Auto merge of rust-lang#95563 - dingxiangfei2009:dxf-rfc66-refactor, …

…r=nikomatsakis

Move the extended lifetime resolution into typeck context

Related to rust-lang#15023

This PR is based on the [idea](rust-lang#15023 (comment)) of rust-lang#15023 by `@nikomatsakis.`

This PR specifically proposes to
- Delay the resolution of scopes of rvalues to a later stage, so that enough type information is available to refine those scopes based on relationships of lifetimes.
- Highlight relevant parts that would help future reviews on the next installments of works to fully implement a solution to RFC 66.

compiler/rustc_hir/src/hir.rs

@@ -1370,7 +1370,7 @@ pub enum UnsafeSource {
     UserProvided,
 }
 
-#[derive(Copy, Clone, PartialEq, Eq, Encodable, Hash, Debug)]
+#[derive(Copy, Clone, PartialEq, Eq, Encodable, Decodable, Hash, Debug)]
 pub struct BodyId {
     pub hir_id: HirId,
 }

compiler/rustc_middle/src/arena.rs

@@ -33,7 +33,6 @@ macro_rules! arena_types {
             [] const_allocs: rustc_middle::mir::interpret::Allocation,
             // Required for the incremental on-disk cache
             [] mir_keys: rustc_hir::def_id::DefIdSet,
-            [] region_scope_tree: rustc_middle::middle::region::ScopeTree,
             [] dropck_outlives:
                 rustc_middle::infer::canonical::Canonical<'tcx,
                     rustc_middle::infer::canonical::QueryResponse<'tcx,

compiler/rustc_middle/src/middle/region.rs

@@ -203,7 +203,7 @@ impl Scope {
 pub type ScopeDepth = u32;
 
 /// The region scope tree encodes information about region relationships.
-#[derive(Default, Debug)]
+#[derive(TyEncodable, TyDecodable, Default, Debug)]
 pub struct ScopeTree {
     /// If not empty, this body is the root of this region hierarchy.
     pub root_body: Option<hir::HirId>,
@@ -223,15 +223,12 @@ pub struct ScopeTree {
     /// Maps from a `NodeId` to the associated destruction scope (if any).
     destruction_scopes: FxIndexMap<hir::ItemLocalId, Scope>,
 
-    /// `rvalue_scopes` includes entries for those expressions whose
-    /// cleanup scope is larger than the default. The map goes from the
-    /// expression ID to the cleanup scope id. For rvalues not present in
-    /// this table, the appropriate cleanup scope is the innermost
-    /// enclosing statement, conditional expression, or repeating
-    /// block (see `terminating_scopes`).
-    /// In constants, None is used to indicate that certain expressions
-    /// escape into 'static and should have no local cleanup scope.
-    rvalue_scopes: FxHashMap<hir::ItemLocalId, Option<Scope>>,
+    /// Identifies expressions which, if captured into a temporary, ought to
+    /// have a temporary whose lifetime extends to the end of the enclosing *block*,
+    /// and not the enclosing *statement*. Expressions that are not present in this
+    /// table are not rvalue candidates. The set of rvalue candidates is computed
+    /// during type check based on a traversal of the AST.
+    pub rvalue_candidates: FxHashMap<hir::HirId, RvalueCandidateType>,
 
     /// If there are any `yield` nested within a scope, this map
     /// stores the `Span` of the last one and its index in the
@@ -315,6 +312,17 @@ pub struct ScopeTree {
     pub body_expr_count: FxHashMap<hir::BodyId, usize>,
 }
 
+/// Identifies the reason that a given expression is an rvalue candidate
+/// (see the `rvalue_candidates` field for more information what rvalue
+/// candidates in general). In constants, the `lifetime` field is None
+/// to indicate that certain expressions escape into 'static and
+/// should have no local cleanup scope.
+#[derive(Debug, Copy, Clone, TyEncodable, TyDecodable, HashStable)]
+pub enum RvalueCandidateType {
+    Borrow { target: hir::ItemLocalId, lifetime: Option<Scope> },
+    Pattern { target: hir::ItemLocalId, lifetime: Option<Scope> },
+}
+
 #[derive(Debug, Copy, Clone, TyEncodable, TyDecodable, HashStable)]
 pub struct YieldData {
     /// The `Span` of the yield.
@@ -349,12 +357,20 @@ impl ScopeTree {
         self.var_map.insert(var, lifetime);
     }
 
-    pub fn record_rvalue_scope(&mut self, var: hir::ItemLocalId, lifetime: Option<Scope>) {
-        debug!("record_rvalue_scope(sub={:?}, sup={:?})", var, lifetime);
-        if let Some(lifetime) = lifetime {
-            assert!(var != lifetime.item_local_id());
+    pub fn record_rvalue_candidate(
+        &mut self,
+        var: hir::HirId,
+        candidate_type: RvalueCandidateType,
+    ) {
+        debug!("record_rvalue_candidate(var={var:?}, type={candidate_type:?})");
+        match &candidate_type {
+            RvalueCandidateType::Borrow { lifetime: Some(lifetime), .. }
+            | RvalueCandidateType::Pattern { lifetime: Some(lifetime), .. } => {
+                assert!(var.local_id != lifetime.item_local_id())
+            }
+            _ => {}
         }
-        self.rvalue_scopes.insert(var, lifetime);
+        self.rvalue_candidates.insert(var, candidate_type);
     }
 
     /// Returns the narrowest scope that encloses `id`, if any.
@@ -367,34 +383,6 @@ impl ScopeTree {
         self.var_map.get(&var_id).cloned()
     }
 
-    /// Returns the scope when the temp created by `expr_id` will be cleaned up.
-    pub fn temporary_scope(&self, expr_id: hir::ItemLocalId) -> Option<Scope> {
-        // Check for a designated rvalue scope.
-        if let Some(&s) = self.rvalue_scopes.get(&expr_id) {
-            debug!("temporary_scope({:?}) = {:?} [custom]", expr_id, s);
-            return s;
-        }
-
-        // Otherwise, locate the innermost terminating scope
-        // if there's one. Static items, for instance, won't
-        // have an enclosing scope, hence no scope will be
-        // returned.
-        let mut id = Scope { id: expr_id, data: ScopeData::Node };
-
-        while let Some(&(p, _)) = self.parent_map.get(&id) {
-            match p.data {
-                ScopeData::Destruction => {
-                    debug!("temporary_scope({:?}) = {:?} [enclosing]", expr_id, id);
-                    return Some(id);
-                }
-                _ => id = p,
-            }
-        }
-
-        debug!("temporary_scope({:?}) = None", expr_id);
-        None
-    }
-
     /// Returns `true` if `subscope` is equal to or is lexically nested inside `superscope`, and
     /// `false` otherwise.
     ///
@@ -439,7 +427,7 @@ impl<'a> HashStable<StableHashingContext<'a>> for ScopeTree {
             ref parent_map,
             ref var_map,
             ref destruction_scopes,
-            ref rvalue_scopes,
+            ref rvalue_candidates,
             ref yield_in_scope,
         } = *self;
 
@@ -448,7 +436,7 @@ impl<'a> HashStable<StableHashingContext<'a>> for ScopeTree {
         parent_map.hash_stable(hcx, hasher);
         var_map.hash_stable(hcx, hasher);
         destruction_scopes.hash_stable(hcx, hasher);
-        rvalue_scopes.hash_stable(hcx, hasher);
+        rvalue_candidates.hash_stable(hcx, hasher);
         yield_in_scope.hash_stable(hcx, hasher);
     }
 }

compiler/rustc_middle/src/query/mod.rs

@@ -1048,12 +1048,6 @@ rustc_queries! {
         desc { "reachability" }
     }
 
-    /// Per-body `region::ScopeTree`. The `DefId` should be the owner `DefId` for the body;
-    /// in the case of closures, this will be redirected to the enclosing function.
-    query region_scope_tree(def_id: DefId) -> &'tcx region::ScopeTree {
-        desc { |tcx| "computing drop scopes for `{}`", tcx.def_path_str(def_id) }
-    }
-
     /// Generates a MIR body for the shim.
     query mir_shims(key: ty::InstanceDef<'tcx>) -> mir::Body<'tcx> {
         storage(ArenaCacheSelector<'tcx>)

compiler/rustc_middle/src/ty/context.rs

@@ -6,6 +6,7 @@ use crate::hir::place::Place as HirPlace;
 use crate::infer::canonical::{Canonical, CanonicalVarInfo, CanonicalVarInfos};
 use crate::lint::{struct_lint_level, LintDiagnosticBuilder, LintLevelSource};
 use crate::middle::codegen_fn_attrs::CodegenFnAttrs;
+use crate::middle::region::ScopeTree;
 use crate::middle::resolve_lifetime::{self, LifetimeScopeForPath};
 use crate::middle::stability;
 use crate::mir::interpret::{self, Allocation, ConstAllocation, ConstValue, Scalar};
@@ -74,6 +75,8 @@ use std::mem;
 use std::ops::{Bound, Deref};
 use std::sync::Arc;
 
+use super::RvalueScopes;
+
 pub trait OnDiskCache<'tcx>: rustc_data_structures::sync::Sync {
     /// Creates a new `OnDiskCache` instance from the serialized data in `data`.
     fn new(sess: &'tcx Session, data: Mmap, start_pos: usize) -> Self
@@ -535,6 +538,17 @@ pub struct TypeckResults<'tcx> {
     /// issue by fake reading `t`.
     pub closure_fake_reads: FxHashMap<DefId, Vec<(HirPlace<'tcx>, FakeReadCause, hir::HirId)>>,
 
+    /// Tracks critical information about regions in a body.
+    /// This includes containment relationship between regions,
+    /// liveness relationship between variables and regions and
+    /// information about yield points.
+    pub region_scope_tree: ScopeTree,
+
+    /// Tracks the rvalue scoping rules which defines finer scoping for rvalue expressions
+    /// by applying extended parameter rules.
+    /// Details may be find in `rustc_typeck::check::rvalue_scopes`.
+    pub rvalue_scopes: RvalueScopes,
+
     /// Stores the type, expression, span and optional scope span of all types
     /// that are live across the yield of this generator (if a generator).
     pub generator_interior_types: ty::Binder<'tcx, Vec<GeneratorInteriorTypeCause<'tcx>>>,
@@ -572,6 +586,8 @@ impl<'tcx> TypeckResults<'tcx> {
             concrete_opaque_types: Default::default(),
             closure_min_captures: Default::default(),
             closure_fake_reads: Default::default(),
+            region_scope_tree: Default::default(),
+            rvalue_scopes: Default::default(),
             generator_interior_types: ty::Binder::dummy(Default::default()),
             treat_byte_string_as_slice: Default::default(),
             closure_size_eval: Default::default(),

compiler/rustc_middle/src/ty/mod.rs

@@ -72,6 +72,7 @@ pub use self::context::{
 };
 pub use self::instance::{Instance, InstanceDef};
 pub use self::list::List;
+pub use self::rvalue_scopes::RvalueScopes;
 pub use self::sty::BoundRegionKind::*;
 pub use self::sty::RegionKind::*;
 pub use self::sty::TyKind::*;
@@ -118,6 +119,7 @@ mod generics;
 mod impls_ty;
 mod instance;
 mod list;
+mod rvalue_scopes;
 mod structural_impls;
 mod sty;
 

compiler/rustc_middle/src/ty/query.rs

@@ -6,7 +6,6 @@ use crate::middle::codegen_fn_attrs::CodegenFnAttrs;
 use crate::middle::exported_symbols::{ExportedSymbol, SymbolExportInfo};
 use crate::middle::lib_features::LibFeatures;
 use crate::middle::privacy::AccessLevels;
-use crate::middle::region;
 use crate::middle::resolve_lifetime::{
     LifetimeScopeForPath, ObjectLifetimeDefault, Region, ResolveLifetimes,
 };

compiler/rustc_middle/src/ty/rvalue_scopes.rs

@@ -0,0 +1,57 @@
+use crate::middle::region::{Scope, ScopeData, ScopeTree};
+use rustc_data_structures::fx::FxHashMap;
+use rustc_hir as hir;
+
+/// `RvalueScopes` is a mapping from sub-expressions to _extended_ lifetime as determined by
+/// rules laid out in `rustc_typeck::check::rvalue_scopes`.
+#[derive(TyEncodable, TyDecodable, Clone, Debug, Default, Eq, PartialEq, HashStable)]
+pub struct RvalueScopes {
+    map: FxHashMap<hir::ItemLocalId, Option<Scope>>,
+}
+
+impl RvalueScopes {
+    pub fn new() -> Self {
+        Self { map: <_>::default() }
+    }
+
+    /// Returns the scope when the temp created by `expr_id` will be cleaned up.
+    pub fn temporary_scope(
+        &self,
+        region_scope_tree: &ScopeTree,
+        expr_id: hir::ItemLocalId,
+    ) -> Option<Scope> {
+        // Check for a designated rvalue scope.
+        if let Some(&s) = self.map.get(&expr_id) {
+            debug!("temporary_scope({expr_id:?}) = {s:?} [custom]");
+            return s;
+        }
+
+        // Otherwise, locate the innermost terminating scope
+        // if there's one. Static items, for instance, won't
+        // have an enclosing scope, hence no scope will be
+        // returned.
+        let mut id = Scope { id: expr_id, data: ScopeData::Node };
+
+        while let Some(&(p, _)) = region_scope_tree.parent_map.get(&id) {
+            match p.data {
+                ScopeData::Destruction => {
+                    debug!("temporary_scope({expr_id:?}) = {id:?} [enclosing]");
+                    return Some(id);
+                }
+                _ => id = p,
+            }
+        }
+
+        debug!("temporary_scope({expr_id:?}) = None");
+        None
+    }
+
+    /// Make an association between a sub-expression and an extended lifetime
+    pub fn record_rvalue_scope(&mut self, var: hir::ItemLocalId, lifetime: Option<Scope>) {
+        debug!("record_rvalue_scope(var={var:?}, lifetime={lifetime:?})");
+        if let Some(lifetime) = lifetime {
+            assert!(var != lifetime.item_local_id());
+        }
+        self.map.insert(var, lifetime);
+    }
+}

compiler/rustc_mir_build/src/build/block.rs

@@ -108,7 +108,8 @@ impl<'a, 'tcx> Builder<'a, 'tcx> {
                     let_scope_stack.push(remainder_scope);
 
                     // Declare the bindings, which may create a source scope.
-                    let remainder_span = remainder_scope.span(this.tcx, this.region_scope_tree);
+                    let remainder_span =
+                        remainder_scope.span(this.tcx, &this.typeck_results.region_scope_tree);
 
                     let visibility_scope =
                         Some(this.new_source_scope(remainder_span, LintLevel::Inherited, None));

compiler/rustc_mir_build/src/build/matches/mod.rs

@@ -699,7 +699,7 @@ impl<'a, 'tcx> Builder<'a, 'tcx> {
         self.cfg.push(block, Statement { source_info, kind: StatementKind::StorageLive(local_id) });
         // Altough there is almost always scope for given variable in corner cases
         // like #92893 we might get variable with no scope.
-        if let Some(region_scope) = self.region_scope_tree.var_scope(var.local_id) && schedule_drop{
+        if let Some(region_scope) = self.typeck_results.region_scope_tree.var_scope(var.local_id) && schedule_drop{
             self.schedule_drop(span, region_scope, local_id, DropKind::Storage);
         }
         Place::from(local_id)
@@ -712,7 +712,7 @@ impl<'a, 'tcx> Builder<'a, 'tcx> {
         for_guard: ForGuard,
     ) {
         let local_id = self.var_local_id(var, for_guard);
-        if let Some(region_scope) = self.region_scope_tree.var_scope(var.local_id) {
+        if let Some(region_scope) = self.typeck_results.region_scope_tree.var_scope(var.local_id) {
             self.schedule_drop(span, region_scope, local_id, DropKind::Value);
         }
     }

compiler/rustc_mir_build/src/build/mod.rs

@@ -398,7 +398,6 @@ struct Builder<'a, 'tcx> {
     tcx: TyCtxt<'tcx>,
     infcx: &'a InferCtxt<'a, 'tcx>,
     typeck_results: &'tcx TypeckResults<'tcx>,
-    region_scope_tree: &'tcx region::ScopeTree,
     param_env: ty::ParamEnv<'tcx>,
 
     thir: &'a Thir<'tcx>,
@@ -881,7 +880,6 @@ impl<'a, 'tcx> Builder<'a, 'tcx> {
             tcx,
             infcx,
             typeck_results: tcx.typeck_opt_const_arg(def),
-            region_scope_tree: tcx.region_scope_tree(def.did),
             param_env,
             def_id: def.did.to_def_id(),
             hir_id,

compiler/rustc_mir_build/src/build/scope.rs

@@ -916,7 +916,8 @@ impl<'a, 'tcx> Builder<'a, 'tcx> {
             }
 
             if scope.region_scope == region_scope {
-                let region_scope_span = region_scope.span(self.tcx, &self.region_scope_tree);
+                let region_scope_span =
+                    region_scope.span(self.tcx, &self.typeck_results.region_scope_tree);
                 // Attribute scope exit drops to scope's closing brace.
                 let scope_end = self.tcx.sess.source_map().end_point(region_scope_span);