interpret: refactor projection handling code

compiler/rustc_const_eval/src/interpret/mod.rs

@@ -9,6 +9,7 @@ mod memory;
 mod operand;
 mod operator;
 mod place;
+mod projection;
 mod step;
 mod terminator;
 mod traits;

compiler/rustc_const_eval/src/interpret/operand.rs

@@ -1,7 +1,6 @@
 //! Functions concerning immediate values and operands, and reading from operands.
 //! All high-level functions to read from memory work on operands as sources.
 
-use std::convert::TryFrom;
 use std::fmt::Write;
 
 use rustc_hir::def::Namespace;
@@ -15,7 +14,7 @@ use rustc_target::abi::{VariantIdx, Variants};
 
 use super::{
     alloc_range, from_known_layout, mir_assign_valid_types, AllocId, ConstValue, Frame, GlobalId,
-    InterpCx, InterpResult, MPlaceTy, Machine, MemPlace, Place, PlaceTy, Pointer,
+    InterpCx, InterpResult, MPlaceTy, Machine, MemPlace, MemPlaceMeta, Place, PlaceTy, Pointer,
     PointerArithmetic, Provenance, Scalar, ScalarMaybeUninit,
 };
 
@@ -253,6 +252,11 @@ impl<'tcx, Tag: Provenance> ImmTy<'tcx, Tag> {
         ImmTy { imm, layout }
     }
 
+    #[inline]
+    pub fn uninit(layout: TyAndLayout<'tcx>) -> Self {
+        ImmTy { imm: Immediate::Uninit, layout }
+    }
+
     #[inline]
     pub fn try_from_uint(i: impl Into<u128>, layout: TyAndLayout<'tcx>) -> Option<Self> {
         Some(Self::from_scalar(Scalar::try_from_uint(i, layout.size)?, layout))
@@ -280,6 +284,41 @@ impl<'tcx, Tag: Provenance> ImmTy<'tcx, Tag> {
     }
 }
 
+impl<'tcx, Tag: Provenance> OpTy<'tcx, Tag> {
+    pub fn len(&self, cx: &impl HasDataLayout) -> InterpResult<'tcx, u64> {
+        if self.layout.is_unsized() {
+            // There are no unsized immediates.
+            self.assert_mem_place().len(cx)
+        } else {
+            match self.layout.fields {
+                abi::FieldsShape::Array { count, .. } => Ok(count),
+                _ => bug!("len not supported on sized type {:?}", self.layout.ty),
+            }
+        }
+    }
+
+    pub fn offset(
+        &self,
+        offset: Size,
+        meta: MemPlaceMeta<Tag>,
+        layout: TyAndLayout<'tcx>,
+        cx: &impl HasDataLayout,
+    ) -> InterpResult<'tcx, Self> {
+        match self.try_as_mplace() {
+            Ok(mplace) => Ok(mplace.offset(offset, meta, layout, cx)?.into()),
+            Err(imm) => {
+                assert!(
+                    matches!(*imm, Immediate::Uninit),
+                    "Scalar/ScalarPair cannot be offset into"
+                );
+                assert!(!meta.has_meta()); // no place to store metadata here
+                // Every part of an uninit is uninit.
+                Ok(ImmTy::uninit(layout).into())
+            }
+        }
+    }
+}
+
 impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
     /// Try reading an immediate in memory; this is interesting particularly for `ScalarPair`.
     /// Returns `None` if the layout does not permit loading this as a value.
@@ -296,11 +335,8 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
         }
 
         let Some(alloc) = self.get_place_alloc(mplace)? else {
-            return Ok(Some(ImmTy {
-                // zero-sized type can be left uninit
-                imm: Immediate::Uninit,
-                layout: mplace.layout,
-            }));
+            // zero-sized type can be left uninit
+            return Ok(Some(ImmTy::uninit(mplace.layout)));
         };
 
         // It may seem like all types with `Scalar` or `ScalarPair` ABI are fair game at this point.
@@ -367,6 +403,7 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
     /// This flag exists only for validity checking.
     ///
     /// This is an internal function that should not usually be used; call `read_immediate` instead.
+    /// ConstProp needs it, though.
     pub fn read_immediate_raw(
         &self,
         src: &OpTy<'tcx, M::PointerTag>,
@@ -421,123 +458,28 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
         Ok(str)
     }
 
-    /// Projection functions
-    pub fn operand_field(
-        &self,
-        op: &OpTy<'tcx, M::PointerTag>,
-        field: usize,
-    ) -> InterpResult<'tcx, OpTy<'tcx, M::PointerTag>> {
-        let base = match op.try_as_mplace() {
-            Ok(ref mplace) => {
-                // We can reuse the mplace field computation logic for indirect operands.
-                let field = self.mplace_field(mplace, field)?;
-                return Ok(field.into());
-            }
-            Err(value) => value,
-        };
-
-        let field_layout = base.layout.field(self, field);
-        let offset = base.layout.fields.offset(field);
-        // This makes several assumptions about what layouts we will encounter; we match what
-        // codegen does as good as we can (see `extract_field` in `rustc_codegen_ssa/src/mir/operand.rs`).
-        let field_val: Immediate<_> = match (*base, base.layout.abi) {
-            // the field contains no information, can be left uninit
-            _ if field_layout.is_zst() => Immediate::Uninit,
-            // the field covers the entire type
-            _ if field_layout.size == base.layout.size => {
-                assert!(match (base.layout.abi, field_layout.abi) {
-                    (Abi::Scalar(..), Abi::Scalar(..)) => true,
-                    (Abi::ScalarPair(..), Abi::ScalarPair(..)) => true,
-                    _ => false,
-                });
-                assert!(offset.bytes() == 0);
-                *base
-            }
-            // extract fields from types with `ScalarPair` ABI
-            (Immediate::ScalarPair(a_val, b_val), Abi::ScalarPair(a, b)) => {
-                assert!(matches!(field_layout.abi, Abi::Scalar(..)));
-                Immediate::from(if offset.bytes() == 0 {
-                    debug_assert_eq!(field_layout.size, a.size(self));
-                    a_val
-                } else {
-                    debug_assert_eq!(offset, a.size(self).align_to(b.align(self).abi));
-                    debug_assert_eq!(field_layout.size, b.size(self));
-                    b_val
-                })
-            }
-            _ => span_bug!(
-                self.cur_span(),
-                "invalid field access on immediate {}, layout {:#?}",
-                base,
-                base.layout
-            ),
-        };
-
-        Ok(OpTy { op: Operand::Immediate(field_val), layout: field_layout, align: None })
-    }
-
-    pub fn operand_index(
-        &self,
-        op: &OpTy<'tcx, M::PointerTag>,
-        index: u64,
-    ) -> InterpResult<'tcx, OpTy<'tcx, M::PointerTag>> {
-        if let Ok(index) = usize::try_from(index) {
-            // We can just treat this as a field.
-            self.operand_field(op, index)
-        } else {
-            // Indexing into a big array. This must be an mplace.
-            let mplace = op.assert_mem_place();
-            Ok(self.mplace_index(&mplace, index)?.into())
-        }
-    }
-
-    pub fn operand_downcast(
-        &self,
-        op: &OpTy<'tcx, M::PointerTag>,
-        variant: VariantIdx,
-    ) -> InterpResult<'tcx, OpTy<'tcx, M::PointerTag>> {
-        Ok(match op.try_as_mplace() {
-            Ok(ref mplace) => self.mplace_downcast(mplace, variant)?.into(),
-            Err(..) => {
-                // Downcasts only change the layout.
-                // (In particular, no check about whether this is even the active variant -- that's by design,
-                // see https://github.com/rust-lang/rust/issues/93688#issuecomment-1032929496.)
-                let layout = op.layout.for_variant(self, variant);
-                OpTy { layout, ..*op }
-            }
-        })
-    }
-
-    #[instrument(skip(self), level = "debug")]
-    pub fn operand_projection(
-        &self,
-        base: &OpTy<'tcx, M::PointerTag>,
-        proj_elem: mir::PlaceElem<'tcx>,
-    ) -> InterpResult<'tcx, OpTy<'tcx, M::PointerTag>> {
-        use rustc_middle::mir::ProjectionElem::*;
-        Ok(match proj_elem {
-            Field(field, _) => self.operand_field(base, field.index())?,
-            Downcast(_, variant) => self.operand_downcast(base, variant)?,
-            Deref => self.deref_operand(base)?.into(),
-            Subslice { .. } | ConstantIndex { .. } | Index(_) => {
-                // The rest should only occur as mplace, we do not use Immediates for types
-                // allowing such operations.  This matches place_projection forcing an allocation.
-                let mplace = base.assert_mem_place();
-                self.mplace_projection(&mplace, proj_elem)?.into()
-            }
-        })
-    }
-
     /// Converts a repr(simd) operand into an operand where `place_index` accesses the SIMD elements.
     /// Also returns the number of elements.
+    ///
+    /// Can (but does not always) trigger UB if `op` is uninitialized.
     pub fn operand_to_simd(
         &self,
-        base: &OpTy<'tcx, M::PointerTag>,
+        op: &OpTy<'tcx, M::PointerTag>,
     ) -> InterpResult<'tcx, (MPlaceTy<'tcx, M::PointerTag>, u64)> {
         // Basically we just transmute this place into an array following simd_size_and_type.
         // This only works in memory, but repr(simd) types should never be immediates anyway.
-        assert!(base.layout.ty.is_simd());
-        self.mplace_to_simd(&base.assert_mem_place())
+        assert!(op.layout.ty.is_simd());
+        match op.try_as_mplace() {
+            Ok(mplace) => self.mplace_to_simd(&mplace),
+            Err(imm) => match *imm {
+                Immediate::Uninit => {
+                    throw_ub!(InvalidUninitBytes(None))
+                }
+                Immediate::Scalar(..) | Immediate::ScalarPair(..) => {
+                    bug!("arrays/slices can never have Scalar/ScalarPair layout")
+                }
+            },
+        }
     }
 
     /// Read from a local. Will not actually access the local if reading from a ZST.
@@ -582,30 +524,34 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
     /// avoid allocations.
     pub fn eval_place_to_op(
         &self,
-        place: mir::Place<'tcx>,
+        mir_place: mir::Place<'tcx>,
         layout: Option<TyAndLayout<'tcx>>,
     ) -> InterpResult<'tcx, OpTy<'tcx, M::PointerTag>> {
         // Do not use the layout passed in as argument if the base we are looking at
         // here is not the entire place.
-        let layout = if place.projection.is_empty() { layout } else { None };
+        let layout = if mir_place.projection.is_empty() { layout } else { None };
 
-        let base_op = self.local_to_op(self.frame(), place.local, layout)?;
-
-        let op = place
-            .projection
-            .iter()
-            .try_fold(base_op, |op, elem| self.operand_projection(&op, elem))?;
+        let mut op = self.local_to_op(self.frame(), mir_place.local, layout)?;
+        // Using `try_fold` turned out to be bad for performance, hence the loop.
+        for elem in mir_place.projection.iter() {
+            op = self.operand_projection(&op, elem)?
+        }
 
         trace!("eval_place_to_op: got {:?}", *op);
         // Sanity-check the type we ended up with.
-        debug_assert!(mir_assign_valid_types(
-            *self.tcx,
-            self.param_env,
-            self.layout_of(self.subst_from_current_frame_and_normalize_erasing_regions(
-                place.ty(&self.frame().body.local_decls, *self.tcx).ty
-            )?)?,
-            op.layout,
-        ));
+        debug_assert!(
+            mir_assign_valid_types(
+                *self.tcx,
+                self.param_env,
+                self.layout_of(self.subst_from_current_frame_and_normalize_erasing_regions(
+                    mir_place.ty(&self.frame().body.local_decls, *self.tcx).ty
+                )?)?,
+                op.layout,
+            ),
+            "eval_place of a MIR place with type {:?} produced an interpreter operand with type {:?}",
+            mir_place.ty(&self.frame().body.local_decls, *self.tcx).ty,
+            op.layout.ty,
+        );
         Ok(op)
     }
 

compiler/rustc_const_eval/src/interpret/projection.rs

@@ -0,0 +1,393 @@
+//! This file implements "place projections"; basically a symmetric API for 3 types: MPlaceTy, OpTy, PlaceTy.
+//!
+//! OpTy and PlaceTy genrally work by "let's see if we are actually an MPlaceTy, and do something custom if not".
+//! For PlaceTy, the custom thing is basically always to call `force_allocation` and then use the MPlaceTy logic anyway.
+//! For OpTy, the custom thing on field pojections has to be pretty clever (since `Operand::Immediate` can have fields),
+//! but for array/slice operations it only has to worry about `Operand::Uninit`. That makes the value part trivial,
+//! but we still need to do bounds checking and adjust the layout. To not duplicate that with MPlaceTy, we actually
+//! implement the logic on OpTy, and MPlaceTy calls that.
+
+use std::hash::Hash;
+
+use rustc_middle::mir;
+use rustc_middle::ty;
+use rustc_middle::ty::layout::LayoutOf;
+use rustc_target::abi::{self, Abi, VariantIdx};
+
+use super::{
+    ImmTy, Immediate, InterpCx, InterpResult, MPlaceTy, Machine, MemPlaceMeta, OpTy, PlaceTy,
+    Provenance, Scalar,
+};
+
+// FIXME: Working around https://github.com/rust-lang/rust/issues/54385
+impl<'mir, 'tcx: 'mir, Tag, M> InterpCx<'mir, 'tcx, M>
+where
+    Tag: Provenance + Eq + Hash + 'static,
+    M: Machine<'mir, 'tcx, PointerTag = Tag>,
+{
+    //# Field access
+
+    /// Offset a pointer to project to a field of a struct/union. Unlike `place_field`, this is
+    /// always possible without allocating, so it can take `&self`. Also return the field's layout.
+    /// This supports both struct and array fields.
+    ///
+    /// This also works for arrays, but then the `usize` index type is restricting.
+    /// For indexing into arrays, use `mplace_index`.
+    pub fn mplace_field(
+        &self,
+        base: &MPlaceTy<'tcx, M::PointerTag>,
+        field: usize,
+    ) -> InterpResult<'tcx, MPlaceTy<'tcx, M::PointerTag>> {
+        let offset = base.layout.fields.offset(field);
+        let field_layout = base.layout.field(self, field);
+
+        // Offset may need adjustment for unsized fields.
+        let (meta, offset) = if field_layout.is_unsized() {
+            // Re-use parent metadata to determine dynamic field layout.
+            // With custom DSTS, this *will* execute user-defined code, but the same
+            // happens at run-time so that's okay.
+            match self.size_and_align_of(&base.meta, &field_layout)? {
+                Some((_, align)) => (base.meta, offset.align_to(align)),
+                None => {
+                    // For unsized types with an extern type tail we perform no adjustments.
+                    // NOTE: keep this in sync with `PlaceRef::project_field` in the codegen backend.
+                    assert!(matches!(base.meta, MemPlaceMeta::None));
+                    (base.meta, offset)
+                }
+            }
+        } else {
+            // base.meta could be present; we might be accessing a sized field of an unsized
+            // struct.
+            (MemPlaceMeta::None, offset)
+        };
+
+        // We do not look at `base.layout.align` nor `field_layout.align`, unlike
+        // codegen -- mostly to see if we can get away with that
+        base.offset(offset, meta, field_layout, self)
+    }
+
+    /// Gets the place of a field inside the place, and also the field's type.
+    /// Just a convenience function, but used quite a bit.
+    /// This is the only projection that might have a side-effect: We cannot project
+    /// into the field of a local `ScalarPair`, we have to first allocate it.
+    pub fn place_field(
+        &mut self,
+        base: &PlaceTy<'tcx, M::PointerTag>,
+        field: usize,
+    ) -> InterpResult<'tcx, PlaceTy<'tcx, M::PointerTag>> {
+        // FIXME: We could try to be smarter and avoid allocation for fields that span the
+        // entire place.
+        let base = self.force_allocation(base)?;
+        Ok(self.mplace_field(&base, field)?.into())
+    }
+
+    pub fn operand_field(
+        &self,
+        base: &OpTy<'tcx, M::PointerTag>,
+        field: usize,
+    ) -> InterpResult<'tcx, OpTy<'tcx, M::PointerTag>> {
+        let base = match base.try_as_mplace() {
+            Ok(ref mplace) => {
+                // We can reuse the mplace field computation logic for indirect operands.
+                let field = self.mplace_field(mplace, field)?;
+                return Ok(field.into());
+            }
+            Err(value) => value,
+        };
+
+        let field_layout = base.layout.field(self, field);
+        let offset = base.layout.fields.offset(field);
+        // This makes several assumptions about what layouts we will encounter; we match what
+        // codegen does as good as we can (see `extract_field` in `rustc_codegen_ssa/src/mir/operand.rs`).
+        let field_val: Immediate<_> = match (*base, base.layout.abi) {
+            // the field contains no information, can be left uninit
+            _ if field_layout.is_zst() => Immediate::Uninit,
+            // the field covers the entire type
+            _ if field_layout.size == base.layout.size => {
+                assert!(match (base.layout.abi, field_layout.abi) {
+                    (Abi::Scalar(..), Abi::Scalar(..)) => true,
+                    (Abi::ScalarPair(..), Abi::ScalarPair(..)) => true,
+                    _ => false,
+                });
+                assert!(offset.bytes() == 0);
+                *base
+            }
+            // extract fields from types with `ScalarPair` ABI
+            (Immediate::ScalarPair(a_val, b_val), Abi::ScalarPair(a, b)) => {
+                assert!(matches!(field_layout.abi, Abi::Scalar(..)));
+                Immediate::from(if offset.bytes() == 0 {
+                    debug_assert_eq!(field_layout.size, a.size(self));
+                    a_val
+                } else {
+                    debug_assert_eq!(offset, a.size(self).align_to(b.align(self).abi));
+                    debug_assert_eq!(field_layout.size, b.size(self));
+                    b_val
+                })
+            }
+            _ => span_bug!(
+                self.cur_span(),
+                "invalid field access on immediate {}, layout {:#?}",
+                base,
+                base.layout
+            ),
+        };
+
+        Ok(ImmTy::from_immediate(field_val, field_layout).into())
+    }
+
+    //# Downcasting
+
+    pub fn mplace_downcast(
+        &self,
+        base: &MPlaceTy<'tcx, M::PointerTag>,
+        variant: VariantIdx,
+    ) -> InterpResult<'tcx, MPlaceTy<'tcx, M::PointerTag>> {
+        // Downcasts only change the layout.
+        // (In particular, no check about whether this is even the active variant -- that's by design,
+        // see https://github.com/rust-lang/rust/issues/93688#issuecomment-1032929496.)
+        assert!(!base.meta.has_meta());
+        let mut base = *base;
+        base.layout = base.layout.for_variant(self, variant);
+        Ok(base)
+    }
+
+    pub fn place_downcast(
+        &self,
+        base: &PlaceTy<'tcx, M::PointerTag>,
+        variant: VariantIdx,
+    ) -> InterpResult<'tcx, PlaceTy<'tcx, M::PointerTag>> {
+        // Downcast just changes the layout
+        let mut base = *base;
+        base.layout = base.layout.for_variant(self, variant);
+        Ok(base)
+    }
+
+    pub fn operand_downcast(
+        &self,
+        base: &OpTy<'tcx, M::PointerTag>,
+        variant: VariantIdx,
+    ) -> InterpResult<'tcx, OpTy<'tcx, M::PointerTag>> {
+        // Downcast just changes the layout
+        let mut base = *base;
+        base.layout = base.layout.for_variant(self, variant);
+        Ok(base)
+    }
+
+    //# Slice indexing
+
+    #[inline(always)]
+    pub fn operand_index(
+        &self,
+        base: &OpTy<'tcx, M::PointerTag>,
+        index: u64,
+    ) -> InterpResult<'tcx, OpTy<'tcx, M::PointerTag>> {
+        // Not using the layout method because we want to compute on u64
+        match base.layout.fields {
+            abi::FieldsShape::Array { stride, count: _ } => {
+                // `count` is nonsense for slices, use the dynamic length instead.
+                let len = base.len(self)?;
+                if index >= len {
+                    // This can only be reached in ConstProp and non-rustc-MIR.
+                    throw_ub!(BoundsCheckFailed { len, index });
+                }
+                let offset = stride * index; // `Size` multiplication
+                // All fields have the same layout.
+                let field_layout = base.layout.field(self, 0);
+                assert!(!field_layout.is_unsized());
+
+                base.offset(offset, MemPlaceMeta::None, field_layout, self)
+            }
+            _ => span_bug!(
+                self.cur_span(),
+                "`mplace_index` called on non-array type {:?}",
+                base.layout.ty
+            ),
+        }
+    }
+
+    // Iterates over all fields of an array. Much more efficient than doing the
+    // same by repeatedly calling `operand_index`.
+    pub fn operand_array_fields<'a>(
+        &self,
+        base: &'a OpTy<'tcx, Tag>,
+    ) -> InterpResult<'tcx, impl Iterator<Item = InterpResult<'tcx, OpTy<'tcx, Tag>>> + 'a> {
+        let len = base.len(self)?; // also asserts that we have a type where this makes sense
+        let abi::FieldsShape::Array { stride, .. } = base.layout.fields else {
+            span_bug!(self.cur_span(), "operand_array_fields: expected an array layout");
+        };
+        let layout = base.layout.field(self, 0);
+        let dl = &self.tcx.data_layout;
+        // `Size` multiplication
+        Ok((0..len).map(move |i| base.offset(stride * i, MemPlaceMeta::None, layout, dl)))
+    }
+
+    /// Index into an array.
+    pub fn mplace_index(
+        &self,
+        base: &MPlaceTy<'tcx, M::PointerTag>,
+        index: u64,
+    ) -> InterpResult<'tcx, MPlaceTy<'tcx, M::PointerTag>> {
+        Ok(self.operand_index(&base.into(), index)?.assert_mem_place())
+    }
+
+    pub fn place_index(
+        &mut self,
+        base: &PlaceTy<'tcx, M::PointerTag>,
+        index: u64,
+    ) -> InterpResult<'tcx, PlaceTy<'tcx, M::PointerTag>> {
+        // There's not a lot we can do here, since we cannot have a place to a part of a local. If
+        // we are accessing the only element of a 1-element array, it's still the entire local...
+        // that doesn't seem worth it.
+        let base = self.force_allocation(base)?;
+        Ok(self.mplace_index(&base, index)?.into())
+    }
+
+    //# ConstantIndex support
+
+    fn operand_constant_index(
+        &self,
+        base: &OpTy<'tcx, M::PointerTag>,
+        offset: u64,
+        min_length: u64,
+        from_end: bool,
+    ) -> InterpResult<'tcx, OpTy<'tcx, M::PointerTag>> {
+        let n = base.len(self)?;
+        if n < min_length {
+            // This can only be reached in ConstProp and non-rustc-MIR.
+            throw_ub!(BoundsCheckFailed { len: min_length, index: n });
+        }
+
+        let index = if from_end {
+            assert!(0 < offset && offset <= min_length);
+            n.checked_sub(offset).unwrap()
+        } else {
+            assert!(offset < min_length);
+            offset
+        };
+
+        self.operand_index(base, index)
+    }
+
+    fn place_constant_index(
+        &mut self,
+        base: &PlaceTy<'tcx, M::PointerTag>,
+        offset: u64,
+        min_length: u64,
+        from_end: bool,
+    ) -> InterpResult<'tcx, PlaceTy<'tcx, M::PointerTag>> {
+        let base = self.force_allocation(base)?;
+        Ok(self
+            .operand_constant_index(&base.into(), offset, min_length, from_end)?
+            .assert_mem_place()
+            .into())
+    }
+
+    //# Subslicing
+
+    fn operand_subslice(
+        &self,
+        base: &OpTy<'tcx, M::PointerTag>,
+        from: u64,
+        to: u64,
+        from_end: bool,
+    ) -> InterpResult<'tcx, OpTy<'tcx, M::PointerTag>> {
+        let len = base.len(self)?; // also asserts that we have a type where this makes sense
+        let actual_to = if from_end {
+            if from.checked_add(to).map_or(true, |to| to > len) {
+                // This can only be reached in ConstProp and non-rustc-MIR.
+                throw_ub!(BoundsCheckFailed { len: len, index: from.saturating_add(to) });
+            }
+            len.checked_sub(to).unwrap()
+        } else {
+            to
+        };
+
+        // Not using layout method because that works with usize, and does not work with slices
+        // (that have count 0 in their layout).
+        let from_offset = match base.layout.fields {
+            abi::FieldsShape::Array { stride, .. } => stride * from, // `Size` multiplication is checked
+            _ => {
+                span_bug!(self.cur_span(), "unexpected layout of index access: {:#?}", base.layout)
+            }
+        };
+
+        // Compute meta and new layout
+        let inner_len = actual_to.checked_sub(from).unwrap();
+        let (meta, ty) = match base.layout.ty.kind() {
+            // It is not nice to match on the type, but that seems to be the only way to
+            // implement this.
+            ty::Array(inner, _) => (MemPlaceMeta::None, self.tcx.mk_array(*inner, inner_len)),
+            ty::Slice(..) => {
+                let len = Scalar::from_machine_usize(inner_len, self);
+                (MemPlaceMeta::Meta(len), base.layout.ty)
+            }
+            _ => {
+                span_bug!(self.cur_span(), "cannot subslice non-array type: `{:?}`", base.layout.ty)
+            }
+        };
+        let layout = self.layout_of(ty)?;
+        base.offset(from_offset, meta, layout, self)
+    }
+
+    pub fn place_subslice(
+        &mut self,
+        base: &PlaceTy<'tcx, M::PointerTag>,
+        from: u64,
+        to: u64,
+        from_end: bool,
+    ) -> InterpResult<'tcx, PlaceTy<'tcx, M::PointerTag>> {
+        let base = self.force_allocation(base)?;
+        Ok(self.operand_subslice(&base.into(), from, to, from_end)?.assert_mem_place().into())
+    }
+
+    //# Applying a general projection
+
+    /// Projects into a place.
+    #[instrument(skip(self), level = "trace")]
+    pub fn place_projection(
+        &mut self,
+        base: &PlaceTy<'tcx, M::PointerTag>,
+        proj_elem: mir::PlaceElem<'tcx>,
+    ) -> InterpResult<'tcx, PlaceTy<'tcx, M::PointerTag>> {
+        use rustc_middle::mir::ProjectionElem::*;
+        Ok(match proj_elem {
+            Field(field, _) => self.place_field(base, field.index())?,
+            Downcast(_, variant) => self.place_downcast(base, variant)?,
+            Deref => self.deref_operand(&self.place_to_op(base)?)?.into(),
+            Index(local) => {
+                let layout = self.layout_of(self.tcx.types.usize)?;
+                let n = self.local_to_op(self.frame(), local, Some(layout))?;
+                let n = self.read_scalar(&n)?.to_machine_usize(self)?;
+                self.place_index(base, n)?
+            }
+            ConstantIndex { offset, min_length, from_end } => {
+                self.place_constant_index(base, offset, min_length, from_end)?
+            }
+            Subslice { from, to, from_end } => self.place_subslice(base, from, to, from_end)?,
+        })
+    }
+
+    #[instrument(skip(self), level = "trace")]
+    pub fn operand_projection(
+        &self,
+        base: &OpTy<'tcx, M::PointerTag>,
+        proj_elem: mir::PlaceElem<'tcx>,
+    ) -> InterpResult<'tcx, OpTy<'tcx, M::PointerTag>> {
+        use rustc_middle::mir::ProjectionElem::*;
+        Ok(match proj_elem {
+            Field(field, _) => self.operand_field(base, field.index())?,
+            Downcast(_, variant) => self.operand_downcast(base, variant)?,
+            Deref => self.deref_operand(base)?.into(),
+            Index(local) => {
+                let layout = self.layout_of(self.tcx.types.usize)?;
+                let n = self.local_to_op(self.frame(), local, Some(layout))?;
+                let n = self.read_scalar(&n)?.to_machine_usize(self)?;
+                self.operand_index(base, n)?
+            }
+            ConstantIndex { offset, min_length, from_end } => {
+                self.operand_constant_index(base, offset, min_length, from_end)?
+            }
+            Subslice { from, to, from_end } => self.operand_subslice(base, from, to, from_end)?,
+        })
+    }
+}

compiler/rustc_const_eval/src/interpret/terminator.rs

@@ -529,7 +529,7 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> InterpCx<'mir, 'tcx, M> {
                 let receiver_place = loop {
                     match receiver.layout.ty.kind() {
                         ty::Ref(..) | ty::RawPtr(..) => break self.deref_operand(&receiver)?,
-                        ty::Dynamic(..) => break receiver.assert_mem_place(),
+                        ty::Dynamic(..) => break receiver.assert_mem_place(), // no immediate unsized values
                         _ => {
                             // Not there yet, search for the only non-ZST field.
                             let mut non_zst_field = None;

compiler/rustc_const_eval/src/interpret/validity.rs

@@ -847,6 +847,8 @@ impl<'rt, 'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> ValueVisitor<'mir, 'tcx, M>
                 );
             }
             Abi::Scalar(scalar_layout) => {
+                // We use a 'forced' read because we always need a `Immediate` here
+                // and treating "partially uninit" as "fully uninit" is fine for us.
                 let scalar = self.read_immediate_forced(op)?.to_scalar_or_uninit();
                 self.visit_scalar(scalar, scalar_layout)?;
             }
@@ -856,6 +858,8 @@ impl<'rt, 'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> ValueVisitor<'mir, 'tcx, M>
                 // is subtle due to enums having ScalarPair layout, where one field
                 // is the discriminant.
                 if cfg!(debug_assertions) {
+                    // We use a 'forced' read because we always need a `Immediate` here
+                    // and treating "partially uninit" as "fully uninit" is fine for us.
                     let (a, b) = self.read_immediate_forced(op)?.to_scalar_or_uninit_pair();
                     self.visit_scalar(a, a_layout)?;
                     self.visit_scalar(b, b_layout)?;
@@ -880,7 +884,7 @@ impl<'rt, 'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> ValueVisitor<'mir, 'tcx, M>
     ) -> InterpResult<'tcx> {
         match op.layout.ty.kind() {
             ty::Str => {
-                let mplace = op.assert_mem_place(); // strings are never immediate
+                let mplace = op.assert_mem_place(); // strings are unsized and hence never immediate
                 let len = mplace.len(self.ecx)?;
                 try_validation!(
                     self.ecx.read_bytes_ptr(mplace.ptr, Size::from_bytes(len)),
@@ -900,14 +904,27 @@ impl<'rt, 'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> ValueVisitor<'mir, 'tcx, M>
             {
                 // Optimized handling for arrays of integer/float type.
 
-                // Arrays cannot be immediate, slices are never immediate.
-                let mplace = op.assert_mem_place();
                 // This is the length of the array/slice.
-                let len = mplace.len(self.ecx)?;
+                let len = op.len(self.ecx)?;
                 // This is the element type size.
                 let layout = self.ecx.layout_of(*tys)?;
                 // This is the size in bytes of the whole array. (This checks for overflow.)
                 let size = layout.size * len;
+                // If the size is 0, there is nothing to check.
+                // (`size` can only be 0 of `len` is 0, and empty arrays are always valid.)
+                if size == Size::ZERO {
+                    return Ok(());
+                }
+                // Now that we definitely have a non-ZST array, we know it lives in memory.
+                let mplace = match op.try_as_mplace() {
+                    Ok(mplace) => mplace,
+                    Err(imm) => match *imm {
+                        Immediate::Uninit =>
+                            throw_validation_failure!(self.path, { "uninitialized bytes" }),
+                        Immediate::Scalar(..) | Immediate::ScalarPair(..) =>
+                            bug!("arrays/slices can never have Scalar/ScalarPair layout"),
+                    }
+                };
 
                 // Optimization: we just check the entire range at once.
                 // NOTE: Keep this in sync with the handling of integer and float
@@ -919,10 +936,7 @@ impl<'rt, 'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> ValueVisitor<'mir, 'tcx, M>
                 // to reject those pointers, we just do not have the machinery to
                 // talk about parts of a pointer.
                 // We also accept uninit, for consistency with the slow path.
-                let Some(alloc) = self.ecx.get_ptr_alloc(mplace.ptr, size, mplace.align)? else {
-                    // Size 0, nothing more to check.
-                    return Ok(());
-                };
+                let alloc = self.ecx.get_ptr_alloc(mplace.ptr, size, mplace.align)?.expect("we already excluded size 0");
 
                 match alloc.check_bytes(
                     alloc_range(Size::ZERO, size),

compiler/rustc_const_eval/src/interpret/visitor.rs

@@ -21,8 +21,10 @@ pub trait Value<'mir, 'tcx, M: Machine<'mir, 'tcx>>: Copy {
     fn to_op(&self, ecx: &InterpCx<'mir, 'tcx, M>)
     -> InterpResult<'tcx, OpTy<'tcx, M::PointerTag>>;
 
-    /// Creates this from an `MPlaceTy`.
-    fn from_mem_place(mplace: MPlaceTy<'tcx, M::PointerTag>) -> Self;
+    /// Creates this from an `OpTy`.
+    ///
+    /// If `to_op` only ever produces `Indirect` operands, then this one is definitely `Indirect`.
+    fn from_op(mplace: OpTy<'tcx, M::PointerTag>) -> Self;
 
     /// Projects to the given enum variant.
     fn project_downcast(
@@ -56,8 +58,8 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> Value<'mir, 'tcx, M> for OpTy<'tc
     }
 
     #[inline(always)]
-    fn from_mem_place(mplace: MPlaceTy<'tcx, M::PointerTag>) -> Self {
-        mplace.into()
+    fn from_op(op: OpTy<'tcx, M::PointerTag>) -> Self {
+        op
     }
 
     #[inline(always)]
@@ -96,8 +98,9 @@ impl<'mir, 'tcx: 'mir, M: Machine<'mir, 'tcx>> Value<'mir, 'tcx, M>
     }
 
     #[inline(always)]
-    fn from_mem_place(mplace: MPlaceTy<'tcx, M::PointerTag>) -> Self {
-        mplace
+    fn from_op(op: OpTy<'tcx, M::PointerTag>) -> Self {
+        // assert is justified because our `to_op` only ever produces `Indirect` operands.
+        op.assert_mem_place()
     }
 
     #[inline(always)]
@@ -218,13 +221,13 @@ macro_rules! make_value_visitor {
                 match *v.layout().ty.kind() {
                     // If it is a trait object, switch to the real type that was used to create it.
                     ty::Dynamic(..) => {
-                        // immediate trait objects are not a thing
+                        // unsized values are never immediate, so we can assert_mem_place
                         let op = v.to_op(self.ecx())?;
                         let dest = op.assert_mem_place();
                         let inner = self.ecx().unpack_dyn_trait(&dest)?.1;
                         trace!("walk_value: dyn object layout: {:#?}", inner.layout);
                         // recurse with the inner type
-                        return self.visit_field(&v, 0, &Value::from_mem_place(inner));
+                        return self.visit_field(&v, 0, &Value::from_op(inner.into()));
                     },
                     // Slices do not need special handling here: they have `Array` field
                     // placement with length 0, so we enter the `Array` case below which
@@ -292,13 +295,12 @@ macro_rules! make_value_visitor {
                     FieldsShape::Array { .. } => {
                         // Let's get an mplace first.
                         let op = v.to_op(self.ecx())?;
-                        let mplace = op.assert_mem_place();
                         // Now we can go over all the fields.
                         // This uses the *run-time length*, i.e., if we are a slice,
                         // the dynamic info from the metadata is used.
-                        let iter = self.ecx().mplace_array_fields(&mplace)?
+                        let iter = self.ecx().operand_array_fields(&op)?
                             .map(|f| f.and_then(|f| {
-                                Ok(Value::from_mem_place(f))
+                                Ok(Value::from_op(f))
                             }));
                         self.visit_aggregate(v, iter)?;
                     }