we should - // be careful to reconstruct the correct constant pattern here. However a string - // literal pattern will never be reported as a non-exhaustiveness witness, so we - // can ignore this issue. - ty::Ref(..) => PatKind::Deref { subpattern: subpatterns.next().unwrap() }, - ty::Slice(_) | ty::Array(..) => bug!("bad slice pattern {:?} {:?}", self, pcx.ty), - _ => PatKind::Wild, - }, - Slice(slice) => match slice.kind { - FixedLen(_) => { - PatKind::Slice { prefix: subpatterns.collect(), slice: None, suffix: vec![] } - } - VarLen(prefix, _) => { - let mut prefix: Vec<_> = subpatterns.by_ref().take(prefix as usize).collect(); - if slice.array_len.is_some() { - // Improves diagnostics a bit: if the type is a known-size array, instead - // of reporting `[x, _, .., _, y]`, we prefer to report `[x, .., y]`. - // This is incorrect if the size is not known, since `[_, ..]` captures - // arrays of lengths `>= 1` whereas `[..]` captures any length. - while !prefix.is_empty() && prefix.last().unwrap().is_wildcard() { - prefix.pop(); - } - } - let suffix: Vec<_> = if slice.array_len.is_some() { - // Same as above. - subpatterns.skip_while(Pat::is_wildcard).collect() - } else { - subpatterns.collect() - }; - let wild = Pat::wildcard_from_ty(pcx.ty); - PatKind::Slice { prefix, slice: Some(wild), suffix } - } - }, - &Str(value) => PatKind::Constant { value }, - &FloatRange(lo, hi, end) => PatKind::Range(PatRange { lo, hi, end }), - IntRange(range) => return range.to_pat(pcx.cx.tcx), - NonExhaustive => PatKind::Wild, - Opaque => bug!("we should not try to apply an opaque constructor"), - Wildcard => bug!( - "trying to apply a wildcard constructor; this should have been done in `apply_constructors`" - ), - }; + if is_secretly_empty || is_declared_nonexhaustive { + vec![NonExhaustive] + } else if cx.tcx.features().exhaustive_patterns { + // If `exhaustive_patterns` is enabled, we exclude variants known to be + // uninhabited. + def.variants + .iter() + .filter(|v| { + !v.uninhabited_from(cx.tcx, substs, def.adt_kind(), cx.param_env) + .contains(cx.tcx, cx.module) + }) + .map(|v| Variant(v.def_id)) + .collect() + } else { + def.variants.iter().map(|v| Variant(v.def_id)).collect() + } + } + ty::Char => { + vec![ + // The valid Unicode Scalar Value ranges. + make_range('\u{0000}' as u128, '\u{D7FF}' as u128), + make_range('\u{E000}' as u128, '\u{10FFFF}' as u128), + ] + } + ty::Int(_) | ty::Uint(_) + if pcx.ty.is_ptr_sized_integral() + && !cx.tcx.features().precise_pointer_size_matching => + { + // `usize`/`isize` are not allowed to be matched exhaustively unless the + // `precise_pointer_size_matching` feature is enabled. So we treat those types like + // `#[non_exhaustive]` enums by returning a special unmatcheable constructor. + vec![NonExhaustive] + } + &ty::Int(ity) => { + let bits = Integer::from_attr(&cx.tcx, SignedInt(ity)).size().bits() as u128; + let min = 1u128 << (bits - 1); + let max = min - 1; + vec![make_range(min, max)] + } + &ty::Uint(uty) => { + let size = Integer::from_attr(&cx.tcx, UnsignedInt(uty)).size(); + let max = size.truncate(u128::MAX); + vec![make_range(0, max)] + } + // If `exhaustive_patterns` is disabled and our scrutinee is the never type, we cannot + // expose its emptiness. The exception is if the pattern is at the top level, because we + // want empty matches to be considered exhaustive. + ty::Never if !cx.tcx.features().exhaustive_patterns && !pcx.is_top_level => { + vec![NonExhaustive] + } + ty::Never => vec![], + _ if cx.is_uninhabited(pcx.ty) => vec![], + ty::Adt(..) | ty::Tuple(..) | ty::Ref(..) => vec![Single], + // This type is one for which we cannot list constructors, like `str` or `f64`. + _ => vec![NonExhaustive], + } +} - Pat { ty: pcx.ty, span: DUMMY_SP, kind: Box::new(pat) } +// A struct to compute a set of constructors equivalent to `all_ctors \ used_ctors`. +#[derive(Debug)] +pub(super) struct MissingConstructors<'tcx> { + all_ctors: SmallVec<[Constructor<'tcx>; 1]>, + used_ctors: Vec<Constructor<'tcx>>, +} + +impl<'tcx> MissingConstructors<'tcx> { + pub(super) fn new<'p>(pcx: PatCtxt<'_, 'p, 'tcx>) -> Self { + let used_ctors: Vec<Constructor<'_>> = + pcx.matrix.head_ctors(pcx.cx).cloned().filter(|c| !c.is_wildcard()).collect(); + // Since `all_ctors` never contains wildcards, this won't recurse further. + let all_ctors = + all_constructors(pcx).into_iter().flat_map(|ctor| ctor.split(pcx, None)).collect(); + + MissingConstructors { all_ctors, used_ctors } + } + + fn is_empty<'p>(&self, pcx: PatCtxt<'_, 'p, 'tcx>) -> bool { + self.iter(pcx).next().is_none() + } + + /// Iterate over all_ctors \ used_ctors + fn iter<'a, 'p>( + &'a self, + pcx: PatCtxt<'a, 'p, 'tcx>, + ) -> impl Iterator<Item = &'a Constructor<'tcx>> + Captures<'p> { + self.all_ctors.iter().filter(move |ctor| !ctor.is_covered_by_any(pcx, &self.used_ctors)) + } + + /// List the patterns corresponding to the missing constructors. In some cases, instead of + /// listing all constructors of a given type, we prefer to simply report a wildcard. + pub(super) fn report_patterns<'p>( + &self, + pcx: PatCtxt<'_, 'p, 'tcx>, + ) -> SmallVec<[Pat<'tcx>; 1]> { + // There are 2 ways we can report a witness here. + // Commonly, we can report all the "free" + // constructors as witnesses, e.g., if we have: + // + // ``` + // enum Direction { N, S, E, W } + // let Direction::N = ...; + // ``` + // + // we can report 3 witnesses: `S`, `E`, and `W`. + // + // However, there is a case where we don't want + // to do this and instead report a single `_` witness: + // if the user didn't actually specify a constructor + // in this arm, e.g., in + // + // ``` + // let x: (Direction, Direction, bool) = ...; + // let (_, _, false) = x; + // ``` + // + // we don't want to show all 16 possible witnesses + // `(<direction-1>, <direction-2>, true)` - we are + // satisfied with `(_, _, true)`. In this case, + // `used_ctors` is empty. + // The exception is: if we are at the top-level, for example in an empty match, we + // sometimes prefer reporting the list of constructors instead of just `_`. + let report_when_all_missing = pcx.is_top_level && !IntRange::is_integral(pcx.ty); + if self.used_ctors.is_empty() && !report_when_all_missing { + // All constructors are unused. Report only a wildcard + // rather than each individual constructor. + smallvec![Pat::wildcard_from_ty(pcx.ty)] + } else { + // Construct for each missing constructor a "wild" version of this + // constructor, that matches everything that can be built with + // it. For example, if `ctor` is a `Constructor::Variant` for + // `Option::Some`, we get the pattern `Some(_)`. + self.iter(pcx) + .map(|missing_ctor| Fields::wildcards(pcx, &missing_ctor).apply(pcx, missing_ctor)) + .collect() + } } } @@ -1086,7 +1020,7 @@ impl<'tcx> Constructor<'tcx> { /// `Fields` struct. This struct represents such a potentially-hidden field. When a field is hidden /// we still keep its type around. #[derive(Debug, Copy, Clone)] -enum FilteredField<'p, 'tcx> { +pub(super) enum FilteredField<'p, 'tcx> { Kept(&'p Pat<'tcx>), Hidden(Ty<'tcx>), } @@ -1119,7 +1053,7 @@ impl<'p, 'tcx> FilteredField<'p, 'tcx> { /// This filtering is uncommon in practice, because uninhabited fields are rarely used, so we avoid /// it when possible to preserve performance. #[derive(Debug, Clone)] -enum Fields<'p, 'tcx> { +pub(super) enum Fields<'p, 'tcx> { /// Lists of patterns that don't contain any filtered fields. /// `Slice` and `Vec` behave the same; the difference is only to avoid allocating and /// triple-dereferences when possible. Frankly this is premature optimization, I (Nadrieril) @@ -1156,7 +1090,7 @@ impl<'p, 'tcx> Fields<'p, 'tcx> { } /// Creates a new list of wildcard fields for a given constructor. - fn wildcards(pcx: PatCtxt<'_, 'p, 'tcx>, constructor: &Constructor<'tcx>) -> Self { + pub(super) fn wildcards(pcx: PatCtxt<'_, 'p, 'tcx>, constructor: &Constructor<'tcx>) -> Self { let ty = pcx.ty; let cx = pcx.cx; let wildcard_from_ty = |ty| &*cx.pattern_arena.alloc(Pat::wildcard_from_ty(ty)); @@ -1228,10 +1162,97 @@ impl<'p, 'tcx> Fields<'p, 'tcx> { ret } + /// Apply a constructor to a list of patterns, yielding a new pattern. `self` + /// must have as many elements as this constructor's arity. + /// + /// This is roughly the inverse of `specialize_constructor`. + /// + /// Examples: + /// `ctor`: `Constructor::Single` + /// `ty`: `Foo(u32, u32, u32)` + /// `self`: `[10, 20, _]` + /// returns `Foo(10, 20, _)` + /// + /// `ctor`: `Constructor::Variant(Option::Some)` + /// `ty`: `Option<bool>` + /// `self`: `[false]` + /// returns `Some(false)` + pub(super) fn apply(self, pcx: PatCtxt<'_, 'p, 'tcx>, ctor: &Constructor<'tcx>) -> Pat<'tcx> { + let mut subpatterns = self.all_patterns(); + + let pat = match ctor { + Single | Variant(_) => match pcx.ty.kind() { + ty::Adt(..) | ty::Tuple(..) => { + let subpatterns = subpatterns + .enumerate() + .map(|(i, p)| FieldPat { field: Field::new(i), pattern: p }) + .collect(); + + if let ty::Adt(adt, substs) = pcx.ty.kind() { + if adt.is_enum() { + PatKind::Variant { + adt_def: adt, + substs, + variant_index: ctor.variant_index_for_adt(adt), + subpatterns, + } + } else { + PatKind::Leaf { subpatterns } + } + } else { + PatKind::Leaf { subpatterns } + } + } + // Note: given the expansion of `&str` patterns done in `expand_pattern`, we should + // be careful to reconstruct the correct constant pattern here. However a string + // literal pattern will never be reported as a non-exhaustiveness witness, so we + // can ignore this issue. + ty::Ref(..) => PatKind::Deref { subpattern: subpatterns.next().unwrap() }, + ty::Slice(_) | ty::Array(..) => bug!("bad slice pattern {:?} {:?}", ctor, pcx.ty), + _ => PatKind::Wild, + }, + Slice(slice) => match slice.kind { + FixedLen(_) => { + PatKind::Slice { prefix: subpatterns.collect(), slice: None, suffix: vec![] } + } + VarLen(prefix, _) => { + let mut prefix: Vec<_> = subpatterns.by_ref().take(prefix as usize).collect(); + if slice.array_len.is_some() { + // Improves diagnostics a bit: if the type is a known-size array, instead + // of reporting `[x, _, .., _, y]`, we prefer to report `[x, .., y]`. + // This is incorrect if the size is not known, since `[_, ..]` captures + // arrays of lengths `>= 1` whereas `[..]` captures any length. + while !prefix.is_empty() && prefix.last().unwrap().is_wildcard() { + prefix.pop(); + } + } + let suffix: Vec<_> = if slice.array_len.is_some() { + // Same as above. + subpatterns.skip_while(Pat::is_wildcard).collect() + } else { + subpatterns.collect() + }; + let wild = Pat::wildcard_from_ty(pcx.ty); + PatKind::Slice { prefix, slice: Some(wild), suffix } + } + }, + &Str(value) => PatKind::Constant { value }, + &FloatRange(lo, hi, end) => PatKind::Range(PatRange { lo, hi, end }), + IntRange(range) => return range.to_pat(pcx.cx.tcx), + NonExhaustive => PatKind::Wild, + Opaque => bug!("we should not try to apply an opaque constructor"), + Wildcard => bug!( + "trying to apply a wildcard constructor; this should have been done in `apply_constructors`" + ), + }; + + Pat { ty: pcx.ty, span: DUMMY_SP, kind: Box::new(pat) } + } + /// Returns the number of patterns from the viewpoint of match-checking, i.e. excluding hidden /// fields. This is what we want in most cases in this file, the only exception being /// conversion to/from `Pat`. - fn len(&self) -> usize { + pub(super) fn len(&self) -> usize { match self { Fields::Slice(pats) => pats.len(), Fields::Vec(pats) => pats.len(), @@ -1252,6 +1273,18 @@ impl<'p, 'tcx> Fields<'p, 'tcx> { pats.into_iter() } + /// Returns the filtered list of patterns, not including hidden fields. + pub(super) fn filtered_patterns(self) -> SmallVec<[&'p Pat<'tcx>; 2]> { + match self { + Fields::Slice(pats) => pats.iter().collect(), + Fields::Vec(pats) => pats, + Fields::Filtered { fields, .. } => { + // We skip hidden fields here + fields.into_iter().filter_map(|p| p.kept()).collect() + } + } + } + /// Overrides some of the fields with the provided patterns. Exactly like /// `replace_fields_indexed`, except that it takes `FieldPat`s as input. fn replace_with_fieldpats( @@ -1297,7 +1330,7 @@ impl<'p, 'tcx> Fields<'p, 'tcx> { /// Replaces contained fields with the given filtered list of patterns, e.g. taken from the /// matrix. There must be `len()` patterns in `pats`. - fn replace_fields( + pub(super) fn replace_fields( &self, cx: &MatchCheckCtxt<'p, 'tcx>, pats: impl IntoIterator<Item = Pat<'tcx>>, @@ -1335,7 +1368,7 @@ impl<'p, 'tcx> Fields<'p, 'tcx> { /// } /// ``` /// This is guaranteed to preserve the number of patterns in `self`. - fn replace_with_pattern_arguments(&self, pat: &'p Pat<'tcx>) -> Self { + pub(super) fn replace_with_pattern_arguments(&self, pat: &'p Pat<'tcx>) -> Self { match pat.kind.as_ref() { PatKind::Deref { subpattern } => { assert_eq!(self.len(), 1); @@ -1358,1005 +1391,4 @@ impl<'p, 'tcx> Fields<'p, 'tcx> { _ => self.clone(), } } - - fn push_on_patstack(self, stack: &[&'p Pat<'tcx>]) -> PatStack<'p, 'tcx> { - let pats: SmallVec<_> = match self { - Fields::Slice(pats) => pats.iter().chain(stack.iter().copied()).collect(), - Fields::Vec(mut pats) => { - pats.extend_from_slice(stack); - pats - } - Fields::Filtered { fields, .. } => { - // We skip hidden fields here - fields.into_iter().filter_map(|p| p.kept()).chain(stack.iter().copied()).collect() - } - }; - PatStack::from_vec(pats) - } -} - -#[derive(Clone, Debug)] -crate enum Usefulness<'tcx> { - /// Carries, for each column in the matrix, a set of sub-branches that have been found to be - /// unreachable. Used only in the presence of or-patterns, otherwise it stays empty. - Useful(Vec<FxHashSet<Span>>), - /// Carries a list of witnesses of non-exhaustiveness. - UsefulWithWitness(Vec<Witness<'tcx>>), - NotUseful, -} - -impl<'tcx> Usefulness<'tcx> { - fn new_useful(preference: WitnessPreference) -> Self { - match preference { - ConstructWitness => UsefulWithWitness(vec![Witness(vec![])]), - LeaveOutWitness => Useful(vec![]), - } - } - - fn is_useful(&self) -> bool { - !matches!(*self, NotUseful) - } - - fn apply_constructor<'p>( - self, - pcx: PatCtxt<'_, 'p, 'tcx>, - ctor: &Constructor<'tcx>, - ctor_wild_subpatterns: &Fields<'p, 'tcx>, - ) -> Self { - match self { - UsefulWithWitness(witnesses) => { - let new_witnesses = if ctor.is_wildcard() { - let missing_ctors = MissingConstructors::new(pcx); - let new_patterns = missing_ctors.report_patterns(pcx); - witnesses - .into_iter() - .flat_map(|witness| { - new_patterns.iter().map(move |pat| { - let mut witness = witness.clone(); - witness.0.push(pat.clone()); - witness - }) - }) - .collect() - } else { - witnesses - .into_iter() - .map(|witness| witness.apply_constructor(pcx, &ctor, ctor_wild_subpatterns)) - .collect() - }; - UsefulWithWitness(new_witnesses) - } - Useful(mut unreachables) => { - if !unreachables.is_empty() { - // When we apply a constructor, there are `arity` columns of the matrix that - // corresponded to its arguments. All the unreachables found in these columns - // will, after `apply`, come from the first column. So we take the union of all - // the corresponding sets and put them in the first column. - // Note that `arity` may be 0, in which case we just push a new empty set. - let len = unreachables.len(); - let arity = ctor_wild_subpatterns.len(); - let mut unioned = FxHashSet::default(); - for set in unreachables.drain((len - arity)..) { - unioned.extend(set) - } - unreachables.push(unioned); - } - Useful(unreachables) - } - x => x, - } - } -} - -#[derive(Copy, Clone, Debug)] -enum WitnessPreference { - ConstructWitness, - LeaveOutWitness, -} - -#[derive(Copy, Clone)] -struct PatCtxt<'a, 'p, 'tcx> { - cx: &'a MatchCheckCtxt<'p, 'tcx>, - /// Current state of the matrix. - matrix: &'a Matrix<'p, 'tcx>, - /// Type of the current column under investigation. - ty: Ty<'tcx>, - /// Span of the current pattern under investigation. - span: Span, - /// Whether the current pattern is the whole pattern as found in a match arm, or if it's a - /// subpattern. - is_top_level: bool, -} - -/// A witness of non-exhaustiveness for error reporting, represented -/// as a list of patterns (in reverse order of construction) with -/// wildcards inside to represent elements that can take any inhabitant -/// of the type as a value. -/// -/// A witness against a list of patterns should have the same types -/// and length as the pattern matched against. Because Rust `match` -/// is always against a single pattern, at the end the witness will -/// have length 1, but in the middle of the algorithm, it can contain -/// multiple patterns. -/// -/// For example, if we are constructing a witness for the match against -/// -/// ``` -/// struct Pair(Option<(u32, u32)>, bool); -/// -/// match (p: Pair) { -/// Pair(None, _) => {} -/// Pair(_, false) => {} -/// } -/// ``` -/// -/// We'll perform the following steps: -/// 1. Start with an empty witness -/// `Witness(vec![])` -/// 2. Push a witness `Some(_)` against the `None` -/// `Witness(vec![Some(_)])` -/// 3. Push a witness `true` against the `false` -/// `Witness(vec![Some(_), true])` -/// 4. Apply the `Pair` constructor to the witnesses -/// `Witness(vec![Pair(Some(_), true)])` -/// -/// The final `Pair(Some(_), true)` is then the resulting witness. -#[derive(Clone, Debug)] -crate struct Witness<'tcx>(Vec<Pat<'tcx>>); - -impl<'tcx> Witness<'tcx> { - /// Asserts that the witness contains a single pattern, and returns it. - fn single_pattern(self) -> Pat<'tcx> { - assert_eq!(self.0.len(), 1); - self.0.into_iter().next().unwrap() - } - - /// Constructs a partial witness for a pattern given a list of - /// patterns expanded by the specialization step. - /// - /// When a pattern P is discovered to be useful, this function is used bottom-up - /// to reconstruct a complete witness, e.g., a pattern P' that covers a subset - /// of values, V, where each value in that set is not covered by any previously - /// used patterns and is covered by the pattern P'. Examples: - /// - /// left_ty: tuple of 3 elements - /// pats: [10, 20, _] => (10, 20, _) - /// - /// left_ty: struct X { a: (bool, &'static str), b: usize} - /// pats: [(false, "foo"), 42] => X { a: (false, "foo"), b: 42 } - fn apply_constructor<'p>( - mut self, - pcx: PatCtxt<'_, 'p, 'tcx>, - ctor: &Constructor<'tcx>, - ctor_wild_subpatterns: &Fields<'p, 'tcx>, - ) -> Self { - let pat = { - let len = self.0.len(); - let arity = ctor_wild_subpatterns.len(); - let pats = self.0.drain((len - arity)..).rev(); - let fields = ctor_wild_subpatterns.replace_fields(pcx.cx, pats); - ctor.apply(pcx, fields) - }; - - self.0.push(pat); - - self - } -} - -/// This determines the set of all possible constructors of a pattern matching -/// values of type `left_ty`. For vectors, this would normally be an infinite set -/// but is instead bounded by the maximum fixed length of slice patterns in -/// the column of patterns being analyzed. -/// -/// We make sure to omit constructors that are statically impossible. E.g., for -/// `Option<!>`, we do not include `Some(_)` in the returned list of constructors. -/// Invariant: this returns an empty `Vec` if and only if the type is uninhabited (as determined by -/// `cx.is_uninhabited()`). -fn all_constructors<'p, 'tcx>(pcx: PatCtxt<'_, 'p, 'tcx>) -> Vec<Constructor<'tcx>> { - debug!("all_constructors({:?})", pcx.ty); - let cx = pcx.cx; - let make_range = |start, end| { - IntRange( - // `unwrap()` is ok because we know the type is an integer. - IntRange::from_range(cx.tcx, start, end, pcx.ty, &RangeEnd::Included, pcx.span) - .unwrap(), - ) - }; - match pcx.ty.kind() { - ty::Bool => vec![make_range(0, 1)], - ty::Array(sub_ty, len) if len.try_eval_usize(cx.tcx, cx.param_env).is_some() => { - let len = len.eval_usize(cx.tcx, cx.param_env); - if len != 0 && cx.is_uninhabited(sub_ty) { - vec![] - } else { - vec![Slice(Slice::new(Some(len), VarLen(0, 0)))] - } - } - // Treat arrays of a constant but unknown length like slices. - ty::Array(sub_ty, _) | ty::Slice(sub_ty) => { - let kind = if cx.is_uninhabited(sub_ty) { FixedLen(0) } else { VarLen(0, 0) }; - vec![Slice(Slice::new(None, kind))] - } - ty::Adt(def, substs) if def.is_enum() => { - // If the enum is declared as `#[non_exhaustive]`, we treat it as if it had an - // additional "unknown" constructor. - // There is no point in enumerating all possible variants, because the user can't - // actually match against them all themselves. So we always return only the fictitious - // constructor. - // E.g., in an example like: - // - // ``` - // let err: io::ErrorKind = ...; - // match err { - // io::ErrorKind::NotFound => {}, - // } - // ``` - // - // we don't want to show every possible IO error, but instead have only `_` as the - // witness. - let is_declared_nonexhaustive = cx.is_foreign_non_exhaustive_enum(pcx.ty); - - // If `exhaustive_patterns` is disabled and our scrutinee is an empty enum, we treat it - // as though it had an "unknown" constructor to avoid exposing its emptiness. The - // exception is if the pattern is at the top level, because we want empty matches to be - // considered exhaustive. - let is_secretly_empty = def.variants.is_empty() - && !cx.tcx.features().exhaustive_patterns - && !pcx.is_top_level; - - if is_secretly_empty || is_declared_nonexhaustive { - vec![NonExhaustive] - } else if cx.tcx.features().exhaustive_patterns { - // If `exhaustive_patterns` is enabled, we exclude variants known to be - // uninhabited. - def.variants - .iter() - .filter(|v| { - !v.uninhabited_from(cx.tcx, substs, def.adt_kind(), cx.param_env) - .contains(cx.tcx, cx.module) - }) - .map(|v| Variant(v.def_id)) - .collect() - } else { - def.variants.iter().map(|v| Variant(v.def_id)).collect() - } - } - ty::Char => { - vec![ - // The valid Unicode Scalar Value ranges. - make_range('\u{0000}' as u128, '\u{D7FF}' as u128), - make_range('\u{E000}' as u128, '\u{10FFFF}' as u128), - ] - } - ty::Int(_) | ty::Uint(_) - if pcx.ty.is_ptr_sized_integral() - && !cx.tcx.features().precise_pointer_size_matching => - { - // `usize`/`isize` are not allowed to be matched exhaustively unless the - // `precise_pointer_size_matching` feature is enabled. So we treat those types like - // `#[non_exhaustive]` enums by returning a special unmatcheable constructor. - vec![NonExhaustive] - } - &ty::Int(ity) => { - let bits = Integer::from_attr(&cx.tcx, SignedInt(ity)).size().bits() as u128; - let min = 1u128 << (bits - 1); - let max = min - 1; - vec![make_range(min, max)] - } - &ty::Uint(uty) => { - let size = Integer::from_attr(&cx.tcx, UnsignedInt(uty)).size(); - let max = size.truncate(u128::MAX); - vec![make_range(0, max)] - } - // If `exhaustive_patterns` is disabled and our scrutinee is the never type, we cannot - // expose its emptiness. The exception is if the pattern is at the top level, because we - // want empty matches to be considered exhaustive. - ty::Never if !cx.tcx.features().exhaustive_patterns && !pcx.is_top_level => { - vec![NonExhaustive] - } - ty::Never => vec![], - _ if cx.is_uninhabited(pcx.ty) => vec![], - ty::Adt(..) | ty::Tuple(..) | ty::Ref(..) => vec![Single], - // This type is one for which we cannot list constructors, like `str` or `f64`. - _ => vec![NonExhaustive], - } -} - -/// An inclusive interval, used for precise integer exhaustiveness checking. -/// `IntRange`s always store a contiguous range. This means that values are -/// encoded such that `0` encodes the minimum value for the integer, -/// regardless of the signedness. -/// For example, the pattern `-128..=127i8` is encoded as `0..=255`. -/// This makes comparisons and arithmetic on interval endpoints much more -/// straightforward. See `signed_bias` for details. -/// -/// `IntRange` is never used to encode an empty range or a "range" that wraps -/// around the (offset) space: i.e., `range.lo <= range.hi`. -#[derive(Clone, Debug)] -struct IntRange<'tcx> { - range: RangeInclusive<u128>, - ty: Ty<'tcx>, - span: Span, -} - -impl<'tcx> IntRange<'tcx> { - #[inline] - fn is_integral(ty: Ty<'_>) -> bool { - matches!(ty.kind(), ty::Char | ty::Int(_) | ty::Uint(_) | ty::Bool) - } - - fn is_singleton(&self) -> bool { - self.range.start() == self.range.end() - } - - fn boundaries(&self) -> (u128, u128) { - (*self.range.start(), *self.range.end()) - } - - /// Don't treat `usize`/`isize` exhaustively unless the `precise_pointer_size_matching` feature - /// is enabled. - fn treat_exhaustively(&self, tcx: TyCtxt<'tcx>) -> bool { - !self.ty.is_ptr_sized_integral() || tcx.features().precise_pointer_size_matching - } - - #[inline] - fn integral_size_and_signed_bias(tcx: TyCtxt<'tcx>, ty: Ty<'_>) -> Option<(Size, u128)> { - match *ty.kind() { - ty::Bool => Some((Size::from_bytes(1), 0)), - ty::Char => Some((Size::from_bytes(4), 0)), - ty::Int(ity) => { - let size = Integer::from_attr(&tcx, SignedInt(ity)).size(); - Some((size, 1u128 << (size.bits() as u128 - 1))) - } - ty::Uint(uty) => Some((Integer::from_attr(&tcx, UnsignedInt(uty)).size(), 0)), - _ => None, - } - } - - #[inline] - fn from_const( - tcx: TyCtxt<'tcx>, - param_env: ty::ParamEnv<'tcx>, - value: &Const<'tcx>, - span: Span, - ) -> Option<IntRange<'tcx>> { - if let Some((target_size, bias)) = Self::integral_size_and_signed_bias(tcx, value.ty) { - let ty = value.ty; - let val = (|| { - if let ty::ConstKind::Value(ConstValue::Scalar(scalar)) = value.val { - // For this specific pattern we can skip a lot of effort and go - // straight to the result, after doing a bit of checking. (We - // could remove this branch and just fall through, which - // is more general but much slower.) - if let Ok(bits) = scalar.to_bits_or_ptr(target_size, &tcx) { - return Some(bits); - } - } - // This is a more general form of the previous case. - value.try_eval_bits(tcx, param_env, ty) - })()?; - let val = val ^ bias; - Some(IntRange { range: val..=val, ty, span }) - } else { - None - } - } - - #[inline] - fn from_range( - tcx: TyCtxt<'tcx>, - lo: u128, - hi: u128, - ty: Ty<'tcx>, - end: &RangeEnd, - span: Span, - ) -> Option<IntRange<'tcx>> { - if Self::is_integral(ty) { - // Perform a shift if the underlying types are signed, - // which makes the interval arithmetic simpler. - let bias = IntRange::signed_bias(tcx, ty); - let (lo, hi) = (lo ^ bias, hi ^ bias); - let offset = (*end == RangeEnd::Excluded) as u128; - if lo > hi || (lo == hi && *end == RangeEnd::Excluded) { - // This should have been caught earlier by E0030. - bug!("malformed range pattern: {}..={}", lo, (hi - offset)); - } - Some(IntRange { range: lo..=(hi - offset), ty, span }) - } else { - None - } - } - - // The return value of `signed_bias` should be XORed with an endpoint to encode/decode it. - fn signed_bias(tcx: TyCtxt<'tcx>, ty: Ty<'tcx>) -> u128 { - match *ty.kind() { - ty::Int(ity) => { - let bits = Integer::from_attr(&tcx, SignedInt(ity)).size().bits() as u128; - 1u128 << (bits - 1) - } - _ => 0, - } - } - - fn is_subrange(&self, other: &Self) -> bool { - other.range.start() <= self.range.start() && self.range.end() <= other.range.end() - } - - fn intersection(&self, tcx: TyCtxt<'tcx>, other: &Self) -> Option<Self> { - let ty = self.ty; - let (lo, hi) = self.boundaries(); - let (other_lo, other_hi) = other.boundaries(); - if self.treat_exhaustively(tcx) { - if lo <= other_hi && other_lo <= hi { - let span = other.span; - Some(IntRange { range: max(lo, other_lo)..=min(hi, other_hi), ty, span }) - } else { - None - } - } else { - // If the range should not be treated exhaustively, fallback to checking for inclusion. - if self.is_subrange(other) { Some(self.clone()) } else { None } - } - } - - fn suspicious_intersection(&self, other: &Self) -> bool { - // `false` in the following cases: - // 1 ---- // 1 ---------- // 1 ---- // 1 ---- - // 2 ---------- // 2 ---- // 2 ---- // 2 ---- - // - // The following are currently `false`, but could be `true` in the future (#64007): - // 1 --------- // 1 --------- - // 2 ---------- // 2 ---------- - // - // `true` in the following cases: - // 1 ------- // 1 ------- - // 2 -------- // 2 ------- - let (lo, hi) = self.boundaries(); - let (other_lo, other_hi) = other.boundaries(); - lo == other_hi || hi == other_lo - } - - fn to_pat(&self, tcx: TyCtxt<'tcx>) -> Pat<'tcx> { - let (lo, hi) = self.boundaries(); - - let bias = IntRange::signed_bias(tcx, self.ty); - let (lo, hi) = (lo ^ bias, hi ^ bias); - - let ty = ty::ParamEnv::empty().and(self.ty); - let lo_const = ty::Const::from_bits(tcx, lo, ty); - let hi_const = ty::Const::from_bits(tcx, hi, ty); - - let kind = if lo == hi { - PatKind::Constant { value: lo_const } - } else { - PatKind::Range(PatRange { lo: lo_const, hi: hi_const, end: RangeEnd::Included }) - }; - - // This is a brand new pattern, so we don't reuse `self.span`. - Pat { ty: self.ty, span: DUMMY_SP, kind: Box::new(kind) } - } - - /// For exhaustive integer matching, some constructors are grouped within other constructors - /// (namely integer typed values are grouped within ranges). However, when specialising these - /// constructors, we want to be specialising for the underlying constructors (the integers), not - /// the groups (the ranges). Thus we need to split the groups up. Splitting them up naïvely would - /// mean creating a separate constructor for every single value in the range, which is clearly - /// impractical. However, observe that for some ranges of integers, the specialisation will be - /// identical across all values in that range (i.e., there are equivalence classes of ranges of - /// constructors based on their `U(S(c, P), S(c, p))` outcome). These classes are grouped by - /// the patterns that apply to them (in the matrix `P`). We can split the range whenever the - /// patterns that apply to that range (specifically: the patterns that *intersect* with that range) - /// change. - /// Our solution, therefore, is to split the range constructor into subranges at every single point - /// the group of intersecting patterns changes (using the method described below). - /// And voilà! We're testing precisely those ranges that we need to, without any exhaustive matching - /// on actual integers. The nice thing about this is that the number of subranges is linear in the - /// number of rows in the matrix (i.e., the number of cases in the `match` statement), so we don't - /// need to be worried about matching over gargantuan ranges. - /// - /// Essentially, given the first column of a matrix representing ranges, looking like the following: - /// - /// |------| |----------| |-------| || - /// |-------| |-------| |----| || - /// |---------| - /// - /// We split the ranges up into equivalence classes so the ranges are no longer overlapping: - /// - /// |--|--|||-||||--||---|||-------| |-|||| || - /// - /// The logic for determining how to split the ranges is fairly straightforward: we calculate - /// boundaries for each interval range, sort them, then create constructors for each new interval - /// between every pair of boundary points. (This essentially sums up to performing the intuitive - /// merging operation depicted above.) - fn split<'p>( - &self, - pcx: PatCtxt<'_, 'p, 'tcx>, - hir_id: Option<HirId>, - ) -> SmallVec<[Constructor<'tcx>; 1]> { - let ty = pcx.ty; - - /// Represents a border between 2 integers. Because the intervals spanning borders - /// must be able to cover every integer, we need to be able to represent - /// 2^128 + 1 such borders. - #[derive(Clone, Copy, PartialEq, Eq, PartialOrd, Ord, Debug)] - enum Border { - JustBefore(u128), - AfterMax, - } - - // A function for extracting the borders of an integer interval. - fn range_borders(r: IntRange<'_>) -> impl Iterator<Item = Border> { - let (lo, hi) = r.range.into_inner(); - let from = Border::JustBefore(lo); - let to = match hi.checked_add(1) { - Some(m) => Border::JustBefore(m), - None => Border::AfterMax, - }; - vec![from, to].into_iter() - } - - // Collect the span and range of all the intersecting ranges to lint on likely - // incorrect range patterns. (#63987) - let mut overlaps = vec![]; - let row_len = pcx.matrix.patterns.get(0).map(|r| r.len()).unwrap_or(0); - // `borders` is the set of borders between equivalence classes: each equivalence - // class lies between 2 borders. - let row_borders = pcx - .matrix - .head_ctors(pcx.cx) - .filter_map(|ctor| ctor.as_int_range()) - .filter_map(|range| { - let intersection = self.intersection(pcx.cx.tcx, &range); - let should_lint = self.suspicious_intersection(&range); - if let (Some(range), 1, true) = (&intersection, row_len, should_lint) { - // FIXME: for now, only check for overlapping ranges on simple range - // patterns. Otherwise with the current logic the following is detected - // as overlapping: - // match (10u8, true) { - // (0 ..= 125, false) => {} - // (126 ..= 255, false) => {} - // (0 ..= 255, true) => {} - // } - overlaps.push(range.clone()); - } - intersection - }) - .flat_map(range_borders); - let self_borders = range_borders(self.clone()); - let mut borders: Vec<_> = row_borders.chain(self_borders).collect(); - borders.sort_unstable(); - - self.lint_overlapping_patterns(pcx.cx.tcx, hir_id, ty, overlaps); - - // We're going to iterate through every adjacent pair of borders, making sure that - // each represents an interval of nonnegative length, and convert each such - // interval into a constructor. - borders - .array_windows() - .filter_map(|&pair| match pair { - [Border::JustBefore(n), Border::JustBefore(m)] => { - if n < m { - Some(n..=(m - 1)) - } else { - None - } - } - [Border::JustBefore(n), Border::AfterMax] => Some(n..=u128::MAX), - [Border::AfterMax, _] => None, - }) - .map(|range| IntRange { range, ty, span: pcx.span }) - .map(IntRange) - .collect() - } - - fn lint_overlapping_patterns( - &self, - tcx: TyCtxt<'tcx>, - hir_id: Option<HirId>, - ty: Ty<'tcx>, - overlaps: Vec<IntRange<'tcx>>, - ) { - if let (true, Some(hir_id)) = (!overlaps.is_empty(), hir_id) { - tcx.struct_span_lint_hir( - lint::builtin::OVERLAPPING_PATTERNS, - hir_id, - self.span, - |lint| { - let mut err = lint.build("multiple patterns covering the same range"); - err.span_label(self.span, "overlapping patterns"); - for int_range in overlaps { - // Use the real type for user display of the ranges: - err.span_label( - int_range.span, - &format!( - "this range overlaps on `{}`", - IntRange { range: int_range.range, ty, span: DUMMY_SP }.to_pat(tcx), - ), - ); - } - err.emit(); - }, - ); - } - } - - /// See `Constructor::is_covered_by` - fn is_covered_by<'p>(&self, pcx: PatCtxt<'_, 'p, 'tcx>, other: &Self) -> bool { - if self.intersection(pcx.cx.tcx, other).is_some() { - // Constructor splitting should ensure that all intersections we encounter are actually - // inclusions. - assert!(self.is_subrange(other)); - true - } else { - false - } - } -} - -/// Ignore spans when comparing, they don't carry semantic information as they are only for lints. -impl<'tcx> std::cmp::PartialEq for IntRange<'tcx> { - fn eq(&self, other: &Self) -> bool { - self.range == other.range && self.ty == other.ty - } -} - -// A struct to compute a set of constructors equivalent to `all_ctors \ used_ctors`. -#[derive(Debug)] -struct MissingConstructors<'tcx> { - all_ctors: SmallVec<[Constructor<'tcx>; 1]>, - used_ctors: Vec<Constructor<'tcx>>, -} - -impl<'tcx> MissingConstructors<'tcx> { - fn new<'p>(pcx: PatCtxt<'_, 'p, 'tcx>) -> Self { - let used_ctors: Vec<Constructor<'_>> = - pcx.matrix.head_ctors(pcx.cx).cloned().filter(|c| !c.is_wildcard()).collect(); - // Since `all_ctors` never contains wildcards, this won't recurse further. - let all_ctors = - all_constructors(pcx).into_iter().flat_map(|ctor| ctor.split(pcx, None)).collect(); - - MissingConstructors { all_ctors, used_ctors } - } - - fn is_empty<'p>(&self, pcx: PatCtxt<'_, 'p, 'tcx>) -> bool { - self.iter(pcx).next().is_none() - } - - /// Iterate over all_ctors \ used_ctors - fn iter<'a, 'p>( - &'a self, - pcx: PatCtxt<'a, 'p, 'tcx>, - ) -> impl Iterator<Item = &'a Constructor<'tcx>> + Captures<'p> { - self.all_ctors.iter().filter(move |ctor| !ctor.is_covered_by_any(pcx, &self.used_ctors)) - } - - /// List the patterns corresponding to the missing constructors. In some cases, instead of - /// listing all constructors of a given type, we prefer to simply report a wildcard. - fn report_patterns<'p>(&self, pcx: PatCtxt<'_, 'p, 'tcx>) -> SmallVec<[Pat<'tcx>; 1]> { - // There are 2 ways we can report a witness here. - // Commonly, we can report all the "free" - // constructors as witnesses, e.g., if we have: - // - // ``` - // enum Direction { N, S, E, W } - // let Direction::N = ...; - // ``` - // - // we can report 3 witnesses: `S`, `E`, and `W`. - // - // However, there is a case where we don't want - // to do this and instead report a single `_` witness: - // if the user didn't actually specify a constructor - // in this arm, e.g., in - // - // ``` - // let x: (Direction, Direction, bool) = ...; - // let (_, _, false) = x; - // ``` - // - // we don't want to show all 16 possible witnesses - // `(<direction-1>, <direction-2>, true)` - we are - // satisfied with `(_, _, true)`. In this case, - // `used_ctors` is empty. - // The exception is: if we are at the top-level, for example in an empty match, we - // sometimes prefer reporting the list of constructors instead of just `_`. - let report_when_all_missing = pcx.is_top_level && !IntRange::is_integral(pcx.ty); - if self.used_ctors.is_empty() && !report_when_all_missing { - // All constructors are unused. Report only a wildcard - // rather than each individual constructor. - smallvec![Pat::wildcard_from_ty(pcx.ty)] - } else { - // Construct for each missing constructor a "wild" version of this - // constructor, that matches everything that can be built with - // it. For example, if `ctor` is a `Constructor::Variant` for - // `Option::Some`, we get the pattern `Some(_)`. - self.iter(pcx) - .map(|missing_ctor| { - let fields = Fields::wildcards(pcx, &missing_ctor); - missing_ctor.apply(pcx, fields) - }) - .collect() - } - } -} - -/// Algorithm from <http://moscova.inria.fr/~maranget/papers/warn/index.html>. -/// The algorithm from the paper has been modified to correctly handle empty -/// types. The changes are: -/// (0) We don't exit early if the pattern matrix has zero rows. We just -/// continue to recurse over columns. -/// (1) all_constructors will only return constructors that are statically -/// possible. E.g., it will only return `Ok` for `Result<T, !>`. -/// -/// This finds whether a (row) vector `v` of patterns is 'useful' in relation -/// to a set of such vectors `m` - this is defined as there being a set of -/// inputs that will match `v` but not any of the sets in `m`. -/// -/// All the patterns at each column of the `matrix ++ v` matrix must have the same type. -/// -/// This is used both for reachability checking (if a pattern isn't useful in -/// relation to preceding patterns, it is not reachable) and exhaustiveness -/// checking (if a wildcard pattern is useful in relation to a matrix, the -/// matrix isn't exhaustive). -/// -/// `is_under_guard` is used to inform if the pattern has a guard. If it -/// has one it must not be inserted into the matrix. This shouldn't be -/// relied on for soundness. -fn is_useful<'p, 'tcx>( - cx: &MatchCheckCtxt<'p, 'tcx>, - matrix: &Matrix<'p, 'tcx>, - v: &PatStack<'p, 'tcx>, - witness_preference: WitnessPreference, - hir_id: HirId, - is_under_guard: bool, - is_top_level: bool, -) -> Usefulness<'tcx> { - let Matrix { patterns: rows, .. } = matrix; - debug!("is_useful({:#?}, {:#?})", matrix, v); - - // The base case. We are pattern-matching on () and the return value is - // based on whether our matrix has a row or not. - // NOTE: This could potentially be optimized by checking rows.is_empty() - // first and then, if v is non-empty, the return value is based on whether - // the type of the tuple we're checking is inhabited or not. - if v.is_empty() { - return if rows.is_empty() { - Usefulness::new_useful(witness_preference) - } else { - NotUseful - }; - }; - - assert!(rows.iter().all(|r| r.len() == v.len())); - - // If the first pattern is an or-pattern, expand it. - if let Some(vs) = v.expand_or_pat() { - // We expand the or pattern, trying each of its branches in turn and keeping careful track - // of possible unreachable sub-branches. - // - // If two branches have detected some unreachable sub-branches, we need to be careful. If - // they were detected in columns that are not the current one, we want to keep only the - // sub-branches that were unreachable in _all_ branches. Eg. in the following, the last - // `true` is unreachable in the second branch of the first or-pattern, but not otherwise. - // Therefore we don't want to lint that it is unreachable. - // - // ``` - // match (true, true) { - // (true, true) => {} - // (false | true, false | true) => {} - // } - // ``` - // If however the sub-branches come from the current column, they come from the inside of - // the current or-pattern, and we want to keep them all. Eg. in the following, we _do_ want - // to lint that the last `false` is unreachable. - // ``` - // match None { - // Some(false) => {} - // None | Some(true | false) => {} - // } - // ``` - - let mut matrix = matrix.clone(); - // We keep track of sub-branches separately depending on whether they come from this column - // or from others. - let mut unreachables_this_column: FxHashSet<Span> = FxHashSet::default(); - let mut unreachables_other_columns: Vec<FxHashSet<Span>> = Vec::default(); - // Whether at least one branch is reachable. - let mut any_is_useful = false; - - for v in vs { - let res = is_useful(cx, &matrix, &v, witness_preference, hir_id, is_under_guard, false); - match res { - Useful(unreachables) => { - if let Some((this_column, other_columns)) = unreachables.split_last() { - // We keep the union of unreachables found in the first column. - unreachables_this_column.extend(this_column); - // We keep the intersection of unreachables found in other columns. - if unreachables_other_columns.is_empty() { - unreachables_other_columns = other_columns.to_vec(); - } else { - unreachables_other_columns = unreachables_other_columns - .into_iter() - .zip(other_columns) - .map(|(x, y)| x.intersection(&y).copied().collect()) - .collect(); - } - } - any_is_useful = true; - } - NotUseful => { - unreachables_this_column.insert(v.head().span); - } - UsefulWithWitness(_) => bug!( - "encountered or-pat in the expansion of `_` during exhaustiveness checking" - ), - } - - // If pattern has a guard don't add it to the matrix. - if !is_under_guard { - // We push the already-seen patterns into the matrix in order to detect redundant - // branches like `Some(_) | Some(0)`. - matrix.push(v); - } - } - - return if any_is_useful { - let mut unreachables = if unreachables_other_columns.is_empty() { - let n_columns = v.len(); - (0..n_columns - 1).map(|_| FxHashSet::default()).collect() - } else { - unreachables_other_columns - }; - unreachables.push(unreachables_this_column); - Useful(unreachables) - } else { - NotUseful - }; - } - - // FIXME(Nadrieril): Hack to work around type normalization issues (see #72476). - let ty = matrix.heads().next().map(|r| r.ty).unwrap_or(v.head().ty); - let pcx = PatCtxt { cx, matrix, ty, span: v.head().span, is_top_level }; - - debug!("is_useful_expand_first_col: ty={:#?}, expanding {:#?}", pcx.ty, v.head()); - - let ret = v - .head_ctor(cx) - .split(pcx, Some(hir_id)) - .into_iter() - .map(|ctor| { - // We cache the result of `Fields::wildcards` because it is used a lot. - let ctor_wild_subpatterns = Fields::wildcards(pcx, &ctor); - let matrix = pcx.matrix.specialize_constructor(pcx, &ctor, &ctor_wild_subpatterns); - let v = v.pop_head_constructor(&ctor_wild_subpatterns); - let usefulness = - is_useful(pcx.cx, &matrix, &v, witness_preference, hir_id, is_under_guard, false); - usefulness.apply_constructor(pcx, &ctor, &ctor_wild_subpatterns) - }) - .find(|result| result.is_useful()) - .unwrap_or(NotUseful); - debug!("is_useful::returns({:#?}, {:#?}) = {:?}", matrix, v, ret); - ret -} - -/// Determines the constructor that the given pattern can be specialized to. -/// Returns `None` in case of a catch-all, which can't be specialized. -fn pat_constructor<'p, 'tcx>( - cx: &MatchCheckCtxt<'p, 'tcx>, - pat: &'p Pat<'tcx>, -) -> Constructor<'tcx> { - match pat.kind.as_ref() { - PatKind::AscribeUserType { .. } => bug!(), // Handled by `expand_pattern` - PatKind::Binding { .. } | PatKind::Wild => Wildcard, - PatKind::Leaf { .. } | PatKind::Deref { .. } => Single, - &PatKind::Variant { adt_def, variant_index, .. } => { - Variant(adt_def.variants[variant_index].def_id) - } - PatKind::Constant { value } => { - if let Some(int_range) = IntRange::from_const(cx.tcx, cx.param_env, value, pat.span) { - IntRange(int_range) - } else { - match pat.ty.kind() { - ty::Float(_) => FloatRange(value, value, RangeEnd::Included), - // In `expand_pattern`, we convert string literals to `&CONST` patterns with - // `CONST` a pattern of type `str`. In truth this contains a constant of type - // `&str`. - ty::Str => Str(value), - // All constants that can be structurally matched have already been expanded - // into the corresponding `Pat`s by `const_to_pat`. Constants that remain are - // opaque. - _ => Opaque, - } - } - } - &PatKind::Range(PatRange { lo, hi, end }) => { - let ty = lo.ty; - if let Some(int_range) = IntRange::from_range( - cx.tcx, - lo.eval_bits(cx.tcx, cx.param_env, lo.ty), - hi.eval_bits(cx.tcx, cx.param_env, hi.ty), - ty, - &end, - pat.span, - ) { - IntRange(int_range) - } else { - FloatRange(lo, hi, end) - } - } - PatKind::Array { prefix, slice, suffix } | PatKind::Slice { prefix, slice, suffix } => { - let array_len = match pat.ty.kind() { - ty::Array(_, length) => Some(length.eval_usize(cx.tcx, cx.param_env)), - ty::Slice(_) => None, - _ => span_bug!(pat.span, "bad ty {:?} for slice pattern", pat.ty), - }; - let prefix = prefix.len() as u64; - let suffix = suffix.len() as u64; - let kind = - if slice.is_some() { VarLen(prefix, suffix) } else { FixedLen(prefix + suffix) }; - Slice(Slice::new(array_len, kind)) - } - PatKind::Or { .. } => bug!("Or-pattern should have been expanded earlier on."), - } -} - -/// The arm of a match expression. -#[derive(Clone, Copy)] -crate struct MatchArm<'p, 'tcx> { - /// The pattern must have been lowered through `MatchVisitor::lower_pattern`. - crate pat: &'p super::Pat<'tcx>, - crate hir_id: HirId, - crate has_guard: bool, -} - -/// The output of checking a match for exhaustiveness and arm reachability. -crate struct UsefulnessReport<'p, 'tcx> { - /// For each arm of the input, whether that arm is reachable after the arms above it. - crate arm_usefulness: Vec<(MatchArm<'p, 'tcx>, Usefulness<'tcx>)>, - /// If the match is exhaustive, this is empty. If not, this contains witnesses for the lack of - /// exhaustiveness. - crate non_exhaustiveness_witnesses: Vec<super::Pat<'tcx>>, -} - -/// The entrypoint for the usefulness algorithm. Computes whether a match is exhaustive and which -/// of its arms are reachable. -/// -/// Note: the input patterns must have been lowered through `MatchVisitor::lower_pattern`. -crate fn compute_match_usefulness<'p, 'tcx>( - cx: &MatchCheckCtxt<'p, 'tcx>, - arms: &[MatchArm<'p, 'tcx>], - scrut_hir_id: HirId, - scrut_ty: Ty<'tcx>, -) -> UsefulnessReport<'p, 'tcx> { - let mut matrix = Matrix::empty(); - let arm_usefulness: Vec<_> = arms - .iter() - .copied() - .map(|arm| { - let v = PatStack::from_pattern(arm.pat); - let usefulness = - is_useful(cx, &matrix, &v, LeaveOutWitness, arm.hir_id, arm.has_guard, true); - if !arm.has_guard { - matrix.push(v); - } - (arm, usefulness) - }) - .collect(); - - let wild_pattern = cx.pattern_arena.alloc(super::Pat::wildcard_from_ty(scrut_ty)); - let v = PatStack::from_pattern(wild_pattern); - let usefulness = is_useful(cx, &matrix, &v, ConstructWitness, scrut_hir_id, false, true); - let non_exhaustiveness_witnesses = match usefulness { - NotUseful => vec![], // Wildcard pattern isn't useful, so the match is exhaustive. - UsefulWithWitness(pats) => { - if pats.is_empty() { - bug!("Exhaustiveness check returned no witnesses") - } else { - pats.into_iter().map(|w| w.single_pattern()).collect() - } - } - Useful(_) => bug!(), - }; - UsefulnessReport { arm_usefulness, non_exhaustiveness_witnesses } } diff --git a/compiler/rustc_mir_build/src/thir/pattern/mod.rs b/compiler/rustc_mir_build/src/thir/pattern/mod.rs index db0ecd701bc..7e9a3a37278 100644 --- a/compiler/rustc_mir_build/src/thir/pattern/mod.rs +++ b/compiler/rustc_mir_build/src/thir/pattern/mod.rs @@ -1,8 +1,9 @@ //! Validation of patterns/matches. -mod _match; mod check_match; mod const_to_pat; +mod deconstruct_pat; +mod usefulness; pub(crate) use self::check_match::check_match; diff --git a/compiler/rustc_mir_build/src/thir/pattern/usefulness.rs b/compiler/rustc_mir_build/src/thir/pattern/usefulness.rs new file mode 100644 index 00000000000..be96d5ae816 --- /dev/null +++ b/compiler/rustc_mir_build/src/thir/pattern/usefulness.rs @@ -0,0 +1,992 @@ +//! Note: tests specific to this file can be found in: +//! +//! - `ui/pattern/usefulness` +//! - `ui/or-patterns` +//! - `ui/consts/const_in_pattern` +//! - `ui/rfc-2008-non-exhaustive` +//! - `ui/half-open-range-patterns` +//! - probably many others +//! +//! I (Nadrieril) prefer to put new tests in `ui/pattern/usefulness` unless there's a specific +//! reason not to, for example if they depend on a particular feature like `or_patterns`. +//! +//! ----- +//! +//! This file includes the logic for exhaustiveness and usefulness checking for +//! pattern-matching. Specifically, given a list of patterns for a type, we can +//! tell whether: +//! (a) the patterns cover every possible constructor for the type (exhaustiveness) +//! (b) each pattern is necessary (usefulness) +//! +//! The algorithm implemented here is a modified version of the one described in +//! [this paper](http://moscova.inria.fr/~maranget/papers/warn/index.html). +//! However, to save future implementors from reading the original paper, we +//! summarise the algorithm here to hopefully save time and be a little clearer +//! (without being so rigorous). +//! +//! # Premise +//! +//! The core of the algorithm revolves about a "usefulness" check. In particular, we +//! are trying to compute a predicate `U(P, p)` where `P` is a list of patterns (we refer to this as +//! a matrix). `U(P, p)` represents whether, given an existing list of patterns +//! `P_1 ..= P_m`, adding a new pattern `p` will be "useful" (that is, cover previously- +//! uncovered values of the type). +//! +//! If we have this predicate, then we can easily compute both exhaustiveness of an +//! entire set of patterns and the individual usefulness of each one. +//! (a) the set of patterns is exhaustive iff `U(P, _)` is false (i.e., adding a wildcard +//! match doesn't increase the number of values we're matching) +//! (b) a pattern `P_i` is not useful if `U(P[0..=(i-1), P_i)` is false (i.e., adding a +//! pattern to those that have come before it doesn't increase the number of values +//! we're matching). +//! +//! # Core concept +//! +//! The idea that powers everything that is done in this file is the following: a value is made +//! from a constructor applied to some fields. Examples of constructors are `Some`, `None`, `(,)` +//! (the 2-tuple constructor), `Foo {..}` (the constructor for a struct `Foo`), and `2` (the +//! constructor for the number `2`). Fields are just a (possibly empty) list of values. +//! +//! Some of the constructors listed above might feel weird: `None` and `2` don't take any +//! arguments. This is part of what makes constructors so general: we will consider plain values +//! like numbers and string literals to be constructors that take no arguments, also called "0-ary +//! constructors"; they are the simplest case of constructors. This allows us to see any value as +//! made up from a tree of constructors, each having a given number of children. For example: +//! `(None, Ok(0))` is made from 4 different constructors. +//! +//! This idea can be extended to patterns: a pattern captures a set of possible values, and we can +//! describe this set using constructors. For example, `Err(_)` captures all values of the type +//! `Result<T, E>` that start with the `Err` constructor (for some choice of `T` and `E`). The +//! wildcard `_` captures all values of the given type starting with any of the constructors for +//! that type. +//! +//! We use this to compute whether different patterns might capture a same value. Do the patterns +//! `Ok("foo")` and `Err(_)` capture a common value? The answer is no, because the first pattern +//! captures only values starting with the `Ok` constructor and the second only values starting +//! with the `Err` constructor. Do the patterns `Some(42)` and `Some(1..10)` intersect? They might, +//! since they both capture values starting with `Some`. To be certain, we need to dig under the +//! `Some` constructor and continue asking the question. This is the main idea behind the +//! exhaustiveness algorithm: by looking at patterns constructor-by-constructor, we can efficiently +//! figure out if some new pattern might capture a value that hadn't been captured by previous +//! patterns. +//! +//! Constructors are represented by the `Constructor` enum, and its fields by the `Fields` enum. +//! Most of the complexity of this file resides in transforming between patterns and +//! (`Constructor`, `Fields`) pairs, handling all the special cases correctly. +//! +//! Caveat: this constructors/fields distinction doesn't quite cover every Rust value. For example +//! a value of type `Rc<u64>` doesn't fit this idea very well, nor do various other things. +//! However, this idea covers most of the cases that are relevant to exhaustiveness checking. +//! +//! +//! # Algorithm +//! +//! Recall that `U(P, p)` represents whether, given an existing list of patterns (aka matrix) `P`, +//! adding a new pattern `p` will cover previously-uncovered values of the type. +//! During the course of the algorithm, the rows of the matrix won't just be individual patterns, +//! but rather partially-deconstructed patterns in the form of a list of fields. The paper +//! calls those pattern-vectors, and we will call them pattern-stacks. The same holds for the +//! new pattern `p`. +//! +//! For example, say we have the following: +//! +//! ``` +//! // x: (Option<bool>, Result<()>) +//! match x { +//! (Some(true), _) => {} +//! (None, Err(())) => {} +//! (None, Err(_)) => {} +//! } +//! ``` +//! +//! Here, the matrix `P` starts as: +//! +//! ``` +//! [ +//! [(Some(true), _)], +//! [(None, Err(()))], +//! [(None, Err(_))], +//! ] +//! ``` +//! +//! We can tell it's not exhaustive, because `U(P, _)` is true (we're not covering +//! `[(Some(false), _)]`, for instance). In addition, row 3 is not useful, because +//! all the values it covers are already covered by row 2. +//! +//! A list of patterns can be thought of as a stack, because we are mainly interested in the top of +//! the stack at any given point, and we can pop or apply constructors to get new pattern-stacks. +//! To match the paper, the top of the stack is at the beginning / on the left. +//! +//! There are two important operations on pattern-stacks necessary to understand the algorithm: +//! +//! 1. We can pop a given constructor off the top of a stack. This operation is called +//! `specialize`, and is denoted `S(c, p)` where `c` is a constructor (like `Some` or +//! `None`) and `p` a pattern-stack. +//! If the pattern on top of the stack can cover `c`, this removes the constructor and +//! pushes its arguments onto the stack. It also expands OR-patterns into distinct patterns. +//! Otherwise the pattern-stack is discarded. +//! This essentially filters those pattern-stacks whose top covers the constructor `c` and +//! discards the others. +//! +//! For example, the first pattern above initially gives a stack `[(Some(true), _)]`. If we +//! pop the tuple constructor, we are left with `[Some(true), _]`, and if we then pop the +//! `Some` constructor we get `[true, _]`. If we had popped `None` instead, we would get +//! nothing back. +//! +//! This returns zero or more new pattern-stacks, as follows. We look at the pattern `p_1` +//! on top of the stack, and we have four cases: +//! +//! 1.1. `p_1 = c(r_1, .., r_a)`, i.e. the top of the stack has constructor `c`. We +//! push onto the stack the arguments of this constructor, and return the result: +//! `r_1, .., r_a, p_2, .., p_n` +//! +//! 1.2. `p_1 = c'(r_1, .., r_a')` where `c ≠ c'`. We discard the current stack and +//! return nothing. +//! +//! 1.3. `p_1 = _`. We push onto the stack as many wildcards as the constructor `c` has +//! arguments (its arity), and return the resulting stack: +//! `_, .., _, p_2, .., p_n` +//! +//! 1.4. `p_1 = r_1 | r_2`. We expand the OR-pattern and then recurse on each resulting +//! stack: +//! - `S(c, (r_1, p_2, .., p_n))` +//! - `S(c, (r_2, p_2, .., p_n))` +//! +//! 2. We can pop a wildcard off the top of the stack. This is called `S(_, p)`, where `p` is +//! a pattern-stack. Note: the paper calls this `D(p)`. +//! This is used when we know there are missing constructor cases, but there might be +//! existing wildcard patterns, so to check the usefulness of the matrix, we have to check +//! all its *other* components. +//! +//! It is computed as follows. We look at the pattern `p_1` on top of the stack, +//! and we have three cases: +//! 2.1. `p_1 = c(r_1, .., r_a)`. We discard the current stack and return nothing. +//! 2.2. `p_1 = _`. We return the rest of the stack: +//! p_2, .., p_n +//! 2.3. `p_1 = r_1 | r_2`. We expand the OR-pattern and then recurse on each resulting +//! stack. +//! - `S(_, (r_1, p_2, .., p_n))` +//! - `S(_, (r_2, p_2, .., p_n))` +//! +//! Note that the OR-patterns are not always used directly in Rust, but are used to derive the +//! exhaustive integer matching rules, so they're written here for posterity. +//! +//! Both those operations extend straightforwardly to a list or pattern-stacks, i.e. a matrix, by +//! working row-by-row. Popping a constructor ends up keeping only the matrix rows that start with +//! the given constructor, and popping a wildcard keeps those rows that start with a wildcard. +//! +//! +//! The algorithm for computing `U` +//! ------------------------------- +//! The algorithm is inductive (on the number of columns: i.e., components of tuple patterns). +//! That means we're going to check the components from left-to-right, so the algorithm +//! operates principally on the first component of the matrix and new pattern-stack `p`. +//! This algorithm is realised in the `is_useful` function. +//! +//! Base case. (`n = 0`, i.e., an empty tuple pattern) +//! - If `P` already contains an empty pattern (i.e., if the number of patterns `m > 0`), +//! then `U(P, p)` is false. +//! - Otherwise, `P` must be empty, so `U(P, p)` is true. +//! +//! Inductive step. (`n > 0`, i.e., whether there's at least one column +//! [which may then be expanded into further columns later]) +//! We're going to match on the top of the new pattern-stack, `p_1`. +//! - If `p_1 == c(r_1, .., r_a)`, i.e. we have a constructor pattern. +//! Then, the usefulness of `p_1` can be reduced to whether it is useful when +//! we ignore all the patterns in the first column of `P` that involve other constructors. +//! This is where `S(c, P)` comes in: +//! `U(P, p) := U(S(c, P), S(c, p))` +//! +//! For example, if `P` is: +//! +//! ``` +//! [ +//! [Some(true), _], +//! [None, 0], +//! ] +//! ``` +//! +//! and `p` is `[Some(false), 0]`, then we don't care about row 2 since we know `p` only +//! matches values that row 2 doesn't. For row 1 however, we need to dig into the +//! arguments of `Some` to know whether some new value is covered. So we compute +//! `U([[true, _]], [false, 0])`. +//! +//! - If `p_1 == _`, then we look at the list of constructors that appear in the first +//! component of the rows of `P`: +//! + If there are some constructors that aren't present, then we might think that the +//! wildcard `_` is useful, since it covers those constructors that weren't covered +//! before. +//! That's almost correct, but only works if there were no wildcards in those first +//! components. So we need to check that `p` is useful with respect to the rows that +//! start with a wildcard, if there are any. This is where `S(_, x)` comes in: +//! `U(P, p) := U(S(_, P), S(_, p))` +//! +//! For example, if `P` is: +//! +//! ``` +//! [ +//! [_, true, _], +//! [None, false, 1], +//! ] +//! ``` +//! +//! and `p` is `[_, false, _]`, the `Some` constructor doesn't appear in `P`. So if we +//! only had row 2, we'd know that `p` is useful. However row 1 starts with a +//! wildcard, so we need to check whether `U([[true, _]], [false, 1])`. +//! +//! + Otherwise, all possible constructors (for the relevant type) are present. In this +//! case we must check whether the wildcard pattern covers any unmatched value. For +//! that, we can think of the `_` pattern as a big OR-pattern that covers all +//! possible constructors. For `Option`, that would mean `_ = None | Some(_)` for +//! example. The wildcard pattern is useful in this case if it is useful when +//! specialized to one of the possible constructors. So we compute: +//! `U(P, p) := ∃(k ϵ constructors) U(S(k, P), S(k, p))` +//! +//! For example, if `P` is: +//! +//! ``` +//! [ +//! [Some(true), _], +//! [None, false], +//! ] +//! ``` +//! +//! and `p` is `[_, false]`, both `None` and `Some` constructors appear in the first +//! components of `P`. We will therefore try popping both constructors in turn: we +//! compute `U([[true, _]], [_, false])` for the `Some` constructor, and `U([[false]], +//! [false])` for the `None` constructor. The first case returns true, so we know that +//! `p` is useful for `P`. Indeed, it matches `[Some(false), _]` that wasn't matched +//! before. +//! +//! - If `p_1 == r_1 | r_2`, then the usefulness depends on each `r_i` separately: +//! `U(P, p) := U(P, (r_1, p_2, .., p_n)) +//! || U(P, (r_2, p_2, .., p_n))` +//! +//! Modifications to the algorithm +//! ------------------------------ +//! The algorithm in the paper doesn't cover some of the special cases that arise in Rust, for +//! example uninhabited types and variable-length slice patterns. These are drawn attention to +//! throughout the code below. I'll make a quick note here about how exhaustive integer matching is +//! accounted for, though. +//! +//! Exhaustive integer matching +//! --------------------------- +//! An integer type can be thought of as a (huge) sum type: 1 | 2 | 3 | ... +//! So to support exhaustive integer matching, we can make use of the logic in the paper for +//! OR-patterns. However, we obviously can't just treat ranges x..=y as individual sums, because +//! they are likely gigantic. So we instead treat ranges as constructors of the integers. This means +//! that we have a constructor *of* constructors (the integers themselves). We then need to work +//! through all the inductive step rules above, deriving how the ranges would be treated as +//! OR-patterns, and making sure that they're treated in the same way even when they're ranges. +//! There are really only four special cases here: +//! - When we match on a constructor that's actually a range, we have to treat it as if we would +//! an OR-pattern. +//! + It turns out that we can simply extend the case for single-value patterns in +//! `specialize` to either be *equal* to a value constructor, or *contained within* a range +//! constructor. +//! + When the pattern itself is a range, you just want to tell whether any of the values in +//! the pattern range coincide with values in the constructor range, which is precisely +//! intersection. +//! Since when encountering a range pattern for a value constructor, we also use inclusion, it +//! means that whenever the constructor is a value/range and the pattern is also a value/range, +//! we can simply use intersection to test usefulness. +//! - When we're testing for usefulness of a pattern and the pattern's first component is a +//! wildcard. +//! + If all the constructors appear in the matrix, we have a slight complication. By default, +//! the behaviour (i.e., a disjunction over specialised matrices for each constructor) is +//! invalid, because we want a disjunction over every *integer* in each range, not just a +//! disjunction over every range. This is a bit more tricky to deal with: essentially we need +//! to form equivalence classes of subranges of the constructor range for which the behaviour +//! of the matrix `P` and new pattern `p` are the same. This is described in more +//! detail in `Constructor::split`. +//! + If some constructors are missing from the matrix, it turns out we don't need to do +//! anything special (because we know none of the integers are actually wildcards: i.e., we +//! can't span wildcards using ranges). + +use self::Usefulness::*; +use self::WitnessPreference::*; + +use super::deconstruct_pat::{Constructor, Fields, MissingConstructors}; +use super::{Pat, PatKind}; +use super::{PatternFoldable, PatternFolder}; + +use rustc_data_structures::captures::Captures; +use rustc_data_structures::fx::FxHashSet; +use rustc_data_structures::sync::OnceCell; + +use rustc_arena::TypedArena; +use rustc_hir::def_id::DefId; +use rustc_hir::HirId; +use rustc_middle::ty::{self, Ty, TyCtxt}; +use rustc_span::Span; + +use smallvec::{smallvec, SmallVec}; +use std::fmt; +use std::iter::{FromIterator, IntoIterator}; + +crate struct MatchCheckCtxt<'a, 'tcx> { + crate tcx: TyCtxt<'tcx>, + /// The module in which the match occurs. This is necessary for + /// checking inhabited-ness of types because whether a type is (visibly) + /// inhabited can depend on whether it was defined in the current module or + /// not. E.g., `struct Foo { _private: ! }` cannot be seen to be empty + /// outside its module and should not be matchable with an empty match statement. + crate module: DefId, + crate param_env: ty::ParamEnv<'tcx>, + crate pattern_arena: &'a TypedArena<Pat<'tcx>>, +} + +impl<'a, 'tcx> MatchCheckCtxt<'a, 'tcx> { + pub(super) fn is_uninhabited(&self, ty: Ty<'tcx>) -> bool { + if self.tcx.features().exhaustive_patterns { + self.tcx.is_ty_uninhabited_from(self.module, ty, self.param_env) + } else { + false + } + } + + /// Returns whether the given type is an enum from another crate declared `#[non_exhaustive]`. + pub(super) fn is_foreign_non_exhaustive_enum(&self, ty: Ty<'tcx>) -> bool { + match ty.kind() { + ty::Adt(def, ..) => { + def.is_enum() && def.is_variant_list_non_exhaustive() && !def.did.is_local() + } + _ => false, + } + } +} + +#[derive(Copy, Clone)] +pub(super) struct PatCtxt<'a, 'p, 'tcx> { + pub(super) cx: &'a MatchCheckCtxt<'p, 'tcx>, + /// Current state of the matrix. + pub(super) matrix: &'a Matrix<'p, 'tcx>, + /// Type of the current column under investigation. + pub(super) ty: Ty<'tcx>, + /// Span of the current pattern under investigation. + pub(super) span: Span, + /// Whether the current pattern is the whole pattern as found in a match arm, or if it's a + /// subpattern. + pub(super) is_top_level: bool, +} + +crate fn expand_pattern<'tcx>(pat: Pat<'tcx>) -> Pat<'tcx> { + LiteralExpander.fold_pattern(&pat) +} + +struct LiteralExpander; + +impl<'tcx> PatternFolder<'tcx> for LiteralExpander { + fn fold_pattern(&mut self, pat: &Pat<'tcx>) -> Pat<'tcx> { + debug!("fold_pattern {:?} {:?} {:?}", pat, pat.ty.kind(), pat.kind); + match (pat.ty.kind(), pat.kind.as_ref()) { + (_, PatKind::Binding { subpattern: Some(s), .. }) => s.fold_with(self), + (_, PatKind::AscribeUserType { subpattern: s, .. }) => s.fold_with(self), + (ty::Ref(_, t, _), PatKind::Constant { .. }) if t.is_str() => { + // Treat string literal patterns as deref patterns to a `str` constant, i.e. + // `&CONST`. This expands them like other const patterns. This could have been done + // in `const_to_pat`, but that causes issues with the rest of the matching code. + let mut new_pat = pat.super_fold_with(self); + // Make a fake const pattern of type `str` (instead of `&str`). That the carried + // constant value still knows it is of type `&str`. + new_pat.ty = t; + Pat { + kind: Box::new(PatKind::Deref { subpattern: new_pat }), + span: pat.span, + ty: pat.ty, + } + } + _ => pat.super_fold_with(self), + } + } +} + +impl<'tcx> Pat<'tcx> { + pub(super) fn is_wildcard(&self) -> bool { + matches!(*self.kind, PatKind::Binding { subpattern: None, .. } | PatKind::Wild) + } +} + +/// A row of a matrix. Rows of len 1 are very common, which is why `SmallVec[_; 2]` +/// works well. +#[derive(Debug, Clone)] +struct PatStack<'p, 'tcx> { + pats: SmallVec<[&'p Pat<'tcx>; 2]>, + /// Cache for the constructor of the head + head_ctor: OnceCell<Constructor<'tcx>>, +} + +impl<'p, 'tcx> PatStack<'p, 'tcx> { + fn from_pattern(pat: &'p Pat<'tcx>) -> Self { + Self::from_vec(smallvec![pat]) + } + + fn from_vec(vec: SmallVec<[&'p Pat<'tcx>; 2]>) -> Self { + PatStack { pats: vec, head_ctor: OnceCell::new() } + } + + fn is_empty(&self) -> bool { + self.pats.is_empty() + } + + fn len(&self) -> usize { + self.pats.len() + } + + fn head(&self) -> &'p Pat<'tcx> { + self.pats[0] + } + + fn head_ctor<'a>(&'a self, cx: &MatchCheckCtxt<'p, 'tcx>) -> &'a Constructor<'tcx> { + self.head_ctor.get_or_init(|| Constructor::from_pat(cx, self.head())) + } + + fn iter(&self) -> impl Iterator<Item = &Pat<'tcx>> { + self.pats.iter().copied() + } + + // If the first pattern is an or-pattern, expand this pattern. Otherwise, return `None`. + fn expand_or_pat(&self) -> Option<Vec<Self>> { + if self.is_empty() { + None + } else if let PatKind::Or { pats } = &*self.head().kind { + Some( + pats.iter() + .map(|pat| { + let mut new_patstack = PatStack::from_pattern(pat); + new_patstack.pats.extend_from_slice(&self.pats[1..]); + new_patstack + }) + .collect(), + ) + } else { + None + } + } + + /// This computes `S(self.head_ctor(), self)`. See top of the file for explanations. + /// + /// Structure patterns with a partial wild pattern (Foo { a: 42, .. }) have their missing + /// fields filled with wild patterns. + /// + /// This is roughly the inverse of `Constructor::apply`. + fn pop_head_constructor(&self, ctor_wild_subpatterns: &Fields<'p, 'tcx>) -> PatStack<'p, 'tcx> { + // We pop the head pattern and push the new fields extracted from the arguments of + // `self.head()`. + let mut new_fields = + ctor_wild_subpatterns.replace_with_pattern_arguments(self.head()).filtered_patterns(); + new_fields.extend_from_slice(&self.pats[1..]); + PatStack::from_vec(new_fields) + } +} + +impl<'p, 'tcx> Default for PatStack<'p, 'tcx> { + fn default() -> Self { + Self::from_vec(smallvec![]) + } +} + +impl<'p, 'tcx> PartialEq for PatStack<'p, 'tcx> { + fn eq(&self, other: &Self) -> bool { + self.pats == other.pats + } +} + +impl<'p, 'tcx> FromIterator<&'p Pat<'tcx>> for PatStack<'p, 'tcx> { + fn from_iter<T>(iter: T) -> Self + where + T: IntoIterator<Item = &'p Pat<'tcx>>, + { + Self::from_vec(iter.into_iter().collect()) + } +} + +/// A 2D matrix. +#[derive(Clone, PartialEq)] +pub(super) struct Matrix<'p, 'tcx> { + patterns: Vec<PatStack<'p, 'tcx>>, +} + +impl<'p, 'tcx> Matrix<'p, 'tcx> { + fn empty() -> Self { + Matrix { patterns: vec![] } + } + + /// Number of columns of this matrix. `None` is the matrix is empty. + pub(super) fn column_count(&self) -> Option<usize> { + self.patterns.get(0).map(|r| r.len()) + } + + /// Pushes a new row to the matrix. If the row starts with an or-pattern, this expands it. + fn push(&mut self, row: PatStack<'p, 'tcx>) { + if let Some(rows) = row.expand_or_pat() { + for row in rows { + // We recursively expand the or-patterns of the new rows. + // This is necessary as we might have `0 | (1 | 2)` or e.g., `x @ 0 | x @ (1 | 2)`. + self.push(row) + } + } else { + self.patterns.push(row); + } + } + + /// Iterate over the first component of each row + fn heads<'a>(&'a self) -> impl Iterator<Item = &'a Pat<'tcx>> + Captures<'p> { + self.patterns.iter().map(|r| r.head()) + } + + /// Iterate over the first constructor of each row + pub(super) fn head_ctors<'a>( + &'a self, + cx: &'a MatchCheckCtxt<'p, 'tcx>, + ) -> impl Iterator<Item = &'a Constructor<'tcx>> + Captures<'a> + Captures<'p> { + self.patterns.iter().map(move |r| r.head_ctor(cx)) + } + + /// This computes `S(constructor, self)`. See top of the file for explanations. + fn specialize_constructor( + &self, + pcx: PatCtxt<'_, 'p, 'tcx>, + ctor: &Constructor<'tcx>, + ctor_wild_subpatterns: &Fields<'p, 'tcx>, + ) -> Matrix<'p, 'tcx> { + self.patterns + .iter() + .filter(|r| ctor.is_covered_by(pcx, r.head_ctor(pcx.cx))) + .map(|r| r.pop_head_constructor(ctor_wild_subpatterns)) + .collect() + } +} + +/// Pretty-printer for matrices of patterns, example: +/// +/// ```text +/// +++++++++++++++++++++++++++++ +/// + _ + [] + +/// +++++++++++++++++++++++++++++ +/// + true + [First] + +/// +++++++++++++++++++++++++++++ +/// + true + [Second(true)] + +/// +++++++++++++++++++++++++++++ +/// + false + [_] + +/// +++++++++++++++++++++++++++++ +/// + _ + [_, _, tail @ ..] + +/// +++++++++++++++++++++++++++++ +/// ``` +impl<'p, 'tcx> fmt::Debug for Matrix<'p, 'tcx> { + fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result { + write!(f, "\n")?; + + let Matrix { patterns: m, .. } = self; + let pretty_printed_matrix: Vec<Vec<String>> = + m.iter().map(|row| row.iter().map(|pat| format!("{:?}", pat)).collect()).collect(); + + let column_count = m.iter().map(|row| row.len()).max().unwrap_or(0); + assert!(m.iter().all(|row| row.len() == column_count)); + let column_widths: Vec<usize> = (0..column_count) + .map(|col| pretty_printed_matrix.iter().map(|row| row[col].len()).max().unwrap_or(0)) + .collect(); + + let total_width = column_widths.iter().cloned().sum::<usize>() + column_count * 3 + 1; + let br = "+".repeat(total_width); + write!(f, "{}\n", br)?; + for row in pretty_printed_matrix { + write!(f, "+")?; + for (column, pat_str) in row.into_iter().enumerate() { + write!(f, " ")?; + write!(f, "{:1$}", pat_str, column_widths[column])?; + write!(f, " +")?; + } + write!(f, "\n")?; + write!(f, "{}\n", br)?; + } + Ok(()) + } +} + +impl<'p, 'tcx> FromIterator<PatStack<'p, 'tcx>> for Matrix<'p, 'tcx> { + fn from_iter<T>(iter: T) -> Self + where + T: IntoIterator<Item = PatStack<'p, 'tcx>>, + { + let mut matrix = Matrix::empty(); + for x in iter { + // Using `push` ensures we correctly expand or-patterns. + matrix.push(x); + } + matrix + } +} + +#[derive(Clone, Debug)] +crate enum Usefulness<'tcx> { + /// Carries, for each column in the matrix, a set of sub-branches that have been found to be + /// unreachable. Used only in the presence of or-patterns, otherwise it stays empty. + Useful(Vec<FxHashSet<Span>>), + /// Carries a list of witnesses of non-exhaustiveness. + UsefulWithWitness(Vec<Witness<'tcx>>), + NotUseful, +} + +impl<'tcx> Usefulness<'tcx> { + fn new_useful(preference: WitnessPreference) -> Self { + match preference { + ConstructWitness => UsefulWithWitness(vec![Witness(vec![])]), + LeaveOutWitness => Useful(vec![]), + } + } + + fn is_useful(&self) -> bool { + !matches!(*self, NotUseful) + } + + fn apply_constructor<'p>( + self, + pcx: PatCtxt<'_, 'p, 'tcx>, + ctor: &Constructor<'tcx>, + ctor_wild_subpatterns: &Fields<'p, 'tcx>, + ) -> Self { + match self { + UsefulWithWitness(witnesses) => { + let new_witnesses = if ctor.is_wildcard() { + let missing_ctors = MissingConstructors::new(pcx); + let new_patterns = missing_ctors.report_patterns(pcx); + witnesses + .into_iter() + .flat_map(|witness| { + new_patterns.iter().map(move |pat| { + let mut witness = witness.clone(); + witness.0.push(pat.clone()); + witness + }) + }) + .collect() + } else { + witnesses + .into_iter() + .map(|witness| witness.apply_constructor(pcx, &ctor, ctor_wild_subpatterns)) + .collect() + }; + UsefulWithWitness(new_witnesses) + } + Useful(mut unreachables) => { + if !unreachables.is_empty() { + // When we apply a constructor, there are `arity` columns of the matrix that + // corresponded to its arguments. All the unreachables found in these columns + // will, after `apply`, come from the first column. So we take the union of all + // the corresponding sets and put them in the first column. + // Note that `arity` may be 0, in which case we just push a new empty set. + let len = unreachables.len(); + let arity = ctor_wild_subpatterns.len(); + let mut unioned = FxHashSet::default(); + for set in unreachables.drain((len - arity)..) { + unioned.extend(set) + } + unreachables.push(unioned); + } + Useful(unreachables) + } + x => x, + } + } +} + +#[derive(Copy, Clone, Debug)] +enum WitnessPreference { + ConstructWitness, + LeaveOutWitness, +} + +/// A witness of non-exhaustiveness for error reporting, represented +/// as a list of patterns (in reverse order of construction) with +/// wildcards inside to represent elements that can take any inhabitant +/// of the type as a value. +/// +/// A witness against a list of patterns should have the same types +/// and length as the pattern matched against. Because Rust `match` +/// is always against a single pattern, at the end the witness will +/// have length 1, but in the middle of the algorithm, it can contain +/// multiple patterns. +/// +/// For example, if we are constructing a witness for the match against +/// +/// ``` +/// struct Pair(Option<(u32, u32)>, bool); +/// +/// match (p: Pair) { +/// Pair(None, _) => {} +/// Pair(_, false) => {} +/// } +/// ``` +/// +/// We'll perform the following steps: +/// 1. Start with an empty witness +/// `Witness(vec![])` +/// 2. Push a witness `Some(_)` against the `None` +/// `Witness(vec![Some(_)])` +/// 3. Push a witness `true` against the `false` +/// `Witness(vec![Some(_), true])` +/// 4. Apply the `Pair` constructor to the witnesses +/// `Witness(vec![Pair(Some(_), true)])` +/// +/// The final `Pair(Some(_), true)` is then the resulting witness. +#[derive(Clone, Debug)] +crate struct Witness<'tcx>(Vec<Pat<'tcx>>); + +impl<'tcx> Witness<'tcx> { + /// Asserts that the witness contains a single pattern, and returns it. + fn single_pattern(self) -> Pat<'tcx> { + assert_eq!(self.0.len(), 1); + self.0.into_iter().next().unwrap() + } + + /// Constructs a partial witness for a pattern given a list of + /// patterns expanded by the specialization step. + /// + /// When a pattern P is discovered to be useful, this function is used bottom-up + /// to reconstruct a complete witness, e.g., a pattern P' that covers a subset + /// of values, V, where each value in that set is not covered by any previously + /// used patterns and is covered by the pattern P'. Examples: + /// + /// left_ty: tuple of 3 elements + /// pats: [10, 20, _] => (10, 20, _) + /// + /// left_ty: struct X { a: (bool, &'static str), b: usize} + /// pats: [(false, "foo"), 42] => X { a: (false, "foo"), b: 42 } + fn apply_constructor<'p>( + mut self, + pcx: PatCtxt<'_, 'p, 'tcx>, + ctor: &Constructor<'tcx>, + ctor_wild_subpatterns: &Fields<'p, 'tcx>, + ) -> Self { + let pat = { + let len = self.0.len(); + let arity = ctor_wild_subpatterns.len(); + let pats = self.0.drain((len - arity)..).rev(); + ctor_wild_subpatterns.replace_fields(pcx.cx, pats).apply(pcx, ctor) + }; + + self.0.push(pat); + + self + } +} + +/// Algorithm from <http://moscova.inria.fr/~maranget/papers/warn/index.html>. +/// The algorithm from the paper has been modified to correctly handle empty +/// types. The changes are: +/// (0) We don't exit early if the pattern matrix has zero rows. We just +/// continue to recurse over columns. +/// (1) all_constructors will only return constructors that are statically +/// possible. E.g., it will only return `Ok` for `Result<T, !>`. +/// +/// This finds whether a (row) vector `v` of patterns is 'useful' in relation +/// to a set of such vectors `m` - this is defined as there being a set of +/// inputs that will match `v` but not any of the sets in `m`. +/// +/// All the patterns at each column of the `matrix ++ v` matrix must have the same type. +/// +/// This is used both for reachability checking (if a pattern isn't useful in +/// relation to preceding patterns, it is not reachable) and exhaustiveness +/// checking (if a wildcard pattern is useful in relation to a matrix, the +/// matrix isn't exhaustive). +/// +/// `is_under_guard` is used to inform if the pattern has a guard. If it +/// has one it must not be inserted into the matrix. This shouldn't be +/// relied on for soundness. +fn is_useful<'p, 'tcx>( + cx: &MatchCheckCtxt<'p, 'tcx>, + matrix: &Matrix<'p, 'tcx>, + v: &PatStack<'p, 'tcx>, + witness_preference: WitnessPreference, + hir_id: HirId, + is_under_guard: bool, + is_top_level: bool, +) -> Usefulness<'tcx> { + let Matrix { patterns: rows, .. } = matrix; + debug!("is_useful({:#?}, {:#?})", matrix, v); + + // The base case. We are pattern-matching on () and the return value is + // based on whether our matrix has a row or not. + // NOTE: This could potentially be optimized by checking rows.is_empty() + // first and then, if v is non-empty, the return value is based on whether + // the type of the tuple we're checking is inhabited or not. + if v.is_empty() { + return if rows.is_empty() { + Usefulness::new_useful(witness_preference) + } else { + NotUseful + }; + }; + + assert!(rows.iter().all(|r| r.len() == v.len())); + + // If the first pattern is an or-pattern, expand it. + if let Some(vs) = v.expand_or_pat() { + // We expand the or pattern, trying each of its branches in turn and keeping careful track + // of possible unreachable sub-branches. + // + // If two branches have detected some unreachable sub-branches, we need to be careful. If + // they were detected in columns that are not the current one, we want to keep only the + // sub-branches that were unreachable in _all_ branches. Eg. in the following, the last + // `true` is unreachable in the second branch of the first or-pattern, but not otherwise. + // Therefore we don't want to lint that it is unreachable. + // + // ``` + // match (true, true) { + // (true, true) => {} + // (false | true, false | true) => {} + // } + // ``` + // If however the sub-branches come from the current column, they come from the inside of + // the current or-pattern, and we want to keep them all. Eg. in the following, we _do_ want + // to lint that the last `false` is unreachable. + // ``` + // match None { + // Some(false) => {} + // None | Some(true | false) => {} + // } + // ``` + + let mut matrix = matrix.clone(); + // We keep track of sub-branches separately depending on whether they come from this column + // or from others. + let mut unreachables_this_column: FxHashSet<Span> = FxHashSet::default(); + let mut unreachables_other_columns: Vec<FxHashSet<Span>> = Vec::default(); + // Whether at least one branch is reachable. + let mut any_is_useful = false; + + for v in vs { + let res = is_useful(cx, &matrix, &v, witness_preference, hir_id, is_under_guard, false); + match res { + Useful(unreachables) => { + if let Some((this_column, other_columns)) = unreachables.split_last() { + // We keep the union of unreachables found in the first column. + unreachables_this_column.extend(this_column); + // We keep the intersection of unreachables found in other columns. + if unreachables_other_columns.is_empty() { + unreachables_other_columns = other_columns.to_vec(); + } else { + unreachables_other_columns = unreachables_other_columns + .into_iter() + .zip(other_columns) + .map(|(x, y)| x.intersection(&y).copied().collect()) + .collect(); + } + } + any_is_useful = true; + } + NotUseful => { + unreachables_this_column.insert(v.head().span); + } + UsefulWithWitness(_) => bug!( + "encountered or-pat in the expansion of `_` during exhaustiveness checking" + ), + } + + // If pattern has a guard don't add it to the matrix. + if !is_under_guard { + // We push the already-seen patterns into the matrix in order to detect redundant + // branches like `Some(_) | Some(0)`. + matrix.push(v); + } + } + + return if any_is_useful { + let mut unreachables = if unreachables_other_columns.is_empty() { + let n_columns = v.len(); + (0..n_columns - 1).map(|_| FxHashSet::default()).collect() + } else { + unreachables_other_columns + }; + unreachables.push(unreachables_this_column); + Useful(unreachables) + } else { + NotUseful + }; + } + + // FIXME(Nadrieril): Hack to work around type normalization issues (see #72476). + let ty = matrix.heads().next().map(|r| r.ty).unwrap_or(v.head().ty); + let pcx = PatCtxt { cx, matrix, ty, span: v.head().span, is_top_level }; + + debug!("is_useful_expand_first_col: ty={:#?}, expanding {:#?}", pcx.ty, v.head()); + + let ret = v + .head_ctor(cx) + .split(pcx, Some(hir_id)) + .into_iter() + .map(|ctor| { + // We cache the result of `Fields::wildcards` because it is used a lot. + let ctor_wild_subpatterns = Fields::wildcards(pcx, &ctor); + let matrix = pcx.matrix.specialize_constructor(pcx, &ctor, &ctor_wild_subpatterns); + let v = v.pop_head_constructor(&ctor_wild_subpatterns); + let usefulness = + is_useful(pcx.cx, &matrix, &v, witness_preference, hir_id, is_under_guard, false); + usefulness.apply_constructor(pcx, &ctor, &ctor_wild_subpatterns) + }) + .find(|result| result.is_useful()) + .unwrap_or(NotUseful); + debug!("is_useful::returns({:#?}, {:#?}) = {:?}", matrix, v, ret); + ret +} + +/// The arm of a match expression. +#[derive(Clone, Copy)] +crate struct MatchArm<'p, 'tcx> { + /// The pattern must have been lowered through `MatchVisitor::lower_pattern`. + crate pat: &'p super::Pat<'tcx>, + crate hir_id: HirId, + crate has_guard: bool, +} + +/// The output of checking a match for exhaustiveness and arm reachability. +crate struct UsefulnessReport<'p, 'tcx> { + /// For each arm of the input, whether that arm is reachable after the arms above it. + crate arm_usefulness: Vec<(MatchArm<'p, 'tcx>, Usefulness<'tcx>)>, + /// If the match is exhaustive, this is empty. If not, this contains witnesses for the lack of + /// exhaustiveness. + crate non_exhaustiveness_witnesses: Vec<super::Pat<'tcx>>, +} + +/// The entrypoint for the usefulness algorithm. Computes whether a match is exhaustive and which +/// of its arms are reachable. +/// +/// Note: the input patterns must have been lowered through `MatchVisitor::lower_pattern`. +crate fn compute_match_usefulness<'p, 'tcx>( + cx: &MatchCheckCtxt<'p, 'tcx>, + arms: &[MatchArm<'p, 'tcx>], + scrut_hir_id: HirId, + scrut_ty: Ty<'tcx>, +) -> UsefulnessReport<'p, 'tcx> { + let mut matrix = Matrix::empty(); + let arm_usefulness: Vec<_> = arms + .iter() + .copied() + .map(|arm| { + let v = PatStack::from_pattern(arm.pat); + let usefulness = + is_useful(cx, &matrix, &v, LeaveOutWitness, arm.hir_id, arm.has_guard, true); + if !arm.has_guard { + matrix.push(v); + } + (arm, usefulness) + }) + .collect(); + + let wild_pattern = cx.pattern_arena.alloc(super::Pat::wildcard_from_ty(scrut_ty)); + let v = PatStack::from_pattern(wild_pattern); + let usefulness = is_useful(cx, &matrix, &v, ConstructWitness, scrut_hir_id, false, true); + let non_exhaustiveness_witnesses = match usefulness { + NotUseful => vec![], // Wildcard pattern isn't useful, so the match is exhaustive. + UsefulWithWitness(pats) => { + if pats.is_empty() { + bug!("Exhaustiveness check returned no witnesses") + } else { + pats.into_iter().map(|w| w.single_pattern()).collect() + } + } + Useful(_) => bug!(), + }; + UsefulnessReport { arm_usefulness, non_exhaustiveness_witnesses } +} |