Auto merge of rust-lang#89030 - nbdd0121:box2, r=jonas-schievink

Introduce `Rvalue::ShallowInitBox`

Polished version of rust-lang#88700.

Implements MCP rust-lang/compiler-team#460, and should allow rust-lang#43596 to go forward.

In short, creating an empty box is split from a nullary-op `NullOp::Box` into two steps, first a call to `exchange_malloc`, then a `Rvalue::ShallowInitBox` which transmutes `*mut u8` to a shallow-initialized `Box<T>`. This allows the `exchange_malloc` call to unwind. Details can be found in the MCP.

`NullOp::Box` is not yet removed, purely to make reverting easier in case anything goes wrong as the result of this PR. If revert is needed a reversion of "Use Rvalue::ShallowInitBox for box expression" commit followed by a test bless should be sufficient.

Experiments in rust-lang#88700 showed a very slight compile-time perf regression due to (supposedly) slightly more time spent in LLVM. We could omit unwind edge generation (in non-`oom=panic` case) in box expression MIR construction to restore perf; but I don't think it's necessary since runtime perf isn't affected and perf difference is rather small.

Author: bors

compiler/rustc_borrowck/src/invalidation.rs

@@ -316,7 +316,8 @@ impl<'cx, 'tcx> InvalidationGenerator<'cx, 'tcx> {
             Rvalue::Use(ref operand)
             | Rvalue::Repeat(ref operand, _)
             | Rvalue::UnaryOp(_ /*un_op*/, ref operand)
-            | Rvalue::Cast(_ /*cast_kind*/, ref operand, _ /*ty*/) => {
+            | Rvalue::Cast(_ /*cast_kind*/, ref operand, _ /*ty*/)
+            | Rvalue::ShallowInitBox(ref operand, _ /*ty*/) => {
                 self.consume_operand(location, operand)
             }
 

compiler/rustc_borrowck/src/lib.rs

@@ -1361,7 +1361,8 @@ impl<'cx, 'tcx> MirBorrowckCtxt<'cx, 'tcx> {
             Rvalue::Use(ref operand)
             | Rvalue::Repeat(ref operand, _)
             | Rvalue::UnaryOp(_ /*un_op*/, ref operand)
-            | Rvalue::Cast(_ /*cast_kind*/, ref operand, _ /*ty*/) => {
+            | Rvalue::Cast(_ /*cast_kind*/, ref operand, _ /*ty*/)
+            | Rvalue::ShallowInitBox(ref operand, _ /*ty*/) => {
                 self.consume_operand(location, (operand, span), flow_state)
             }
 

compiler/rustc_borrowck/src/type_check/mod.rs

@@ -2024,13 +2024,7 @@ impl<'a, 'tcx> TypeChecker<'a, 'tcx> {
                 }
             }
 
-            Rvalue::NullaryOp(_, ty) => {
-                // Even with unsized locals cannot box an unsized value.
-                if self.unsized_feature_enabled() {
-                    let span = body.source_info(location).span;
-                    self.ensure_place_sized(ty, span);
-                }
-
+            Rvalue::NullaryOp(_, ty) | Rvalue::ShallowInitBox(_, ty) => {
                 let trait_ref = ty::TraitRef {
                     def_id: tcx.require_lang_item(LangItem::Sized, Some(self.last_span)),
                     substs: tcx.mk_substs_trait(ty, &[]),
@@ -2363,6 +2357,7 @@ impl<'a, 'tcx> TypeChecker<'a, 'tcx> {
             | Rvalue::AddressOf(..)
             | Rvalue::Len(..)
             | Rvalue::Cast(..)
+            | Rvalue::ShallowInitBox(..)
             | Rvalue::BinaryOp(..)
             | Rvalue::CheckedBinaryOp(..)
             | Rvalue::NullaryOp(..)

compiler/rustc_codegen_cranelift/src/base.rs

@@ -701,6 +701,13 @@ fn codegen_stmt<'tcx>(
                     let len = codegen_array_len(fx, place);
                     lval.write_cvalue(fx, CValue::by_val(len, usize_layout));
                 }
+                Rvalue::ShallowInitBox(ref operand, content_ty) => {
+                    let content_ty = fx.monomorphize(content_ty);
+                    let box_layout = fx.layout_of(fx.tcx.mk_box(content_ty));
+                    let operand = codegen_operand(fx, operand);
+                    let operand = operand.load_scalar(fx);
+                    lval.write_cvalue(fx, CValue::by_val(operand, box_layout));
+                }
                 Rvalue::NullaryOp(NullOp::Box, content_ty) => {
                     let usize_type = fx.clif_type(fx.tcx.types.usize).unwrap();
                     let content_ty = fx.monomorphize(content_ty);

compiler/rustc_codegen_ssa/src/mir/rvalue.rs

@@ -550,6 +550,18 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
                     OperandRef::new_zst(&mut bx, self.cx.layout_of(self.monomorphize(ty)));
                 (bx, operand)
             }
+            mir::Rvalue::ShallowInitBox(ref operand, content_ty) => {
+                let operand = self.codegen_operand(&mut bx, operand);
+                let lloperand = operand.immediate();
+
+                let content_ty = self.monomorphize(content_ty);
+                let box_layout = bx.cx().layout_of(bx.tcx().mk_box(content_ty));
+                let llty_ptr = bx.cx().backend_type(box_layout);
+
+                let val = bx.pointercast(lloperand, llty_ptr);
+                let operand = OperandRef { val: OperandValue::Immediate(val), layout: box_layout };
+                (bx, operand)
+            }
         }
     }
 
@@ -763,6 +775,7 @@ impl<'a, 'tcx, Bx: BuilderMethods<'a, 'tcx>> FunctionCx<'a, 'tcx, Bx> {
             mir::Rvalue::AddressOf(..) |
             mir::Rvalue::Len(..) |
             mir::Rvalue::Cast(..) | // (*)
+            mir::Rvalue::ShallowInitBox(..) | // (*)
             mir::Rvalue::BinaryOp(..) |
             mir::Rvalue::CheckedBinaryOp(..) |
             mir::Rvalue::UnaryOp(..) |

compiler/rustc_const_eval/src/interpret/step.rs

@@ -289,6 +289,12 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
                 self.write_scalar(Scalar::from_machine_usize(val, self), &dest)?;
             }
 
+            ShallowInitBox(ref operand, _) => {
+                let src = self.eval_operand(operand, None)?;
+                let v = self.read_immediate(&src)?;
+                self.write_immediate(*v, &dest)?;
+            }
+
             Cast(cast_kind, ref operand, cast_ty) => {
                 let src = self.eval_operand(operand, None)?;
                 let cast_ty = self.subst_from_current_frame_and_normalize_erasing_regions(cast_ty);

compiler/rustc_const_eval/src/transform/check_consts/check.rs

@@ -650,6 +650,7 @@ impl Visitor<'tcx> for Checker<'mir, 'tcx> {
 
             Rvalue::NullaryOp(NullOp::SizeOf | NullOp::AlignOf, _) => {}
             Rvalue::NullaryOp(NullOp::Box, _) => self.check_op(ops::HeapAllocation),
+            Rvalue::ShallowInitBox(_, _) => {}
 
             Rvalue::UnaryOp(_, ref operand) => {
                 let ty = operand.ty(self.body, self.tcx);
@@ -912,6 +913,11 @@ impl Visitor<'tcx> for Checker<'mir, 'tcx> {
                     return;
                 }
 
+                if Some(callee) == tcx.lang_items().exchange_malloc_fn() {
+                    self.check_op(ops::HeapAllocation);
+                    return;
+                }
+
                 // `async` blocks get lowered to `std::future::from_generator(/* a closure */)`.
                 let is_async_block = Some(callee) == tcx.lang_items().from_generator_fn();
                 if is_async_block {

compiler/rustc_const_eval/src/transform/check_consts/qualifs.rs

@@ -206,7 +206,8 @@ where
         Rvalue::Use(operand)
         | Rvalue::Repeat(operand, _)
         | Rvalue::UnaryOp(_, operand)
-        | Rvalue::Cast(_, operand, _) => in_operand::<Q, _>(cx, in_local, operand),
+        | Rvalue::Cast(_, operand, _)
+        | Rvalue::ShallowInitBox(operand, _) => in_operand::<Q, _>(cx, in_local, operand),
 
         Rvalue::BinaryOp(_, box (lhs, rhs)) | Rvalue::CheckedBinaryOp(_, box (lhs, rhs)) => {
             in_operand::<Q, _>(cx, in_local, lhs) || in_operand::<Q, _>(cx, in_local, rhs)

compiler/rustc_const_eval/src/transform/promote_consts.rs

@@ -523,6 +523,8 @@ impl<'tcx> Validator<'_, 'tcx> {
                 NullOp::AlignOf => {}
             },
 
+            Rvalue::ShallowInitBox(_, _) => return Err(Unpromotable),
+
             Rvalue::UnaryOp(op, operand) => {
                 match op {
                     // These operations can never fail.

compiler/rustc_middle/src/mir/mod.rs

@@ -2200,6 +2200,12 @@ pub enum Rvalue<'tcx> {
     /// that `Foo` has a destructor. These rvalues can be optimized
     /// away after type-checking and before lowering.
     Aggregate(Box<AggregateKind<'tcx>>, Vec<Operand<'tcx>>),
+
+    /// Transmutes a `*mut u8` into shallow-initialized `Box<T>`.
+    ///
+    /// This is different a normal transmute because dataflow analysis will treat the box
+    /// as initialized but its content as uninitialized.
+    ShallowInitBox(Operand<'tcx>, Ty<'tcx>),
 }
 
 #[cfg(all(target_arch = "x86_64", target_pointer_width = "64"))]
@@ -2450,6 +2456,10 @@ impl<'tcx> Debug for Rvalue<'tcx> {
                     }),
                 }
             }
+
+            ShallowInitBox(ref place, ref ty) => {
+                write!(fmt, "ShallowInitBox({:?}, {:?})", place, ty)
+            }
         }
     }
 }