Give an AllocId to ConstValue::Slice.

Author: cjgillot

compiler/rustc_codegen_cranelift/src/constant.rs

@@ -210,8 +210,7 @@ pub(crate) fn codegen_const_value<'tcx>(
                 .offset_i64(fx, i64::try_from(offset.bytes()).unwrap()),
             layout,
         ),
-        ConstValue::Slice { data, meta } => {
-            let alloc_id = fx.tcx.reserve_and_set_memory_alloc(data);
+        ConstValue::Slice { alloc_id, meta, phantom: _ } => {
             let ptr = pointer_for_allocation(fx, alloc_id).get_addr(fx);
             let len = fx.bcx.ins().iconst(fx.pointer_type, meta as i64);
             CValue::by_val_pair(ptr, len, layout)

compiler/rustc_codegen_ssa/src/mir/operand.rs

@@ -154,14 +154,11 @@ impl<'a, 'tcx, V: CodegenObject> OperandRef<'tcx, V> {
                 OperandValue::Immediate(llval)
             }
             ConstValue::ZeroSized => return OperandRef::zero_sized(layout),
-            ConstValue::Slice { data, meta } => {
+            ConstValue::Slice { alloc_id, meta, phantom: _ } => {
                 let BackendRepr::ScalarPair(a_scalar, _) = layout.backend_repr else {
                     bug!("from_const: invalid ScalarPair layout: {:#?}", layout);
                 };
-                let a = Scalar::from_pointer(
-                    Pointer::new(bx.tcx().reserve_and_set_memory_alloc(data).into(), Size::ZERO),
-                    &bx.tcx(),
-                );
+                let a = Scalar::from_pointer(Pointer::new(alloc_id.into(), Size::ZERO), &bx.tcx());
                 let a_llval = bx.scalar_to_backend(
                     a,
                     a_scalar,

compiler/rustc_const_eval/src/const_eval/eval_queries.rs

@@ -241,10 +241,9 @@ pub(super) fn op_to_const<'tcx>(
                 let (prov, offset) =
                     ptr.into_pointer_or_addr().expect(msg).prov_and_relative_offset();
                 let alloc_id = prov.alloc_id();
-                let data = ecx.tcx.global_alloc(alloc_id).unwrap_memory();
                 assert!(offset == abi::Size::ZERO, "{}", msg);
                 let meta = b.to_target_usize(ecx).expect(msg);
-                ConstValue::Slice { data, meta }
+                ConstValue::Slice { alloc_id, meta, phantom: std::marker::PhantomData }
             }
             Immediate::Uninit => bug!("`Uninit` is not a valid value for {}", op.layout.ty),
         },

compiler/rustc_const_eval/src/interpret/intrinsics.rs

@@ -3,10 +3,11 @@
 //! and miri.
 
 use std::assert_matches::assert_matches;
+use std::marker::PhantomData;
 
 use rustc_abi::{FieldIdx, HasDataLayout, Size};
 use rustc_apfloat::ieee::{Double, Half, Quad, Single};
-use rustc_middle::mir::interpret::{read_target_uint, write_target_uint};
+use rustc_middle::mir::interpret::{CTFE_ALLOC_SALT, read_target_uint, write_target_uint};
 use rustc_middle::mir::{self, BinOp, ConstValue, NonDivergingIntrinsic};
 use rustc_middle::ty::layout::TyAndLayout;
 use rustc_middle::ty::{Ty, TyCtxt};
@@ -17,17 +18,18 @@ use tracing::trace;
 use super::memory::MemoryKind;
 use super::util::ensure_monomorphic_enough;
 use super::{
-    Allocation, CheckInAllocMsg, ConstAllocation, ImmTy, InterpCx, InterpResult, Machine, OpTy,
-    PlaceTy, Pointer, PointerArithmetic, Provenance, Scalar, err_ub_custom, err_unsup_format,
-    interp_ok, throw_inval, throw_ub_custom, throw_ub_format,
+    AllocId, CheckInAllocMsg, ImmTy, InterpCx, InterpResult, Machine, OpTy, PlaceTy, Pointer,
+    PointerArithmetic, Provenance, Scalar, err_ub_custom, err_unsup_format, interp_ok, throw_inval,
+    throw_ub_custom, throw_ub_format,
 };
 use crate::fluent_generated as fluent;
 
 /// Directly returns an `Allocation` containing an absolute path representation of the given type.
-pub(crate) fn alloc_type_name<'tcx>(tcx: TyCtxt<'tcx>, ty: Ty<'tcx>) -> ConstAllocation<'tcx> {
+pub(crate) fn alloc_type_name<'tcx>(tcx: TyCtxt<'tcx>, ty: Ty<'tcx>) -> (AllocId, u64) {
     let path = crate::util::type_name(tcx, ty);
-    let alloc = Allocation::from_bytes_byte_aligned_immutable(path.into_bytes(), ());
-    tcx.mk_const_alloc(alloc)
+    let bytes = path.into_bytes();
+    let len = bytes.len().try_into().unwrap();
+    (tcx.allocate_bytes_dedup(bytes, CTFE_ALLOC_SALT), len)
 }
 impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
     /// Generates a value of `TypeId` for `ty` in-place.
@@ -126,8 +128,8 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
             sym::type_name => {
                 let tp_ty = instance.args.type_at(0);
                 ensure_monomorphic_enough(tcx, tp_ty)?;
-                let alloc = alloc_type_name(tcx, tp_ty);
-                let val = ConstValue::Slice { data: alloc, meta: alloc.inner().size().bytes() };
+                let (alloc_id, meta) = alloc_type_name(tcx, tp_ty);
+                let val = ConstValue::Slice { alloc_id, meta, phantom: PhantomData };
                 let val = self.const_val_to_op(val, dest.layout.ty, Some(dest.layout))?;
                 self.copy_op(&val, dest)?;
             }

compiler/rustc_const_eval/src/interpret/operand.rs

@@ -860,9 +860,8 @@ impl<'tcx, M: Machine<'tcx>> InterpCx<'tcx, M> {
             }
             mir::ConstValue::Scalar(x) => adjust_scalar(x)?.into(),
             mir::ConstValue::ZeroSized => Immediate::Uninit,
-            mir::ConstValue::Slice { data, meta } => {
+            mir::ConstValue::Slice { alloc_id, meta, phantom: _ } => {
                 // This is const data, no mutation allowed.
-                let alloc_id = self.tcx.reserve_and_set_memory_alloc(data);
                 let ptr = Pointer::new(CtfeProvenance::from(alloc_id).as_immutable(), Size::ZERO);
                 Immediate::new_slice(self.global_root_pointer(ptr)?.into(), meta, self)
             }

compiler/rustc_middle/src/mir/consts.rs

@@ -9,9 +9,7 @@ use rustc_span::{DUMMY_SP, Span, Symbol};
 use rustc_type_ir::TypeVisitableExt;
 
 use super::interpret::ReportedErrorInfo;
-use crate::mir::interpret::{
-    AllocId, AllocRange, ConstAllocation, ErrorHandled, GlobalAlloc, Scalar, alloc_range,
-};
+use crate::mir::interpret::{AllocId, AllocRange, ErrorHandled, GlobalAlloc, Scalar, alloc_range};
 use crate::mir::{Promoted, pretty_print_const_value};
 use crate::ty::print::{pretty_print_const, with_no_trimmed_paths};
 use crate::ty::{self, ConstKind, GenericArgsRef, ScalarInt, Ty, TyCtxt};
@@ -32,8 +30,8 @@ pub struct ConstAlloc<'tcx> {
 
 /// Represents a constant value in Rust. `Scalar` and `Slice` are optimizations for
 /// array length computations, enum discriminants and the pattern matching logic.
-#[derive(Copy, Clone, Debug, Eq, PartialEq, TyEncodable, TyDecodable, Hash)]
-#[derive(HashStable, Lift)]
+#[derive(Copy, Clone, Debug, Eq, PartialEq, TyEncodable, TyDecodable, Lift, Hash)]
+#[derive(HashStable)]
 pub enum ConstValue<'tcx> {
     /// Used for types with `layout::abi::Scalar` ABI.
     ///
@@ -52,10 +50,11 @@ pub enum ConstValue<'tcx> {
     Slice {
         /// The allocation storing the slice contents.
         /// This always points to the beginning of the allocation.
-        data: ConstAllocation<'tcx>,
+        alloc_id: AllocId,
         /// The metadata field of the reference.
         /// This is a "target usize", so we use `u64` as in the interpreter.
         meta: u64,
+        phantom: std::marker::PhantomData<&'tcx ()>,
     },
 
     /// A value not representable by the other variants; needs to be stored in-memory.
@@ -77,7 +76,7 @@ pub enum ConstValue<'tcx> {
 #[cfg(target_pointer_width = "64")]
 rustc_data_structures::static_assert_size!(ConstValue<'_>, 24);
 
-impl<'tcx> ConstValue<'tcx> {
+impl ConstValue<'_> {
     #[inline]
     pub fn try_to_scalar(&self) -> Option<Scalar> {
         match *self {
@@ -98,11 +97,11 @@ impl<'tcx> ConstValue<'tcx> {
         self.try_to_scalar_int()?.try_into().ok()
     }
 
-    pub fn try_to_target_usize(&self, tcx: TyCtxt<'tcx>) -> Option<u64> {
+    pub fn try_to_target_usize(&self, tcx: TyCtxt<'_>) -> Option<u64> {
         Some(self.try_to_scalar_int()?.to_target_usize(tcx))
     }
 
-    pub fn try_to_bits_for_ty(
+    pub fn try_to_bits_for_ty<'tcx>(
         &self,
         tcx: TyCtxt<'tcx>,
         typing_env: ty::TypingEnv<'tcx>,
@@ -132,12 +131,15 @@ impl<'tcx> ConstValue<'tcx> {
     }
 
     /// Must only be called on constants of type `&str` or `&[u8]`!
-    pub fn try_get_slice_bytes_for_diagnostics(&self, tcx: TyCtxt<'tcx>) -> Option<&'tcx [u8]> {
-        let (data, start, end) = match self {
+    pub fn try_get_slice_bytes_for_diagnostics<'tcx>(
+        &self,
+        tcx: TyCtxt<'tcx>,
+    ) -> Option<&'tcx [u8]> {
+        let (alloc_id, start, len) = match self {
             ConstValue::Scalar(_) | ConstValue::ZeroSized => {
                 bug!("`try_get_slice_bytes` on non-slice constant")
             }
-            &ConstValue::Slice { data, meta } => (data, 0, meta),
+            &ConstValue::Slice { alloc_id, meta, phantom: _ } => (alloc_id, 0, meta),
             &ConstValue::Indirect { alloc_id, offset } => {
                 // The reference itself is stored behind an indirection.
                 // Load the reference, and then load the actual slice contents.
@@ -170,26 +172,29 @@ impl<'tcx> ConstValue<'tcx> {
                 // Non-empty slice, must have memory. We know this is a relative pointer.
                 let (inner_prov, offset) =
                     ptr.into_pointer_or_addr().ok()?.prov_and_relative_offset();
-                let data = tcx.global_alloc(inner_prov.alloc_id()).unwrap_memory();
-                (data, offset.bytes(), offset.bytes() + len)
+                (inner_prov.alloc_id(), offset.bytes(), len)
             }
         };
 
+        let data = tcx.global_alloc(alloc_id).unwrap_memory();
+
         // This is for diagnostics only, so we are okay to use `inspect_with_uninit_and_ptr_outside_interpreter`.
         let start = start.try_into().unwrap();
-        let end = end.try_into().unwrap();
+        let end = start + usize::try_from(len).unwrap();
         Some(data.inner().inspect_with_uninit_and_ptr_outside_interpreter(start..end))
     }
 
     /// Check if a constant may contain provenance information. This is used by MIR opts.
     /// Can return `true` even if there is no provenance.
-    pub fn may_have_provenance(&self, tcx: TyCtxt<'tcx>, size: Size) -> bool {
+    pub fn may_have_provenance(&self, tcx: TyCtxt<'_>, size: Size) -> bool {
         match *self {
             ConstValue::ZeroSized | ConstValue::Scalar(Scalar::Int(_)) => return false,
             ConstValue::Scalar(Scalar::Ptr(..)) => return true,
             // It's hard to find out the part of the allocation we point to;
             // just conservatively check everything.
-            ConstValue::Slice { data, meta: _ } => !data.inner().provenance().ptrs().is_empty(),
+            ConstValue::Slice { alloc_id, meta: _, phantom: _ } => {
+                !tcx.global_alloc(alloc_id).unwrap_memory().inner().provenance().ptrs().is_empty()
+            }
             ConstValue::Indirect { alloc_id, offset } => !tcx
                 .global_alloc(alloc_id)
                 .unwrap_memory()
@@ -200,7 +205,7 @@ impl<'tcx> ConstValue<'tcx> {
     }
 
     /// Check if a constant only contains uninitialized bytes.
-    pub fn all_bytes_uninit(&self, tcx: TyCtxt<'tcx>) -> bool {
+    pub fn all_bytes_uninit(&self, tcx: TyCtxt<'_>) -> bool {
         let ConstValue::Indirect { alloc_id, .. } = self else {
             return false;
         };
@@ -487,9 +492,8 @@ impl<'tcx> Const<'tcx> {
     /// taking into account even pointer identity tests.
     pub fn is_deterministic(&self) -> bool {
         // Some constants may generate fresh allocations for pointers they contain,
-        // so using the same constant twice can yield two different results:
-        // - valtrees purposefully generate new allocations
-        // - ConstValue::Slice also generate new allocations
+        // so using the same constant twice can yield two different results.
+        // Notably, valtrees purposefully generate new allocations.
         match self {
             Const::Ty(_, c) => match c.kind() {
                 ty::ConstKind::Param(..) => true,
@@ -507,11 +511,11 @@ impl<'tcx> Const<'tcx> {
                 | ty::ConstKind::Placeholder(..) => bug!(),
             },
             Const::Unevaluated(..) => false,
-            // If the same slice appears twice in the MIR, we cannot guarantee that we will
-            // give the same `AllocId` to the data.
-            Const::Val(ConstValue::Slice { .. }, _) => false,
             Const::Val(
-                ConstValue::ZeroSized | ConstValue::Scalar(_) | ConstValue::Indirect { .. },
+                ConstValue::Slice { .. }
+                | ConstValue::ZeroSized
+                | ConstValue::Scalar(_)
+                | ConstValue::Indirect { .. },
                 _,
             ) => true,
         }

compiler/rustc_middle/src/mir/pretty.rs

@@ -1567,11 +1567,7 @@ pub fn write_allocations<'tcx>(
             ConstValue::Scalar(interpret::Scalar::Ptr(ptr, _)) => Some(ptr.provenance.alloc_id()),
             ConstValue::Scalar(interpret::Scalar::Int { .. }) => None,
             ConstValue::ZeroSized => None,
-            ConstValue::Slice { .. } => {
-                // `u8`/`str` slices, shouldn't contain pointers that we want to print.
-                None
-            }
-            ConstValue::Indirect { alloc_id, .. } => {
+            ConstValue::Slice { alloc_id, .. } | ConstValue::Indirect { alloc_id, .. } => {
                 // FIXME: we don't actually want to print all of these, since some are printed nicely directly as values inline in MIR.
                 // Really we'd want `pretty_print_const_value` to decide which allocations to print, instead of having a separate visitor.
                 Some(alloc_id)

compiler/rustc_middle/src/ty/context.rs

@@ -5,7 +5,7 @@
 pub mod tls;
 
 use std::assert_matches::debug_assert_matches;
-use std::borrow::Borrow;
+use std::borrow::{Borrow, Cow};
 use std::cmp::Ordering;
 use std::env::VarError;
 use std::ffi::OsStr;
@@ -1625,7 +1625,11 @@ impl<'tcx> TyCtxt<'tcx> {
 
     /// Allocates a read-only byte or string literal for `mir::interpret` with alignment 1.
     /// Returns the same `AllocId` if called again with the same bytes.
-    pub fn allocate_bytes_dedup(self, bytes: &[u8], salt: usize) -> interpret::AllocId {
+    pub fn allocate_bytes_dedup<'a>(
+        self,
+        bytes: impl Into<Cow<'a, [u8]>>,
+        salt: usize,
+    ) -> interpret::AllocId {
         // Create an allocation that just contains these bytes.
         let alloc = interpret::Allocation::from_bytes_byte_aligned_immutable(bytes, ());
         let alloc = self.mk_const_alloc(alloc);

compiler/rustc_middle/src/ty/structural_impls.rs

@@ -4,6 +4,7 @@
 //! to help with the tedium.
 
 use std::fmt::{self, Debug};
+use std::marker::PhantomData;
 
 use rustc_abi::TyAndLayout;
 use rustc_hir::def::Namespace;
@@ -311,6 +312,13 @@ TrivialTypeTraversalAndLiftImpls! {
 ///////////////////////////////////////////////////////////////////////////
 // Lift implementations
 
+impl<'tcx> Lift<TyCtxt<'tcx>> for PhantomData<&()> {
+    type Lifted = PhantomData<&'tcx ()>;
+    fn lift_to_interner(self, _: TyCtxt<'tcx>) -> Option<Self::Lifted> {
+        Some(PhantomData)
+    }
+}
+
 impl<'tcx, T: Lift<TyCtxt<'tcx>>> Lift<TyCtxt<'tcx>> for Option<T> {
     type Lifted = Option<T::Lifted>;
     fn lift_to_interner(self, tcx: TyCtxt<'tcx>) -> Option<Self::Lifted> {

compiler/rustc_mir_build/src/builder/expr/as_constant.rs

@@ -1,9 +1,11 @@
 //! See docs in build/expr/mod.rs
 
+use std::marker::PhantomData;
+
 use rustc_abi::Size;
 use rustc_ast as ast;
 use rustc_hir::LangItem;
-use rustc_middle::mir::interpret::{Allocation, CTFE_ALLOC_SALT, LitToConstInput, Scalar};
+use rustc_middle::mir::interpret::{CTFE_ALLOC_SALT, LitToConstInput, Scalar};
 use rustc_middle::mir::*;
 use rustc_middle::thir::*;
 use rustc_middle::ty::{
@@ -120,28 +122,41 @@ fn lit_to_mir_constant<'tcx>(tcx: TyCtxt<'tcx>, lit_input: LitToConstInput<'tcx>
 
     let value = match (lit, lit_ty.kind()) {
         (ast::LitKind::Str(s, _), ty::Ref(_, inner_ty, _)) if inner_ty.is_str() => {
-            let s = s.as_str();
-            let allocation = Allocation::from_bytes_byte_aligned_immutable(s.as_bytes(), ());
-            let allocation = tcx.mk_const_alloc(allocation);
-            ConstValue::Slice { data: allocation, meta: allocation.inner().size().bytes() }
+            let s = s.as_str().as_bytes();
+            let len = s.len();
+            let allocation = tcx.allocate_bytes_dedup(s, CTFE_ALLOC_SALT);
+            ConstValue::Slice {
+                alloc_id: allocation,
+                meta: len.try_into().unwrap(),
+                phantom: PhantomData,
+            }
         }
-        (ast::LitKind::ByteStr(data, _), ty::Ref(_, inner_ty, _))
+        (ast::LitKind::ByteStr(byte_sym, _), ty::Ref(_, inner_ty, _))
             if matches!(inner_ty.kind(), ty::Slice(_)) =>
         {
-            let allocation = Allocation::from_bytes_byte_aligned_immutable(data.as_byte_str(), ());
-            let allocation = tcx.mk_const_alloc(allocation);
-            ConstValue::Slice { data: allocation, meta: allocation.inner().size().bytes() }
+            let data = byte_sym.as_byte_str();
+            let len = data.len();
+            let allocation = tcx.allocate_bytes_dedup(data, CTFE_ALLOC_SALT);
+            ConstValue::Slice {
+                alloc_id: allocation,
+                meta: len.try_into().unwrap(),
+                phantom: PhantomData,
+            }
         }
         (ast::LitKind::ByteStr(byte_sym, _), ty::Ref(_, inner_ty, _)) if inner_ty.is_array() => {
             let id = tcx.allocate_bytes_dedup(byte_sym.as_byte_str(), CTFE_ALLOC_SALT);
             ConstValue::Scalar(Scalar::from_pointer(id.into(), &tcx))
         }
         (ast::LitKind::CStr(byte_sym, _), ty::Ref(_, inner_ty, _)) if matches!(inner_ty.kind(), ty::Adt(def, _) if tcx.is_lang_item(def.did(), LangItem::CStr)) =>
         {
-            let allocation =
-                Allocation::from_bytes_byte_aligned_immutable(byte_sym.as_byte_str(), ());
-            let allocation = tcx.mk_const_alloc(allocation);
-            ConstValue::Slice { data: allocation, meta: allocation.inner().size().bytes() }
+            let data = byte_sym.as_byte_str();
+            let len = data.len();
+            let allocation = tcx.allocate_bytes_dedup(data, CTFE_ALLOC_SALT);
+            ConstValue::Slice {
+                alloc_id: allocation,
+                meta: len.try_into().unwrap(),
+                phantom: PhantomData,
+            }
         }
         (ast::LitKind::Byte(n), ty::Uint(ty::UintTy::U8)) => {
             ConstValue::Scalar(Scalar::from_uint(n, Size::from_bytes(1)))