rust-analyzer/crates/hir-ty/src/utils.rs
Chayim Refael Friedman 4e475a3245 Store some hir_def Paths in the type ref source maps
Most paths are types and therefore already are in the source map, but the trait in impl trait and in bounds are not.

We do this by storing them basically as `TypeRef`s. For convenience, I created a wrapper around `TypeRefId` called `PathId` that always stores a path, and implemented indexing from the types map to it.

Fortunately, this change impacts memory usage negligibly (adds 2mb to `analysis-stats .`, but that could be just fluff). Probably because there aren't that many trait bounds and impl traits, and this also shrinks `TypeBound` by 8 bytes.

I also added an accessor to `TypesSourceMap` to get the source code, which will be needed for diagnostics.
2024-12-04 14:09:50 +02:00

392 lines
13 KiB
Rust

//! Helper functions for working with def, which don't need to be a separate
//! query, but can't be computed directly from `*Data` (ie, which need a `db`).
use std::{hash::Hash, iter};
use base_db::CrateId;
use chalk_ir::{
fold::{FallibleTypeFolder, Shift},
DebruijnIndex,
};
use hir_def::{
db::DefDatabase,
generics::{WherePredicate, WherePredicateTypeTarget},
lang_item::LangItem,
resolver::{HasResolver, TypeNs},
type_ref::{TraitBoundModifier, TypeRef},
EnumId, EnumVariantId, FunctionId, Lookup, OpaqueInternableThing, TraitId, TypeAliasId,
TypeOrConstParamId,
};
use hir_expand::name::Name;
use intern::sym;
use rustc_abi::TargetDataLayout;
use rustc_hash::FxHashSet;
use smallvec::{smallvec, SmallVec};
use stdx::never;
use crate::{
consteval::unknown_const,
db::HirDatabase,
layout::{Layout, TagEncoding},
mir::pad16,
ChalkTraitId, Const, ConstScalar, GenericArg, Interner, Substitution, TraitRef, TraitRefExt,
Ty, WhereClause,
};
pub(crate) fn fn_traits(
db: &dyn DefDatabase,
krate: CrateId,
) -> impl Iterator<Item = TraitId> + '_ {
[LangItem::Fn, LangItem::FnMut, LangItem::FnOnce]
.into_iter()
.filter_map(move |lang| db.lang_item(krate, lang))
.flat_map(|it| it.as_trait())
}
/// Returns an iterator over the direct super traits (including the trait itself).
pub fn direct_super_traits(db: &dyn DefDatabase, trait_: TraitId) -> SmallVec<[TraitId; 4]> {
let mut result = smallvec![trait_];
direct_super_traits_cb(db, trait_, |tt| {
if !result.contains(&tt) {
result.push(tt);
}
});
result
}
/// Returns an iterator over the whole super trait hierarchy (including the
/// trait itself).
pub fn all_super_traits(db: &dyn DefDatabase, trait_: TraitId) -> SmallVec<[TraitId; 4]> {
// we need to take care a bit here to avoid infinite loops in case of cycles
// (i.e. if we have `trait A: B; trait B: A;`)
let mut result = smallvec![trait_];
let mut i = 0;
while let Some(&t) = result.get(i) {
// yeah this is quadratic, but trait hierarchies should be flat
// enough that this doesn't matter
direct_super_traits_cb(db, t, |tt| {
if !result.contains(&tt) {
result.push(tt);
}
});
i += 1;
}
result
}
/// Given a trait ref (`Self: Trait`), builds all the implied trait refs for
/// super traits. The original trait ref will be included. So the difference to
/// `all_super_traits` is that we keep track of type parameters; for example if
/// we have `Self: Trait<u32, i32>` and `Trait<T, U>: OtherTrait<U>` we'll get
/// `Self: OtherTrait<i32>`.
pub(super) fn all_super_trait_refs<T>(
db: &dyn HirDatabase,
trait_ref: TraitRef,
cb: impl FnMut(TraitRef) -> Option<T>,
) -> Option<T> {
let seen = iter::once(trait_ref.trait_id).collect();
SuperTraits { db, seen, stack: vec![trait_ref] }.find_map(cb)
}
struct SuperTraits<'a> {
db: &'a dyn HirDatabase,
stack: Vec<TraitRef>,
seen: FxHashSet<ChalkTraitId>,
}
impl SuperTraits<'_> {
fn elaborate(&mut self, trait_ref: &TraitRef) {
direct_super_trait_refs(self.db, trait_ref, |trait_ref| {
if !self.seen.contains(&trait_ref.trait_id) {
self.stack.push(trait_ref);
}
});
}
}
impl Iterator for SuperTraits<'_> {
type Item = TraitRef;
fn next(&mut self) -> Option<Self::Item> {
if let Some(next) = self.stack.pop() {
self.elaborate(&next);
Some(next)
} else {
None
}
}
}
pub(super) fn elaborate_clause_supertraits(
db: &dyn HirDatabase,
clauses: impl Iterator<Item = WhereClause>,
) -> ClauseElaborator<'_> {
let mut elaborator = ClauseElaborator { db, stack: Vec::new(), seen: FxHashSet::default() };
elaborator.extend_deduped(clauses);
elaborator
}
pub(super) struct ClauseElaborator<'a> {
db: &'a dyn HirDatabase,
stack: Vec<WhereClause>,
seen: FxHashSet<WhereClause>,
}
impl ClauseElaborator<'_> {
fn extend_deduped(&mut self, clauses: impl IntoIterator<Item = WhereClause>) {
self.stack.extend(clauses.into_iter().filter(|c| self.seen.insert(c.clone())))
}
fn elaborate_supertrait(&mut self, clause: &WhereClause) {
if let WhereClause::Implemented(trait_ref) = clause {
direct_super_trait_refs(self.db, trait_ref, |t| {
let clause = WhereClause::Implemented(t);
if self.seen.insert(clause.clone()) {
self.stack.push(clause);
}
});
}
}
}
impl Iterator for ClauseElaborator<'_> {
type Item = WhereClause;
fn next(&mut self) -> Option<Self::Item> {
if let Some(next) = self.stack.pop() {
self.elaborate_supertrait(&next);
Some(next)
} else {
None
}
}
}
fn direct_super_traits_cb(db: &dyn DefDatabase, trait_: TraitId, cb: impl FnMut(TraitId)) {
let resolver = trait_.resolver(db);
let generic_params = db.generic_params(trait_.into());
let trait_self = generic_params.trait_self_param();
generic_params
.where_predicates()
.filter_map(|pred| match pred {
WherePredicate::ForLifetime { target, bound, .. }
| WherePredicate::TypeBound { target, bound } => {
let is_trait = match target {
WherePredicateTypeTarget::TypeRef(type_ref) => {
match &generic_params.types_map[*type_ref] {
TypeRef::Path(p) => p.is_self_type(),
_ => false,
}
}
WherePredicateTypeTarget::TypeOrConstParam(local_id) => {
Some(*local_id) == trait_self
}
};
match is_trait {
true => bound.as_path(&generic_params.types_map),
false => None,
}
}
WherePredicate::Lifetime { .. } => None,
})
.filter(|(_, bound_modifier)| matches!(bound_modifier, TraitBoundModifier::None))
.filter_map(|(path, _)| match resolver.resolve_path_in_type_ns_fully(db, path) {
Some(TypeNs::TraitId(t)) => Some(t),
_ => None,
})
.for_each(cb);
}
fn direct_super_trait_refs(db: &dyn HirDatabase, trait_ref: &TraitRef, cb: impl FnMut(TraitRef)) {
let generic_params = db.generic_params(trait_ref.hir_trait_id().into());
let trait_self = match generic_params.trait_self_param() {
Some(p) => TypeOrConstParamId { parent: trait_ref.hir_trait_id().into(), local_id: p },
None => return,
};
db.generic_predicates_for_param(trait_self.parent, trait_self, None)
.iter()
.filter_map(|pred| {
pred.as_ref().filter_map(|pred| match pred.skip_binders() {
// FIXME: how to correctly handle higher-ranked bounds here?
WhereClause::Implemented(tr) => Some(
tr.clone()
.shifted_out_to(Interner, DebruijnIndex::ONE)
.expect("FIXME unexpected higher-ranked trait bound"),
),
_ => None,
})
})
.map(|pred| pred.substitute(Interner, &trait_ref.substitution))
.for_each(cb);
}
pub(super) fn associated_type_by_name_including_super_traits(
db: &dyn HirDatabase,
trait_ref: TraitRef,
name: &Name,
) -> Option<(TraitRef, TypeAliasId)> {
all_super_trait_refs(db, trait_ref, |t| {
let assoc_type = db.trait_data(t.hir_trait_id()).associated_type_by_name(name)?;
Some((t, assoc_type))
})
}
/// It is a bit different from the rustc equivalent. Currently it stores:
/// - 0: the function signature, encoded as a function pointer type
/// - 1..n: generics of the parent
///
/// and it doesn't store the closure types and fields.
///
/// Codes should not assume this ordering, and should always use methods available
/// on this struct for retrieving, and `TyBuilder::substs_for_closure` for creating.
pub(crate) struct ClosureSubst<'a>(pub(crate) &'a Substitution);
impl<'a> ClosureSubst<'a> {
pub(crate) fn parent_subst(&self) -> &'a [GenericArg] {
match self.0.as_slice(Interner) {
[_, x @ ..] => x,
_ => {
never!("Closure missing parameter");
&[]
}
}
}
pub(crate) fn sig_ty(&self) -> &'a Ty {
match self.0.as_slice(Interner) {
[x, ..] => x.assert_ty_ref(Interner),
_ => {
unreachable!("Closure missing sig_ty parameter");
}
}
}
}
pub fn is_fn_unsafe_to_call(db: &dyn HirDatabase, func: FunctionId) -> bool {
let data = db.function_data(func);
if data.is_unsafe() {
return true;
}
let is_intrinsic = db.attrs(func.into()).by_key(&sym::rustc_intrinsic).exists()
|| data.abi.as_ref() == Some(&sym::rust_dash_intrinsic);
let loc = func.lookup(db.upcast());
match loc.container {
hir_def::ItemContainerId::ExternBlockId(block) => {
if is_intrinsic || {
let id = block.lookup(db.upcast()).id;
id.item_tree(db.upcast())[id.value].abi.as_ref() == Some(&sym::rust_dash_intrinsic)
} {
// Intrinsics are unsafe unless they have the rustc_safe_intrinsic attribute
!db.attrs(func.into()).by_key(&sym::rustc_safe_intrinsic).exists()
} else {
// Function in an `extern` block are always unsafe to call, except when
// it is marked as `safe`.
!data.is_safe()
}
}
_ if is_intrinsic => !db.attrs(func.into()).by_key(&sym::rustc_safe_intrinsic).exists(),
_ => false,
}
}
pub(crate) struct UnevaluatedConstEvaluatorFolder<'a> {
pub(crate) db: &'a dyn HirDatabase,
}
impl FallibleTypeFolder<Interner> for UnevaluatedConstEvaluatorFolder<'_> {
type Error = ();
fn as_dyn(&mut self) -> &mut dyn FallibleTypeFolder<Interner, Error = ()> {
self
}
fn interner(&self) -> Interner {
Interner
}
fn try_fold_const(
&mut self,
constant: Const,
_outer_binder: DebruijnIndex,
) -> Result<Const, Self::Error> {
if let chalk_ir::ConstValue::Concrete(c) = &constant.data(Interner).value {
if let ConstScalar::UnevaluatedConst(id, subst) = &c.interned {
if let Ok(eval) = self.db.const_eval(*id, subst.clone(), None) {
return Ok(eval);
} else {
return Ok(unknown_const(constant.data(Interner).ty.clone()));
}
}
}
Ok(constant)
}
}
pub(crate) fn detect_variant_from_bytes<'a>(
layout: &'a Layout,
db: &dyn HirDatabase,
target_data_layout: &TargetDataLayout,
b: &[u8],
e: EnumId,
) -> Option<(EnumVariantId, &'a Layout)> {
let (var_id, var_layout) = match &layout.variants {
hir_def::layout::Variants::Single { index } => {
(db.enum_data(e).variants[index.0].0, layout)
}
hir_def::layout::Variants::Multiple { tag, tag_encoding, variants, .. } => {
let size = tag.size(target_data_layout).bytes_usize();
let offset = layout.fields.offset(0).bytes_usize(); // The only field on enum variants is the tag field
let tag = i128::from_le_bytes(pad16(&b[offset..offset + size], false));
match tag_encoding {
TagEncoding::Direct => {
let (var_idx, layout) =
variants.iter_enumerated().find_map(|(var_idx, v)| {
let def = db.enum_data(e).variants[var_idx.0].0;
(db.const_eval_discriminant(def) == Ok(tag)).then_some((def, v))
})?;
(var_idx, layout)
}
TagEncoding::Niche { untagged_variant, niche_start, .. } => {
let candidate_tag = tag.wrapping_sub(*niche_start as i128) as usize;
let variant = variants
.iter_enumerated()
.map(|(x, _)| x)
.filter(|x| x != untagged_variant)
.nth(candidate_tag)
.unwrap_or(*untagged_variant);
(db.enum_data(e).variants[variant.0].0, &variants[variant])
}
}
}
};
Some((var_id, var_layout))
}
#[derive(Debug, Clone, PartialEq, Eq, Hash)]
pub(crate) struct InTypeConstIdMetadata(pub(crate) Ty);
impl OpaqueInternableThing for InTypeConstIdMetadata {
fn dyn_hash(&self, mut state: &mut dyn std::hash::Hasher) {
self.hash(&mut state);
}
fn dyn_eq(&self, other: &dyn OpaqueInternableThing) -> bool {
other.as_any().downcast_ref::<Self>().map_or(false, |x| self == x)
}
fn dyn_clone(&self) -> Box<dyn OpaqueInternableThing> {
Box::new(self.clone())
}
fn as_any(&self) -> &dyn std::any::Any {
self
}
fn box_any(&self) -> Box<dyn std::any::Any> {
Box::new(self.clone())
}
}