mirror of
https://github.com/rust-lang/rust-analyzer
synced 2025-01-20 09:03:56 +00:00
336a3c6121
Basically, we need to allow variables in the caller self type to unify with the impl's declared self type. That requires some more contortions in the variable handling. I'm looking forward to (hopefully) handling this in a cleaner way when we switch to Chalk's types and unification code.
410 lines
15 KiB
Rust
410 lines
15 KiB
Rust
//! Unification and canonicalization logic.
|
|
|
|
use std::borrow::Cow;
|
|
|
|
use ena::unify::{InPlaceUnificationTable, NoError, UnifyKey, UnifyValue};
|
|
|
|
use test_utils::tested_by;
|
|
|
|
use super::{InferenceContext, Obligation};
|
|
use crate::{db::HirDatabase, Canonical, InEnvironment, InferTy, Substs, Ty, TypeCtor, TypeWalk};
|
|
|
|
impl<'a, D: HirDatabase> InferenceContext<'a, D> {
|
|
pub(super) fn canonicalizer<'b>(&'b mut self) -> Canonicalizer<'a, 'b, D>
|
|
where
|
|
'a: 'b,
|
|
{
|
|
Canonicalizer { ctx: self, free_vars: Vec::new(), var_stack: Vec::new() }
|
|
}
|
|
}
|
|
|
|
pub(super) struct Canonicalizer<'a, 'b, D: HirDatabase>
|
|
where
|
|
'a: 'b,
|
|
{
|
|
ctx: &'b mut InferenceContext<'a, D>,
|
|
free_vars: Vec<InferTy>,
|
|
/// A stack of type variables that is used to detect recursive types (which
|
|
/// are an error, but we need to protect against them to avoid stack
|
|
/// overflows).
|
|
var_stack: Vec<TypeVarId>,
|
|
}
|
|
|
|
pub(super) struct Canonicalized<T> {
|
|
pub value: Canonical<T>,
|
|
free_vars: Vec<InferTy>,
|
|
}
|
|
|
|
impl<'a, 'b, D: HirDatabase> Canonicalizer<'a, 'b, D>
|
|
where
|
|
'a: 'b,
|
|
{
|
|
fn add(&mut self, free_var: InferTy) -> usize {
|
|
self.free_vars.iter().position(|&v| v == free_var).unwrap_or_else(|| {
|
|
let next_index = self.free_vars.len();
|
|
self.free_vars.push(free_var);
|
|
next_index
|
|
})
|
|
}
|
|
|
|
fn do_canonicalize<T: TypeWalk>(&mut self, t: T, binders: usize) -> T {
|
|
t.fold_binders(
|
|
&mut |ty, binders| match ty {
|
|
Ty::Infer(tv) => {
|
|
let inner = tv.to_inner();
|
|
if self.var_stack.contains(&inner) {
|
|
// recursive type
|
|
return tv.fallback_value();
|
|
}
|
|
if let Some(known_ty) =
|
|
self.ctx.table.var_unification_table.inlined_probe_value(inner).known()
|
|
{
|
|
self.var_stack.push(inner);
|
|
let result = self.do_canonicalize(known_ty.clone(), binders);
|
|
self.var_stack.pop();
|
|
result
|
|
} else {
|
|
let root = self.ctx.table.var_unification_table.find(inner);
|
|
let free_var = match tv {
|
|
InferTy::TypeVar(_) => InferTy::TypeVar(root),
|
|
InferTy::IntVar(_) => InferTy::IntVar(root),
|
|
InferTy::FloatVar(_) => InferTy::FloatVar(root),
|
|
InferTy::MaybeNeverTypeVar(_) => InferTy::MaybeNeverTypeVar(root),
|
|
};
|
|
let position = self.add(free_var);
|
|
Ty::Bound((position + binders) as u32)
|
|
}
|
|
}
|
|
_ => ty,
|
|
},
|
|
binders,
|
|
)
|
|
}
|
|
|
|
fn into_canonicalized<T>(self, result: T) -> Canonicalized<T> {
|
|
Canonicalized {
|
|
value: Canonical { value: result, num_vars: self.free_vars.len() },
|
|
free_vars: self.free_vars,
|
|
}
|
|
}
|
|
|
|
pub(crate) fn canonicalize_ty(mut self, ty: Ty) -> Canonicalized<Ty> {
|
|
let result = self.do_canonicalize(ty, 0);
|
|
self.into_canonicalized(result)
|
|
}
|
|
|
|
pub(crate) fn canonicalize_obligation(
|
|
mut self,
|
|
obligation: InEnvironment<Obligation>,
|
|
) -> Canonicalized<InEnvironment<Obligation>> {
|
|
let result = match obligation.value {
|
|
Obligation::Trait(tr) => Obligation::Trait(self.do_canonicalize(tr, 0)),
|
|
Obligation::Projection(pr) => Obligation::Projection(self.do_canonicalize(pr, 0)),
|
|
};
|
|
self.into_canonicalized(InEnvironment {
|
|
value: result,
|
|
environment: obligation.environment,
|
|
})
|
|
}
|
|
}
|
|
|
|
impl<T> Canonicalized<T> {
|
|
pub fn decanonicalize_ty(&self, mut ty: Ty) -> Ty {
|
|
ty.walk_mut_binders(
|
|
&mut |ty, binders| {
|
|
if let &mut Ty::Bound(idx) = ty {
|
|
if idx as usize >= binders && (idx as usize - binders) < self.free_vars.len() {
|
|
*ty = Ty::Infer(self.free_vars[idx as usize - binders]);
|
|
}
|
|
}
|
|
},
|
|
0,
|
|
);
|
|
ty
|
|
}
|
|
|
|
pub fn apply_solution(
|
|
&self,
|
|
ctx: &mut InferenceContext<'_, impl HirDatabase>,
|
|
solution: Canonical<Vec<Ty>>,
|
|
) {
|
|
// the solution may contain new variables, which we need to convert to new inference vars
|
|
let new_vars = Substs((0..solution.num_vars).map(|_| ctx.table.new_type_var()).collect());
|
|
for (i, ty) in solution.value.into_iter().enumerate() {
|
|
let var = self.free_vars[i];
|
|
// eagerly replace projections in the type; we may be getting types
|
|
// e.g. from where clauses where this hasn't happened yet
|
|
let ty = ctx.normalize_associated_types_in(ty.subst_bound_vars(&new_vars));
|
|
ctx.table.unify(&Ty::Infer(var), &ty);
|
|
}
|
|
}
|
|
}
|
|
|
|
pub fn unify(ty1: &Canonical<Ty>, ty2: &Canonical<Ty>) -> Option<Substs> {
|
|
let mut table = InferenceTable::new();
|
|
let num_vars = ty1.num_vars.max(ty2.num_vars);
|
|
let vars =
|
|
Substs::builder(num_vars).fill(std::iter::repeat_with(|| table.new_type_var())).build();
|
|
let ty1_with_vars = ty1.value.clone().subst_bound_vars(&vars);
|
|
let ty2_with_vars = ty2.value.clone().subst_bound_vars(&vars);
|
|
if !table.unify(&ty1_with_vars, &ty2_with_vars) {
|
|
return None;
|
|
}
|
|
// default any type vars that weren't unified back to their original bound vars
|
|
// (kind of hacky)
|
|
for (i, var) in vars.iter().enumerate() {
|
|
if &*table.resolve_ty_shallow(var) == var {
|
|
table.unify(var, &Ty::Bound(i as u32));
|
|
}
|
|
}
|
|
Some(
|
|
Substs::builder(ty1.num_vars)
|
|
.fill(vars.iter().map(|v| table.resolve_ty_completely(v.clone())))
|
|
.build(),
|
|
)
|
|
}
|
|
|
|
#[derive(Clone, Debug)]
|
|
pub(crate) struct InferenceTable {
|
|
pub(super) var_unification_table: InPlaceUnificationTable<TypeVarId>,
|
|
}
|
|
|
|
impl InferenceTable {
|
|
pub fn new() -> Self {
|
|
InferenceTable { var_unification_table: InPlaceUnificationTable::new() }
|
|
}
|
|
|
|
pub fn new_type_var(&mut self) -> Ty {
|
|
Ty::Infer(InferTy::TypeVar(self.var_unification_table.new_key(TypeVarValue::Unknown)))
|
|
}
|
|
|
|
pub fn new_integer_var(&mut self) -> Ty {
|
|
Ty::Infer(InferTy::IntVar(self.var_unification_table.new_key(TypeVarValue::Unknown)))
|
|
}
|
|
|
|
pub fn new_float_var(&mut self) -> Ty {
|
|
Ty::Infer(InferTy::FloatVar(self.var_unification_table.new_key(TypeVarValue::Unknown)))
|
|
}
|
|
|
|
pub fn new_maybe_never_type_var(&mut self) -> Ty {
|
|
Ty::Infer(InferTy::MaybeNeverTypeVar(
|
|
self.var_unification_table.new_key(TypeVarValue::Unknown),
|
|
))
|
|
}
|
|
|
|
pub fn resolve_ty_completely(&mut self, ty: Ty) -> Ty {
|
|
self.resolve_ty_completely_inner(&mut Vec::new(), ty)
|
|
}
|
|
|
|
pub fn resolve_ty_as_possible(&mut self, ty: Ty) -> Ty {
|
|
self.resolve_ty_as_possible_inner(&mut Vec::new(), ty)
|
|
}
|
|
|
|
pub fn unify(&mut self, ty1: &Ty, ty2: &Ty) -> bool {
|
|
self.unify_inner(ty1, ty2, 0)
|
|
}
|
|
|
|
pub fn unify_substs(&mut self, substs1: &Substs, substs2: &Substs, depth: usize) -> bool {
|
|
substs1.0.iter().zip(substs2.0.iter()).all(|(t1, t2)| self.unify_inner(t1, t2, depth))
|
|
}
|
|
|
|
fn unify_inner(&mut self, ty1: &Ty, ty2: &Ty, depth: usize) -> bool {
|
|
if depth > 1000 {
|
|
// prevent stackoverflows
|
|
panic!("infinite recursion in unification");
|
|
}
|
|
if ty1 == ty2 {
|
|
return true;
|
|
}
|
|
// try to resolve type vars first
|
|
let ty1 = self.resolve_ty_shallow(ty1);
|
|
let ty2 = self.resolve_ty_shallow(ty2);
|
|
match (&*ty1, &*ty2) {
|
|
(Ty::Apply(a_ty1), Ty::Apply(a_ty2)) if a_ty1.ctor == a_ty2.ctor => {
|
|
self.unify_substs(&a_ty1.parameters, &a_ty2.parameters, depth + 1)
|
|
}
|
|
_ => self.unify_inner_trivial(&ty1, &ty2),
|
|
}
|
|
}
|
|
|
|
pub(super) fn unify_inner_trivial(&mut self, ty1: &Ty, ty2: &Ty) -> bool {
|
|
match (ty1, ty2) {
|
|
(Ty::Unknown, _) | (_, Ty::Unknown) => true,
|
|
|
|
(Ty::Placeholder(p1), Ty::Placeholder(p2)) if *p1 == *p2 => true,
|
|
|
|
(Ty::Infer(InferTy::TypeVar(tv1)), Ty::Infer(InferTy::TypeVar(tv2)))
|
|
| (Ty::Infer(InferTy::IntVar(tv1)), Ty::Infer(InferTy::IntVar(tv2)))
|
|
| (Ty::Infer(InferTy::FloatVar(tv1)), Ty::Infer(InferTy::FloatVar(tv2)))
|
|
| (
|
|
Ty::Infer(InferTy::MaybeNeverTypeVar(tv1)),
|
|
Ty::Infer(InferTy::MaybeNeverTypeVar(tv2)),
|
|
) => {
|
|
// both type vars are unknown since we tried to resolve them
|
|
self.var_unification_table.union(*tv1, *tv2);
|
|
true
|
|
}
|
|
|
|
// The order of MaybeNeverTypeVar matters here.
|
|
// Unifying MaybeNeverTypeVar and TypeVar will let the latter become MaybeNeverTypeVar.
|
|
// Unifying MaybeNeverTypeVar and other concrete type will let the former become it.
|
|
(Ty::Infer(InferTy::TypeVar(tv)), other)
|
|
| (other, Ty::Infer(InferTy::TypeVar(tv)))
|
|
| (Ty::Infer(InferTy::MaybeNeverTypeVar(tv)), other)
|
|
| (other, Ty::Infer(InferTy::MaybeNeverTypeVar(tv)))
|
|
| (Ty::Infer(InferTy::IntVar(tv)), other @ ty_app!(TypeCtor::Int(_)))
|
|
| (other @ ty_app!(TypeCtor::Int(_)), Ty::Infer(InferTy::IntVar(tv)))
|
|
| (Ty::Infer(InferTy::FloatVar(tv)), other @ ty_app!(TypeCtor::Float(_)))
|
|
| (other @ ty_app!(TypeCtor::Float(_)), Ty::Infer(InferTy::FloatVar(tv))) => {
|
|
// the type var is unknown since we tried to resolve it
|
|
self.var_unification_table.union_value(*tv, TypeVarValue::Known(other.clone()));
|
|
true
|
|
}
|
|
|
|
_ => false,
|
|
}
|
|
}
|
|
|
|
/// If `ty` is a type variable with known type, returns that type;
|
|
/// otherwise, return ty.
|
|
pub fn resolve_ty_shallow<'b>(&mut self, ty: &'b Ty) -> Cow<'b, Ty> {
|
|
let mut ty = Cow::Borrowed(ty);
|
|
// The type variable could resolve to a int/float variable. Hence try
|
|
// resolving up to three times; each type of variable shouldn't occur
|
|
// more than once
|
|
for i in 0..3 {
|
|
if i > 0 {
|
|
tested_by!(type_var_resolves_to_int_var);
|
|
}
|
|
match &*ty {
|
|
Ty::Infer(tv) => {
|
|
let inner = tv.to_inner();
|
|
match self.var_unification_table.inlined_probe_value(inner).known() {
|
|
Some(known_ty) => {
|
|
// The known_ty can't be a type var itself
|
|
ty = Cow::Owned(known_ty.clone());
|
|
}
|
|
_ => return ty,
|
|
}
|
|
}
|
|
_ => return ty,
|
|
}
|
|
}
|
|
log::error!("Inference variable still not resolved: {:?}", ty);
|
|
ty
|
|
}
|
|
|
|
/// Resolves the type as far as currently possible, replacing type variables
|
|
/// by their known types. All types returned by the infer_* functions should
|
|
/// be resolved as far as possible, i.e. contain no type variables with
|
|
/// known type.
|
|
fn resolve_ty_as_possible_inner(&mut self, tv_stack: &mut Vec<TypeVarId>, ty: Ty) -> Ty {
|
|
ty.fold(&mut |ty| match ty {
|
|
Ty::Infer(tv) => {
|
|
let inner = tv.to_inner();
|
|
if tv_stack.contains(&inner) {
|
|
tested_by!(type_var_cycles_resolve_as_possible);
|
|
// recursive type
|
|
return tv.fallback_value();
|
|
}
|
|
if let Some(known_ty) =
|
|
self.var_unification_table.inlined_probe_value(inner).known()
|
|
{
|
|
// known_ty may contain other variables that are known by now
|
|
tv_stack.push(inner);
|
|
let result = self.resolve_ty_as_possible_inner(tv_stack, known_ty.clone());
|
|
tv_stack.pop();
|
|
result
|
|
} else {
|
|
ty
|
|
}
|
|
}
|
|
_ => ty,
|
|
})
|
|
}
|
|
|
|
/// Resolves the type completely; type variables without known type are
|
|
/// replaced by Ty::Unknown.
|
|
fn resolve_ty_completely_inner(&mut self, tv_stack: &mut Vec<TypeVarId>, ty: Ty) -> Ty {
|
|
ty.fold(&mut |ty| match ty {
|
|
Ty::Infer(tv) => {
|
|
let inner = tv.to_inner();
|
|
if tv_stack.contains(&inner) {
|
|
tested_by!(type_var_cycles_resolve_completely);
|
|
// recursive type
|
|
return tv.fallback_value();
|
|
}
|
|
if let Some(known_ty) =
|
|
self.var_unification_table.inlined_probe_value(inner).known()
|
|
{
|
|
// known_ty may contain other variables that are known by now
|
|
tv_stack.push(inner);
|
|
let result = self.resolve_ty_completely_inner(tv_stack, known_ty.clone());
|
|
tv_stack.pop();
|
|
result
|
|
} else {
|
|
tv.fallback_value()
|
|
}
|
|
}
|
|
_ => ty,
|
|
})
|
|
}
|
|
}
|
|
|
|
/// The ID of a type variable.
|
|
#[derive(Copy, Clone, PartialEq, Eq, Hash, Debug)]
|
|
pub struct TypeVarId(pub(super) u32);
|
|
|
|
impl UnifyKey for TypeVarId {
|
|
type Value = TypeVarValue;
|
|
|
|
fn index(&self) -> u32 {
|
|
self.0
|
|
}
|
|
|
|
fn from_index(i: u32) -> Self {
|
|
TypeVarId(i)
|
|
}
|
|
|
|
fn tag() -> &'static str {
|
|
"TypeVarId"
|
|
}
|
|
}
|
|
|
|
/// The value of a type variable: either we already know the type, or we don't
|
|
/// know it yet.
|
|
#[derive(Clone, PartialEq, Eq, Debug)]
|
|
pub enum TypeVarValue {
|
|
Known(Ty),
|
|
Unknown,
|
|
}
|
|
|
|
impl TypeVarValue {
|
|
fn known(&self) -> Option<&Ty> {
|
|
match self {
|
|
TypeVarValue::Known(ty) => Some(ty),
|
|
TypeVarValue::Unknown => None,
|
|
}
|
|
}
|
|
}
|
|
|
|
impl UnifyValue for TypeVarValue {
|
|
type Error = NoError;
|
|
|
|
fn unify_values(value1: &Self, value2: &Self) -> Result<Self, NoError> {
|
|
match (value1, value2) {
|
|
// We should never equate two type variables, both of which have
|
|
// known types. Instead, we recursively equate those types.
|
|
(TypeVarValue::Known(t1), TypeVarValue::Known(t2)) => panic!(
|
|
"equating two type variables, both of which have known types: {:?} and {:?}",
|
|
t1, t2
|
|
),
|
|
|
|
// If one side is known, prefer that one.
|
|
(TypeVarValue::Known(..), TypeVarValue::Unknown) => Ok(value1.clone()),
|
|
(TypeVarValue::Unknown, TypeVarValue::Known(..)) => Ok(value2.clone()),
|
|
|
|
(TypeVarValue::Unknown, TypeVarValue::Unknown) => Ok(TypeVarValue::Unknown),
|
|
}
|
|
}
|
|
}
|