Auto merge of #98910 - Dylan-DPC:rollup-9x82wdg, r=Dylan-DPC

Rollup of 6 pull requests

Successful merges:

 - #97300 (Implement `FusedIterator` for `std::net::[Into]Incoming`)
 - #98761 (more `need_type_info` improvements)
 - #98811 (Interpret: AllocRange Debug impl, and use it more consistently)
 - #98847 (fix interpreter validity check on Box)
 - #98854 (clean up the borrowing in rustc_hir_pretty)
 - #98873 (Suggest `#[derive(Default)]` to enums with `#[default]`)

Failed merges:

r? `@ghost`
`@rustbot` modify labels: rollup

Author: bors

compiler/rustc_const_eval/src/interpret/memory.rs

@@ -276,7 +276,7 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
         kind: MemoryKind<M::MemoryKind>,
     ) -> InterpResult<'tcx> {
         let (alloc_id, offset, tag) = self.ptr_get_alloc_id(ptr)?;
-        trace!("deallocating: {}", alloc_id);
+        trace!("deallocating: {alloc_id:?}");
 
         if offset.bytes() != 0 {
             throw_ub_format!(
@@ -289,10 +289,10 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
             // Deallocating global memory -- always an error
             return Err(match self.tcx.get_global_alloc(alloc_id) {
                 Some(GlobalAlloc::Function(..)) => {
-                    err_ub_format!("deallocating {}, which is a function", alloc_id)
+                    err_ub_format!("deallocating {alloc_id:?}, which is a function")
                 }
                 Some(GlobalAlloc::Static(..) | GlobalAlloc::Memory(..)) => {
-                    err_ub_format!("deallocating {}, which is static memory", alloc_id)
+                    err_ub_format!("deallocating {alloc_id:?}, which is static memory")
                 }
                 None => err_ub!(PointerUseAfterFree(alloc_id)),
             }
@@ -302,21 +302,17 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
         debug!(?alloc);
 
         if alloc.mutability == Mutability::Not {
-            throw_ub_format!("deallocating immutable allocation {}", alloc_id);
+            throw_ub_format!("deallocating immutable allocation {alloc_id:?}");
         }
         if alloc_kind != kind {
             throw_ub_format!(
-                "deallocating {}, which is {} memory, using {} deallocation operation",
-                alloc_id,
-                alloc_kind,
-                kind
+                "deallocating {alloc_id:?}, which is {alloc_kind} memory, using {kind} deallocation operation"
             );
         }
         if let Some((size, align)) = old_size_and_align {
             if size != alloc.size() || align != alloc.align {
                 throw_ub_format!(
-                    "incorrect layout on deallocation: {} has size {} and alignment {}, but gave size {} and alignment {}",
-                    alloc_id,
+                    "incorrect layout on deallocation: {alloc_id:?} has size {} and alignment {}, but gave size {} and alignment {}",
                     alloc.size().bytes(),
                     alloc.align.bytes(),
                     size.bytes(),
@@ -815,7 +811,7 @@ impl<'a, 'mir, 'tcx, M: Machine<'mir, 'tcx>> std::fmt::Debug for DumpAllocs<'a,
                 continue;
             }
 
-            write!(fmt, "{}", id)?;
+            write!(fmt, "{id:?}")?;
             match self.ecx.memory.alloc_map.get(id) {
                 Some(&(kind, ref alloc)) => {
                     // normal alloc
@@ -859,25 +855,21 @@ impl<'a, 'mir, 'tcx, M: Machine<'mir, 'tcx>> std::fmt::Debug for DumpAllocs<'a,
 
 /// Reading and writing.
 impl<'tcx, 'a, Tag: Provenance, Extra> AllocRefMut<'a, 'tcx, Tag, Extra> {
+    /// `range` is relative to this allocation reference, not the base of the allocation.
     pub fn write_scalar(
         &mut self,
         range: AllocRange,
         val: ScalarMaybeUninit<Tag>,
     ) -> InterpResult<'tcx> {
         let range = self.range.subrange(range);
-        debug!(
-            "write_scalar in {} at {:#x}, size {}: {:?}",
-            self.alloc_id,
-            range.start.bytes(),
-            range.size.bytes(),
-            val
-        );
+        debug!("write_scalar at {:?}{range:?}: {val:?}", self.alloc_id);
         Ok(self
             .alloc
             .write_scalar(&self.tcx, range, val)
             .map_err(|e| e.to_interp_error(self.alloc_id))?)
     }
 
+    /// `offset` is relative to this allocation reference, not the base of the allocation.
     pub fn write_ptr_sized(
         &mut self,
         offset: Size,
@@ -896,6 +888,7 @@ impl<'tcx, 'a, Tag: Provenance, Extra> AllocRefMut<'a, 'tcx, Tag, Extra> {
 }
 
 impl<'tcx, 'a, Tag: Provenance, Extra> AllocRef<'a, 'tcx, Tag, Extra> {
+    /// `range` is relative to this allocation reference, not the base of the allocation.
     pub fn read_scalar(
         &self,
         range: AllocRange,
@@ -906,31 +899,24 @@ impl<'tcx, 'a, Tag: Provenance, Extra> AllocRef<'a, 'tcx, Tag, Extra> {
             .alloc
             .read_scalar(&self.tcx, range, read_provenance)
             .map_err(|e| e.to_interp_error(self.alloc_id))?;
-        debug!(
-            "read_scalar in {} at {:#x}, size {}: {:?}",
-            self.alloc_id,
-            range.start.bytes(),
-            range.size.bytes(),
-            res
-        );
+        debug!("read_scalar at {:?}{range:?}: {res:?}", self.alloc_id);
         Ok(res)
     }
 
-    pub fn read_integer(
-        &self,
-        offset: Size,
-        size: Size,
-    ) -> InterpResult<'tcx, ScalarMaybeUninit<Tag>> {
-        self.read_scalar(alloc_range(offset, size), /*read_provenance*/ false)
+    /// `range` is relative to this allocation reference, not the base of the allocation.
+    pub fn read_integer(&self, range: AllocRange) -> InterpResult<'tcx, ScalarMaybeUninit<Tag>> {
+        self.read_scalar(range, /*read_provenance*/ false)
     }
 
+    /// `offset` is relative to this allocation reference, not the base of the allocation.
     pub fn read_pointer(&self, offset: Size) -> InterpResult<'tcx, ScalarMaybeUninit<Tag>> {
         self.read_scalar(
             alloc_range(offset, self.tcx.data_layout().pointer_size),
             /*read_provenance*/ true,
         )
     }
 
+    /// `range` is relative to this allocation reference, not the base of the allocation.
     pub fn check_bytes(
         &self,
         range: AllocRange,

compiler/rustc_const_eval/src/interpret/traits.rs

@@ -1,6 +1,6 @@
 use std::convert::TryFrom;
 
-use rustc_middle::mir::interpret::{InterpResult, Pointer, PointerArithmetic};
+use rustc_middle::mir::interpret::{alloc_range, InterpResult, Pointer, PointerArithmetic};
 use rustc_middle::ty::{
     self, Ty, TyCtxt, COMMON_VTABLE_ENTRIES_ALIGN, COMMON_VTABLE_ENTRIES_DROPINPLACE,
     COMMON_VTABLE_ENTRIES_SIZE,
@@ -102,18 +102,18 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
             )?
             .expect("cannot be a ZST");
         let size = vtable
-            .read_integer(
+            .read_integer(alloc_range(
                 pointer_size * u64::try_from(COMMON_VTABLE_ENTRIES_SIZE).unwrap(),
                 pointer_size,
-            )?
+            ))?
             .check_init()?;
         let size = size.to_machine_usize(self)?;
         let size = Size::from_bytes(size);
         let align = vtable
-            .read_integer(
+            .read_integer(alloc_range(
                 pointer_size * u64::try_from(COMMON_VTABLE_ENTRIES_ALIGN).unwrap(),
                 pointer_size,
-            )?
+            ))?
             .check_init()?;
         let align = align.to_machine_usize(self)?;
         let align = Align::from_bytes(align).map_err(|e| err_ub!(InvalidVtableAlignment(e)))?;

compiler/rustc_const_eval/src/interpret/validity.rs

@@ -593,16 +593,6 @@ impl<'rt, 'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> ValidityVisitor<'rt, 'mir, '
                 self.check_safe_pointer(value, "reference")?;
                 Ok(true)
             }
-            ty::Adt(def, ..) if def.is_box() => {
-                let unique = self.ecx.operand_field(value, 0)?;
-                let nonnull = self.ecx.operand_field(&unique, 0)?;
-                let ptr = self.ecx.operand_field(&nonnull, 0)?;
-                self.check_safe_pointer(&ptr, "box")?;
-
-                // Check other fields of Box
-                self.walk_value(value)?;
-                Ok(true)
-            }
             ty::FnPtr(_sig) => {
                 let value = try_validation!(
                     self.ecx.read_scalar(value).and_then(|v| v.check_init()),
@@ -813,6 +803,12 @@ impl<'rt, 'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> ValueVisitor<'mir, 'tcx, M>
         Ok(())
     }
 
+    #[inline]
+    fn visit_box(&mut self, op: &OpTy<'tcx, M::PointerTag>) -> InterpResult<'tcx> {
+        self.check_safe_pointer(op, "box")?;
+        Ok(())
+    }
+
     #[inline]
     fn visit_value(&mut self, op: &OpTy<'tcx, M::PointerTag>) -> InterpResult<'tcx> {
         trace!("visit_value: {:?}, {:?}", *op, op.layout);
@@ -821,8 +817,6 @@ impl<'rt, 'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> ValueVisitor<'mir, 'tcx, M>
         if self.try_visit_primitive(op)? {
             return Ok(());
         }
-        // Sanity check: `builtin_deref` does not know any pointers that are not primitive.
-        assert!(op.layout.ty.builtin_deref(true).is_none());
 
         // Special check preventing `UnsafeCell` in the inner part of constants
         if let Some(def) = op.layout.ty.ty_adt_def() {

compiler/rustc_const_eval/src/interpret/visitor.rs

@@ -151,6 +151,14 @@ macro_rules! make_value_visitor {
             {
                 Ok(())
             }
+            /// Visits the given value as the pointer of a `Box`. There is nothing to recurse into.
+            /// The type of `v` will be a raw pointer, but this is a field of `Box<T>` and the
+            /// pointee type is the actual `T`.
+            #[inline(always)]
+            fn visit_box(&mut self, _v: &Self::V) -> InterpResult<'tcx>
+            {
+                Ok(())
+            }
             /// Visits this value as an aggregate, you are getting an iterator yielding
             /// all the fields (still in an `InterpResult`, you have to do error handling yourself).
             /// Recurses into the fields.
@@ -221,6 +229,47 @@ macro_rules! make_value_visitor {
                     // Slices do not need special handling here: they have `Array` field
                     // placement with length 0, so we enter the `Array` case below which
                     // indirectly uses the metadata to determine the actual length.
+
+                    // However, `Box`... let's talk about `Box`.
+                    ty::Adt(def, ..) if def.is_box() => {
+                        // `Box` is a hybrid primitive-library-defined type that one the one hand is
+                        // a dereferenceable pointer, on the other hand has *basically arbitrary
+                        // user-defined layout* since the user controls the 'allocator' field. So it
+                        // cannot be treated like a normal pointer, since it does not fit into an
+                        // `Immediate`. Yeah, it is quite terrible. But many visitors want to do
+                        // something with "all boxed pointers", so we handle this mess for them.
+                        //
+                        // When we hit a `Box`, we do not do the usual `visit_aggregate`; instead,
+                        // we (a) call `visit_box` on the pointer value, and (b) recurse on the
+                        // allocator field. We also assert tons of things to ensure we do not miss
+                        // any other fields.
+
+                        // `Box` has two fields: the pointer we care about, and the allocator.
+                        assert_eq!(v.layout().fields.count(), 2, "`Box` must have exactly 2 fields");
+                        let (unique_ptr, alloc) =
+                            (v.project_field(self.ecx(), 0)?, v.project_field(self.ecx(), 1)?);
+                        // Unfortunately there is some type junk in the way here: `unique_ptr` is a `Unique`...
+                        // (which means another 2 fields, the second of which is a `PhantomData`)
+                        assert_eq!(unique_ptr.layout().fields.count(), 2);
+                        let (nonnull_ptr, phantom) = (
+                            unique_ptr.project_field(self.ecx(), 0)?,
+                            unique_ptr.project_field(self.ecx(), 1)?,
+                        );
+                        assert!(
+                            phantom.layout().ty.ty_adt_def().is_some_and(|adt| adt.is_phantom_data()),
+                            "2nd field of `Unique` should be PhantomData but is {:?}",
+                            phantom.layout().ty,
+                        );
+                        // ... that contains a `NonNull`... (gladly, only a single field here)
+                        assert_eq!(nonnull_ptr.layout().fields.count(), 1);
+                        let raw_ptr = nonnull_ptr.project_field(self.ecx(), 0)?; // the actual raw ptr
+                        // ... whose only field finally is a raw ptr we can dereference.
+                        self.visit_box(&raw_ptr)?;
+
+                        // The second `Box` field is the allocator, which we recursively check for validity
+                        // like in regular structs.
+                        self.visit_field(v, 1, &alloc)?;
+                    }
                     _ => {},
                 };
 

compiler/rustc_const_eval/src/lib.rs

@@ -21,6 +21,7 @@ Rust MIR: a lowered representation of Rust.
 #![feature(trusted_step)]
 #![feature(try_blocks)]
 #![feature(yeet_expr)]
+#![feature(is_some_with)]
 #![recursion_limit = "256"]
 #![allow(rustc::potential_query_instability)]