Auto merge of rust-lang#116564 - oli-obk:evaluated_static_in_metadata, r=<try>

Store static initializers in metadata instead of the MIR of statics.

This means that adding generic statics would be even more difficult, as we can't evaluate statics from other crates anymore, but the subtle issues I have encountered make me think that having this be an explicit problem is better. Said issues are:

### Nested allocations of static items get duplicated

which leads to issues for statics like

```rust
static mut FOO: &mut u32 = &mut 42;
static mut BAR = unsafe { FOO };
```

getting different allocations, instead of referring to the same one. This is also true for non-static mut, but promotion makes `static FOO: &u32 = &42;` annoying to demo.

### The main allocation of a static gets duplicated across crates

```rust
// crate a

static mut FOO: Option<u32> = Some(42);

// crate b

static mut BAR: &mut u32 = unsafe {
    match &mut a::FOO {
        Some(x) => x,
        None => panic!(),
    }
};
```

## Why is this being done?

In order to ensure all crates see the same nested allocations (which is the last issue that needs fixing before we can stabilize [`const_mut_refs`](rust-lang#57349)), I am working on creating anonymous (from the Rust side, to LLVM it's like a regular static item) static items for the nested allocations in a static. If we evaluate the static item in a downstream crate again, we will end up duplicating its nested allocations (and in some cases, like the `match` case, even duplicate the main allocation).

Author: bors

compiler/rustc_const_eval/src/lib.rs

@@ -46,6 +46,13 @@ pub fn provide(providers: &mut Providers) {
     const_eval::provide(providers);
     providers.eval_to_const_value_raw = const_eval::eval_to_const_value_raw_provider;
     providers.eval_to_allocation_raw = const_eval::eval_to_allocation_raw_provider;
+    providers.eval_static_initializer_raw = |tcx, def_id| {
+        assert!(tcx.is_static(def_id.to_def_id()));
+        let instance = ty::Instance::mono(tcx, def_id.to_def_id());
+        let gid = rustc_middle::mir::interpret::GlobalId { instance, promoted: None };
+        let param_env = ty::ParamEnv::reveal_all();
+        Ok(tcx.eval_to_allocation_raw(param_env.and(gid))?.alloc_id)
+    };
     providers.const_caller_location = const_eval::const_caller_location;
     providers.eval_to_valtree = |tcx, param_env_and_value| {
         let (param_env, raw) = param_env_and_value.into_parts();

compiler/rustc_metadata/src/rmeta/decoder/cstore_impl.rs

@@ -242,6 +242,15 @@ provide! { tcx, def_id, other, cdata,
     asyncness => { table_direct }
     fn_arg_names => { table }
     generator_kind => { table }
+    eval_static_initializer_raw => {
+        Ok(cdata
+            .root
+            .tables
+            .eval_static_initializer_raw
+            .get(cdata, def_id.index)
+            .map(|lazy| lazy.decode((cdata, tcx)))
+            .unwrap_or_else(|| panic!("{def_id:?} does not have eval_static_initializer_raw")))
+    }
     trait_def => { table }
     deduced_param_attrs => { table }
     is_type_alias_impl_trait => {

compiler/rustc_metadata/src/rmeta/encoder.rs

@@ -1034,11 +1034,9 @@ fn should_encode_mir(
             (true, mir_opt_base)
         }
         // Constants
-        DefKind::AnonConst
-        | DefKind::InlineConst
-        | DefKind::AssocConst
-        | DefKind::Static(..)
-        | DefKind::Const => (true, false),
+        DefKind::AnonConst | DefKind::InlineConst | DefKind::AssocConst | DefKind::Const => {
+            (true, false)
+        }
         // Full-fledged functions + closures
         DefKind::AssocFn | DefKind::Fn | DefKind::Closure => {
             let generics = tcx.generics_of(def_id);
@@ -1439,6 +1437,12 @@ impl<'a, 'tcx> EncodeContext<'a, 'tcx> {
                 let data = self.tcx.generator_kind(def_id).unwrap();
                 record!(self.tables.generator_kind[def_id] <- data);
             }
+            if let DefKind::Static(_) = def_kind {
+                if !self.tcx.is_foreign_item(def_id) {
+                    let data = self.tcx.eval_static_initializer_raw(def_id).unwrap();
+                    record!(self.tables.eval_static_initializer_raw[def_id] <- data);
+                }
+            }
             if let DefKind::Enum | DefKind::Struct | DefKind::Union = def_kind {
                 self.encode_info_for_adt(local_id);
             }

compiler/rustc_metadata/src/rmeta/mod.rs

@@ -442,6 +442,7 @@ define_tables! {
     asyncness: Table<DefIndex, ty::Asyncness>,
     fn_arg_names: Table<DefIndex, LazyArray<Ident>>,
     generator_kind: Table<DefIndex, LazyValue<hir::GeneratorKind>>,
+    eval_static_initializer_raw: Table<DefIndex, LazyValue<mir::interpret::AllocId>>,
     trait_def: Table<DefIndex, LazyValue<ty::TraitDef>>,
     trait_item_def_id: Table<DefIndex, RawDefId>,
     expn_that_defined: Table<DefIndex, LazyValue<ExpnId>>,

compiler/rustc_middle/src/mir/interpret/error.rs

@@ -85,6 +85,7 @@ impl Into<ErrorGuaranteed> for ReportedErrorInfo {
 TrivialTypeTraversalImpls! { ErrorHandled }
 
 pub type EvalToAllocationRawResult<'tcx> = Result<ConstAlloc<'tcx>, ErrorHandled>;
+pub type EvalStaticInitializerRawResult = Result<AllocId, ErrorHandled>;
 pub type EvalToConstValueResult<'tcx> = Result<ConstValue<'tcx>, ErrorHandled>;
 /// `Ok(None)` indicates the constant was fine, but the valtree couldn't be constructed.
 /// This is needed in `thir::pattern::lower_inline_const`.

compiler/rustc_middle/src/mir/interpret/mod.rs

@@ -142,11 +142,11 @@ use crate::ty::GenericArgKind;
 use crate::ty::{self, Instance, Ty, TyCtxt};
 
 pub use self::error::{
-    struct_error, BadBytesAccess, CheckInAllocMsg, ErrorHandled, EvalToAllocationRawResult,
-    EvalToConstValueResult, EvalToValTreeResult, ExpectedKind, InterpError, InterpErrorInfo,
-    InterpResult, InvalidMetaKind, InvalidProgramInfo, MachineStopType, PointerKind,
-    ReportedErrorInfo, ResourceExhaustionInfo, ScalarSizeMismatch, UndefinedBehaviorInfo,
-    UnsupportedOpInfo, ValidationErrorInfo, ValidationErrorKind,
+    struct_error, BadBytesAccess, CheckInAllocMsg, ErrorHandled, EvalStaticInitializerRawResult,
+    EvalToAllocationRawResult, EvalToConstValueResult, EvalToValTreeResult, ExpectedKind,
+    InterpError, InterpErrorInfo, InterpResult, InvalidMetaKind, InvalidProgramInfo,
+    MachineStopType, PointerKind, ReportedErrorInfo, ResourceExhaustionInfo, ScalarSizeMismatch,
+    UndefinedBehaviorInfo, UnsupportedOpInfo, ValidationErrorInfo, ValidationErrorKind,
 };
 
 pub use self::value::Scalar;

compiler/rustc_middle/src/mir/interpret/queries.rs

@@ -194,22 +194,8 @@ impl<'tcx> TyCtxtAt<'tcx> {
     ) -> Result<mir::ConstAllocation<'tcx>, ErrorHandled> {
         trace!("eval_static_initializer: Need to compute {:?}", def_id);
         assert!(self.is_static(def_id));
-        let instance = ty::Instance::mono(*self, def_id);
-        let gid = GlobalId { instance, promoted: None };
-        self.eval_to_allocation(gid, ty::ParamEnv::reveal_all())
-    }
-
-    /// Evaluate anything constant-like, returning the allocation of the final memory.
-    ///
-    /// The span is entirely ignored here, but still helpful for better query cycle errors.
-    fn eval_to_allocation(
-        self,
-        gid: GlobalId<'tcx>,
-        param_env: ty::ParamEnv<'tcx>,
-    ) -> Result<mir::ConstAllocation<'tcx>, ErrorHandled> {
-        trace!("eval_to_allocation: Need to compute {:?}", gid);
-        let raw_const = self.eval_to_allocation_raw(param_env.and(gid))?;
-        Ok(self.global_alloc(raw_const.alloc_id).unwrap_memory())
+        let alloc_id = self.eval_static_initializer_raw(def_id)?;
+        Ok(self.global_alloc(alloc_id).unwrap_memory())
     }
 }
 
@@ -237,10 +223,6 @@ impl<'tcx> TyCtxtEnsure<'tcx> {
     pub fn eval_static_initializer(self, def_id: DefId) {
         trace!("eval_static_initializer: Need to compute {:?}", def_id);
         assert!(self.tcx.is_static(def_id));
-        let instance = ty::Instance::mono(self.tcx, def_id);
-        let gid = GlobalId { instance, promoted: None };
-        let param_env = ty::ParamEnv::reveal_all();
-        trace!("eval_to_allocation: Need to compute {:?}", gid);
-        self.eval_to_allocation_raw(param_env.and(gid))
+        self.eval_static_initializer_raw(def_id);
     }
 }

compiler/rustc_middle/src/query/erase.rs

@@ -256,6 +256,7 @@ trivial! {
     rustc_middle::mir::interpret::AllocId,
     rustc_middle::mir::interpret::ErrorHandled,
     rustc_middle::mir::interpret::LitToConstError,
+    rustc_middle::mir::interpret::EvalStaticInitializerRawResult,
     rustc_middle::thir::ExprId,
     rustc_middle::traits::CodegenObligationError,
     rustc_middle::traits::EvaluationResult,

compiler/rustc_middle/src/query/mod.rs

@@ -20,7 +20,8 @@ use crate::middle::stability::{self, DeprecationEntry};
 use crate::mir;
 use crate::mir::interpret::GlobalId;
 use crate::mir::interpret::{
-    EvalToAllocationRawResult, EvalToConstValueResult, EvalToValTreeResult,
+    EvalStaticInitializerRawResult, EvalToAllocationRawResult, EvalToConstValueResult,
+    EvalToValTreeResult,
 };
 use crate::mir::interpret::{LitToConstError, LitToConstInput};
 use crate::mir::mono::CodegenUnit;
@@ -1073,6 +1074,16 @@ rustc_queries! {
         cache_on_disk_if { true }
     }
 
+    /// Evaluate a static's initializer, returning the allocation of the initializer's memory.
+    query eval_static_initializer_raw(key: DefId) -> EvalStaticInitializerRawResult {
+        desc { |tcx|
+            "evaluating initializer of static `{}`",
+            tcx.def_path_str(key)
+        }
+        cache_on_disk_if { true }
+        separate_provide_extern
+    }
+
     /// Evaluates const items or anonymous constants
     /// (such as enum variant explicit discriminants or array lengths)
     /// into a representation suitable for the type system and const generics.

compiler/rustc_middle/src/ty/parameterized.rs

@@ -61,6 +61,7 @@ trivially_parameterized_over_tcx! {
     crate::middle::exported_symbols::SymbolExportInfo,
     crate::middle::resolve_bound_vars::ObjectLifetimeDefault,
     crate::mir::ConstQualifs,
+    crate::mir::interpret::AllocId,
     ty::AssocItemContainer,
     ty::Asyncness,
     ty::DeducedParamAttrs,