Merge branch 'rust-lang:master' into master

Author: YingkaiLi-VM

compiler/rustc_abi/src/canon_abi.rs

@@ -0,0 +1,136 @@
+use std::fmt;
+
+#[cfg(feature = "nightly")]
+use rustc_macros::HashStable_Generic;
+
+use crate::ExternAbi;
+
+/// Calling convention to determine codegen
+///
+/// CanonAbi erases certain distinctions ExternAbi preserves, but remains target-dependent.
+/// There are still both target-specific variants and aliasing variants, though much fewer.
+/// The reason for this step is the frontend may wish to show an ExternAbi but implement that ABI
+/// using a different ABI than the string per se, or describe irrelevant differences, e.g.
+/// - extern "system"
+/// - extern "cdecl"
+/// - extern "C-unwind"
+/// In that sense, this erases mere syntactic distinctions to create a canonical *directive*,
+/// rather than picking the "actual" ABI.
+#[derive(Copy, Clone, Debug)]
+#[derive(PartialOrd, Ord, PartialEq, Eq, Hash)]
+#[cfg_attr(feature = "nightly", derive(HashStable_Generic))]
+pub enum CanonAbi {
+    // NOTE: the use of nested variants for some ABIs is for many targets they don't matter,
+    // and this pushes the complexity of their reasoning to target-specific code,
+    // allowing a `match` to easily exhaustively ignore these subcategories of variants.
+    // Otherwise it is very tempting to avoid matching exhaustively!
+    C,
+    Rust,
+    RustCold,
+
+    /// ABIs relevant to 32-bit Arm targets
+    Arm(ArmCall),
+    /// ABI relevant to GPUs: the entry point for a GPU kernel
+    GpuKernel,
+
+    /// ABIs relevant to bare-metal interrupt targets
+    // FIXME(workingjubilee): a particular reason for this nesting is we might not need these?
+    // interrupt ABIs should have the same properties:
+    // - uncallable by Rust calls, as LLVM rejects it in most cases
+    // - uses a preserve-all-registers *callee* convention
+    // - should always return `-> !` (effectively... it can't use normal `ret`)
+    // what differs between targets is
+    // - allowed arguments: x86 differs slightly, having 2-3 arguments which are handled magically
+    // - may need special prologues/epilogues for some interrupts, without affecting "call ABI"
+    Interrupt(InterruptKind),
+
+    /// ABIs relevant to Windows or x86 targets
+    X86(X86Call),
+}
+
+impl fmt::Display for CanonAbi {
+    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+        self.to_erased_extern_abi().as_str().fmt(f)
+    }
+}
+
+impl CanonAbi {
+    /// convert to the ExternAbi that *shares a string* with this CanonAbi
+    ///
+    /// A target-insensitive mapping of CanonAbi to ExternAbi, convenient for "forwarding" impls.
+    /// Importantly, the set of CanonAbi values is a logical *subset* of ExternAbi values,
+    /// so this is injective: if you take an ExternAbi to a CanonAbi and back, you have lost data.
+    const fn to_erased_extern_abi(self) -> ExternAbi {
+        match self {
+            CanonAbi::C => ExternAbi::C { unwind: false },
+            CanonAbi::Rust => ExternAbi::Rust,
+            CanonAbi::RustCold => ExternAbi::RustCold,
+            CanonAbi::Arm(arm_call) => match arm_call {
+                ArmCall::Aapcs => ExternAbi::Aapcs { unwind: false },
+                ArmCall::CCmseNonSecureCall => ExternAbi::CCmseNonSecureCall,
+                ArmCall::CCmseNonSecureEntry => ExternAbi::CCmseNonSecureEntry,
+            },
+            CanonAbi::GpuKernel => ExternAbi::GpuKernel,
+            CanonAbi::Interrupt(interrupt_kind) => match interrupt_kind {
+                InterruptKind::Avr => ExternAbi::AvrInterrupt,
+                InterruptKind::AvrNonBlocking => ExternAbi::AvrNonBlockingInterrupt,
+                InterruptKind::Msp430 => ExternAbi::Msp430Interrupt,
+                InterruptKind::RiscvMachine => ExternAbi::RiscvInterruptM,
+                InterruptKind::RiscvSupervisor => ExternAbi::RiscvInterruptS,
+                InterruptKind::X86 => ExternAbi::X86Interrupt,
+            },
+            CanonAbi::X86(x86_call) => match x86_call {
+                X86Call::Fastcall => ExternAbi::Fastcall { unwind: false },
+                X86Call::Stdcall => ExternAbi::Stdcall { unwind: false },
+                X86Call::SysV64 => ExternAbi::SysV64 { unwind: false },
+                X86Call::Thiscall => ExternAbi::Thiscall { unwind: false },
+                X86Call::Vectorcall => ExternAbi::Vectorcall { unwind: false },
+                X86Call::Win64 => ExternAbi::Win64 { unwind: false },
+            },
+        }
+    }
+}
+
+/// Callee codegen for interrupts
+///
+/// This is named differently from the "Call" enums because it is different:
+/// these "ABI" differences are not relevant to callers, since there is "no caller".
+/// These only affect callee codegen. making their categorization as distinct ABIs a bit peculiar.
+#[derive(Copy, Clone, Debug)]
+#[derive(PartialOrd, Ord, PartialEq, Eq, Hash)]
+#[cfg_attr(feature = "nightly", derive(HashStable_Generic))]
+pub enum InterruptKind {
+    Avr,
+    AvrNonBlocking,
+    Msp430,
+    RiscvMachine,
+    RiscvSupervisor,
+    X86,
+}
+
+/// ABIs defined for x86-{32,64}
+///
+/// One of SysV64 or Win64 may alias the C ABI, and arguably Win64 is cross-platform now?
+#[derive(Clone, Copy, Debug)]
+#[derive(PartialOrd, Ord, PartialEq, Eq, Hash)]
+#[cfg_attr(feature = "nightly", derive(HashStable_Generic))]
+pub enum X86Call {
+    /// "fastcall" has both GNU and Windows variants
+    Fastcall,
+    /// "stdcall" has both GNU and Windows variants
+    Stdcall,
+    SysV64,
+    Thiscall,
+    Vectorcall,
+    Win64,
+}
+
+/// ABIs defined for 32-bit Arm
+#[derive(Copy, Clone, Debug)]
+#[derive(PartialOrd, Ord, PartialEq, Eq, Hash)]
+#[cfg_attr(feature = "nightly", derive(HashStable_Generic))]
+pub enum ArmCall {
+    Aapcs,
+    CCmseNonSecureCall,
+    CCmseNonSecureEntry,
+}

compiler/rustc_abi/src/extern_abi.rs

@@ -7,6 +7,8 @@ use rustc_data_structures::stable_hasher::{HashStable, StableHasher, StableOrd};
 #[cfg(feature = "nightly")]
 use rustc_macros::{Decodable, Encodable};
 
+use crate::AbiFromStrErr;
+
 #[cfg(test)]
 mod tests;
 
@@ -99,11 +101,6 @@ macro_rules! abi_impls {
     }
 }
 
-#[derive(Debug)]
-pub enum AbiFromStrErr {
-    Unknown,
-}
-
 abi_impls! {
     ExternAbi = {
             C { unwind: false } =><= "C",

compiler/rustc_abi/src/layout.rs

@@ -758,7 +758,7 @@ impl<Cx: HasDataLayout> LayoutCalculator<Cx> {
                         niche_variants,
                         niche_start,
                     },
-                    tag_field: 0,
+                    tag_field: FieldIdx::new(0),
                     variants: IndexVec::new(),
                 },
                 fields: FieldsShape::Arbitrary {
@@ -1072,7 +1072,7 @@ impl<Cx: HasDataLayout> LayoutCalculator<Cx> {
             variants: Variants::Multiple {
                 tag,
                 tag_encoding: TagEncoding::Direct,
-                tag_field: 0,
+                tag_field: FieldIdx::new(0),
                 variants: IndexVec::new(),
             },
             fields: FieldsShape::Arbitrary {

compiler/rustc_abi/src/layout/coroutine.rs

@@ -158,7 +158,7 @@ pub(super) fn layout<
     // Build a prefix layout, including "promoting" all ineligible
     // locals as part of the prefix. We compute the layout of all of
     // these fields at once to get optimal packing.
-    let tag_index = prefix_layouts.len();
+    let tag_index = prefix_layouts.next_index();
 
     // `variant_fields` already accounts for the reserved variants, so no need to add them.
     let max_discr = (variant_fields.len() - 1) as u128;
@@ -187,7 +187,7 @@ pub(super) fn layout<
 
             // "a" (`0..b_start`) and "b" (`b_start..`) correspond to
             // "outer" and "promoted" fields respectively.
-            let b_start = FieldIdx::new(tag_index + 1);
+            let b_start = tag_index.plus(1);
             let offsets_b = IndexVec::from_raw(offsets.raw.split_off(b_start.index()));
             let offsets_a = offsets;
 

compiler/rustc_abi/src/lib.rs

@@ -55,13 +55,14 @@ use rustc_index::{Idx, IndexSlice, IndexVec};
 use rustc_macros::{Decodable_NoContext, Encodable_NoContext, HashStable_Generic};
 
 mod callconv;
+mod canon_abi;
+mod extern_abi;
 mod layout;
 #[cfg(test)]
 mod tests;
 
-mod extern_abi;
-
 pub use callconv::{Heterogeneous, HomogeneousAggregate, Reg, RegKind};
+pub use canon_abi::{ArmCall, CanonAbi, InterruptKind, X86Call};
 pub use extern_abi::{ExternAbi, all_names};
 #[cfg(feature = "nightly")]
 pub use layout::{FIRST_VARIANT, FieldIdx, Layout, TyAbiInterface, TyAndLayout, VariantIdx};
@@ -1572,7 +1573,7 @@ pub enum Variants<FieldIdx: Idx, VariantIdx: Idx> {
     Multiple {
         tag: Scalar,
         tag_encoding: TagEncoding<VariantIdx>,
-        tag_field: usize,
+        tag_field: FieldIdx,
         variants: IndexVec<VariantIdx, LayoutData<FieldIdx, VariantIdx>>,
     },
 }
@@ -1895,3 +1896,11 @@ pub enum StructKind {
     /// A univariant, but with a prefix of an arbitrary size & alignment (e.g., enum tag).
     Prefixed(Size, Align),
 }
+
+#[derive(Clone, Debug)]
+pub enum AbiFromStrErr {
+    /// not a known ABI
+    Unknown,
+    /// no "-unwind" variant can be used here
+    NoExplicitUnwind,
+}

compiler/rustc_ast/src/ast.rs

@@ -99,8 +99,15 @@ pub struct Path {
 
 impl PartialEq<Symbol> for Path {
     #[inline]
-    fn eq(&self, symbol: &Symbol) -> bool {
-        matches!(&self.segments[..], [segment] if segment.ident.name == *symbol)
+    fn eq(&self, name: &Symbol) -> bool {
+        if let [segment] = self.segments.as_ref()
+            && segment.args.is_none()
+            && segment.ident.name == *name
+        {
+            true
+        } else {
+            false
+        }
     }
 }
 
@@ -120,17 +127,6 @@ impl Path {
         Path { segments: thin_vec![PathSegment::from_ident(ident)], span: ident.span, tokens: None }
     }
 
-    pub fn is_ident(&self, name: Symbol) -> bool {
-        if let [segment] = self.segments.as_ref()
-            && segment.args.is_none()
-            && segment.ident.name == name
-        {
-            true
-        } else {
-            false
-        }
-    }
-
     pub fn is_global(&self) -> bool {
         self.segments.first().is_some_and(|segment| segment.ident.name == kw::PathRoot)
     }
@@ -2465,6 +2461,39 @@ impl TyKind {
             None
         }
     }
+
+    /// Returns `true` if this type is considered a scalar primitive (e.g.,
+    /// `i32`, `u8`, `bool`, etc).
+    ///
+    /// This check is based on **symbol equality** and does **not** remove any
+    /// path prefixes or references. If a type alias or shadowing is present
+    /// (e.g., `type i32 = CustomType;`), this method will still return `true`
+    /// for `i32`, even though it may not refer to the primitive type.
+    pub fn maybe_scalar(&self) -> bool {
+        let Some(ty_sym) = self.is_simple_path() else {
+            // unit type
+            return self.is_unit();
+        };
+        matches!(
+            ty_sym,
+            sym::i8
+                | sym::i16
+                | sym::i32
+                | sym::i64
+                | sym::i128
+                | sym::u8
+                | sym::u16
+                | sym::u32
+                | sym::u64
+                | sym::u128
+                | sym::f16
+                | sym::f32
+                | sym::f64
+                | sym::f128
+                | sym::char
+                | sym::bool
+        )
+    }
 }
 
 /// A pattern type pattern.

compiler/rustc_ast/src/tokenstream.rs

@@ -57,7 +57,9 @@ impl TokenTree {
         match (self, other) {
             (TokenTree::Token(token, _), TokenTree::Token(token2, _)) => token.kind == token2.kind,
             (TokenTree::Delimited(.., delim, tts), TokenTree::Delimited(.., delim2, tts2)) => {
-                delim == delim2 && tts.eq_unspanned(tts2)
+                delim == delim2
+                    && tts.len() == tts2.len()
+                    && tts.iter().zip(tts2.iter()).all(|(a, b)| a.eq_unspanned(b))
             }
             _ => false,
         }
@@ -694,18 +696,6 @@ impl TokenStream {
         TokenStreamIter::new(self)
     }
 
-    /// Compares two `TokenStream`s, checking equality without regarding span information.
-    pub fn eq_unspanned(&self, other: &TokenStream) -> bool {
-        let mut iter1 = self.iter();
-        let mut iter2 = other.iter();
-        for (tt1, tt2) in iter::zip(&mut iter1, &mut iter2) {
-            if !tt1.eq_unspanned(tt2) {
-                return false;
-            }
-        }
-        iter1.next().is_none() && iter2.next().is_none()
-    }
-
     /// Create a token stream containing a single token with alone spacing. The
     /// spacing used for the final token in a constructed stream doesn't matter
     /// because it's never used. In practice we arbitrarily use

compiler/rustc_ast/src/visit.rs

@@ -121,6 +121,10 @@ pub enum LifetimeCtxt {
 /// explicitly, you need to override each method. (And you also need
 /// to monitor future changes to `Visitor` in case a new method with a
 /// new default implementation gets introduced.)
+///
+/// Every `walk_*` method uses deconstruction to access fields of structs and
+/// enums. This will result in a compile error if a field is added, which makes
+/// it more likely the appropriate visit call will be added for it.
 pub trait Visitor<'ast>: Sized {
     /// The result type of the `visit_*` methods. Can be either `()`,
     /// or `ControlFlow<T>`.

compiler/rustc_ast_lowering/src/expr.rs

@@ -1,4 +1,3 @@
-use std::assert_matches::assert_matches;
 use std::ops::ControlFlow;
 use std::sync::Arc;
 
@@ -1199,11 +1198,13 @@ impl<'hir> LoweringContext<'_, 'hir> {
         let closure_def_id = self.local_def_id(closure_id);
         let (binder_clause, generic_params) = self.lower_closure_binder(binder);
 
-        assert_matches!(
-            coroutine_kind,
-            CoroutineKind::Async { .. },
-            "only async closures are supported currently"
-        );
+        let coroutine_desugaring = match coroutine_kind {
+            CoroutineKind::Async { .. } => hir::CoroutineDesugaring::Async,
+            CoroutineKind::Gen { .. } => hir::CoroutineDesugaring::Gen,
+            CoroutineKind::AsyncGen { span, .. } => {
+                span_bug!(span, "only async closures and `iter!` closures are supported currently")
+            }
+        };
 
         let body = self.with_new_scopes(fn_decl_span, |this| {
             let inner_decl =
@@ -1247,7 +1248,7 @@ impl<'hir> LoweringContext<'_, 'hir> {
             // Lower this as a `CoroutineClosure`. That will ensure that HIR typeck
             // knows that a `FnDecl` output type like `-> &str` actually means
             // "coroutine that returns &str", rather than directly returning a `&str`.
-            kind: hir::ClosureKind::CoroutineClosure(hir::CoroutineDesugaring::Async),
+            kind: hir::ClosureKind::CoroutineClosure(coroutine_desugaring),
             constness: hir::Constness::NotConst,
         });
         hir::ExprKind::Closure(c)

compiler/rustc_ast_passes/src/feature_gate.rs

@@ -477,11 +477,12 @@ pub fn check_crate(krate: &ast::Crate, sess: &Session, features: &Features) {
         for span in spans {
             if (!visitor.features.coroutines() && !span.allows_unstable(sym::coroutines))
                 && (!visitor.features.gen_blocks() && !span.allows_unstable(sym::gen_blocks))
+                && (!visitor.features.yield_expr() && !span.allows_unstable(sym::yield_expr))
             {
                 #[allow(rustc::untranslatable_diagnostic)]
-                // Don't know which of the two features to include in the
-                // error message, so I am arbitrarily picking one.
-                feature_err(&visitor.sess, sym::coroutines, *span, "yield syntax is experimental")
+                // Emit yield_expr as the error, since that will be sufficient. You can think of it
+                // as coroutines and gen_blocks imply yield_expr.
+                feature_err(&visitor.sess, sym::yield_expr, *span, "yield syntax is experimental")
                     .emit();
             }
         }