Auto merge of #135031 - RalfJung:intrinsics-without-body, r=oli-obk

rustc_intrinsic: support functions without body

We synthesize a HIR body `loop {}` but such bodyless intrinsics.

Most of the diff is due to turning `ItemKind::Fn` into a brace (named-field) enum variant, because it carries a `bool`-typed field now. This is to remember whether the function has a body. MIR building panics to avoid ever translating the fake `loop {}` body, and the intrinsic logic uses the lack of a body to implicitly mark that intrinsic as must-be-overridden.

I first tried actually having no body rather than generating the fake body, but there's a *lot* of code that assumes that all function items have HIR and MIR, so this didn't work very well. Then I noticed that even `rustc_intrinsic_must_be_overridden` intrinsics have MIR generated (they are filled with an `Unreachable` terminator) so I guess I am not the first to discover this. ;)

r? `@oli-obk`

Author: bors

compiler/rustc_ast_lowering/src/item.rs

@@ -222,6 +222,7 @@ impl<'hir> LoweringContext<'_, 'hir> {
                         decl,
                         coroutine_kind,
                         body.as_deref(),
+                        attrs,
                     );
 
                     let itctx = ImplTraitContext::Universal;
@@ -233,7 +234,7 @@ impl<'hir> LoweringContext<'_, 'hir> {
                         header: this.lower_fn_header(*header, hir::Safety::Safe),
                         span: this.lower_span(*fn_sig_span),
                     };
-                    hir::ItemKind::Fn(sig, generics, body_id)
+                    hir::ItemKind::Fn { sig, generics, body: body_id, has_body: body.is_some() }
                 })
             }
             ItemKind::Mod(_, mod_kind) => match mod_kind {
@@ -435,11 +436,12 @@ impl<'hir> LoweringContext<'_, 'hir> {
             }
             ItemKind::Delegation(box delegation) => {
                 let delegation_results = self.lower_delegation(delegation, id);
-                hir::ItemKind::Fn(
-                    delegation_results.sig,
-                    delegation_results.generics,
-                    delegation_results.body_id,
-                )
+                hir::ItemKind::Fn {
+                    sig: delegation_results.sig,
+                    generics: delegation_results.generics,
+                    body: delegation_results.body_id,
+                    has_body: true,
+                }
             }
             ItemKind::MacCall(..) | ItemKind::DelegationMac(..) => {
                 panic!("macros should have been expanded by now")
@@ -747,7 +749,7 @@ impl<'hir> LoweringContext<'_, 'hir> {
 
     fn lower_trait_item(&mut self, i: &AssocItem) -> &'hir hir::TraitItem<'hir> {
         let hir_id = hir::HirId::make_owner(self.current_hir_id_owner.def_id);
-        self.lower_attrs(hir_id, &i.attrs);
+        let attrs = self.lower_attrs(hir_id, &i.attrs);
         let trait_item_def_id = hir_id.expect_owner();
 
         let (generics, kind, has_default) = match &i.kind {
@@ -785,6 +787,7 @@ impl<'hir> LoweringContext<'_, 'hir> {
                     &sig.decl,
                     sig.header.coroutine_kind,
                     Some(body),
+                    attrs,
                 );
                 let (generics, sig) = self.lower_method_sig(
                     generics,
@@ -877,7 +880,7 @@ impl<'hir> LoweringContext<'_, 'hir> {
         let has_value = true;
         let (defaultness, _) = self.lower_defaultness(i.kind.defaultness(), has_value);
         let hir_id = hir::HirId::make_owner(self.current_hir_id_owner.def_id);
-        self.lower_attrs(hir_id, &i.attrs);
+        let attrs = self.lower_attrs(hir_id, &i.attrs);
 
         let (generics, kind) = match &i.kind {
             AssocItemKind::Const(box ConstItem { generics, ty, expr, .. }) => self.lower_generics(
@@ -900,6 +903,7 @@ impl<'hir> LoweringContext<'_, 'hir> {
                     &sig.decl,
                     sig.header.coroutine_kind,
                     body.as_deref(),
+                    attrs,
                 );
                 let (generics, sig) = self.lower_method_sig(
                     generics,
@@ -1054,20 +1058,8 @@ impl<'hir> LoweringContext<'_, 'hir> {
         })
     }
 
-    fn lower_fn_body_block(
-        &mut self,
-        span: Span,
-        decl: &FnDecl,
-        body: Option<&Block>,
-    ) -> hir::BodyId {
-        self.lower_fn_body(decl, |this| this.lower_block_expr_opt(span, body))
-    }
-
-    fn lower_block_expr_opt(&mut self, span: Span, block: Option<&Block>) -> hir::Expr<'hir> {
-        match block {
-            Some(block) => self.lower_block_expr(block),
-            None => self.expr_err(span, self.dcx().has_errors().unwrap()),
-        }
+    fn lower_fn_body_block(&mut self, decl: &FnDecl, body: &Block) -> hir::BodyId {
+        self.lower_fn_body(decl, |this| this.lower_block_expr(body))
     }
 
     pub(super) fn lower_const_body(&mut self, span: Span, expr: Option<&Expr>) -> hir::BodyId {
@@ -1089,9 +1081,37 @@ impl<'hir> LoweringContext<'_, 'hir> {
         decl: &FnDecl,
         coroutine_kind: Option<CoroutineKind>,
         body: Option<&Block>,
+        attrs: &'hir [hir::Attribute],
     ) -> hir::BodyId {
-        let (Some(coroutine_kind), Some(body)) = (coroutine_kind, body) else {
-            return self.lower_fn_body_block(span, decl, body);
+        let Some(body) = body else {
+            // Functions without a body are an error, except if this is an intrinsic. For those we
+            // create a fake body so that the entire rest of the compiler doesn't have to deal with
+            // this as a special case.
+            return self.lower_fn_body(decl, |this| {
+                if attrs.iter().any(|a| a.name_or_empty() == sym::rustc_intrinsic) {
+                    let empty_block = hir::Block {
+                        hir_id: this.next_id(),
+                        stmts: &[],
+                        expr: None,
+                        rules: hir::BlockCheckMode::DefaultBlock,
+                        span,
+                        targeted_by_break: false,
+                    };
+                    let loop_ = hir::ExprKind::Loop(
+                        this.arena.alloc(empty_block),
+                        None,
+                        hir::LoopSource::Loop,
+                        span,
+                    );
+                    hir::Expr { hir_id: this.next_id(), kind: loop_, span }
+                } else {
+                    this.expr_err(span, this.dcx().has_errors().unwrap())
+                }
+            });
+        };
+        let Some(coroutine_kind) = coroutine_kind else {
+            // Typical case: not a coroutine.
+            return self.lower_fn_body_block(decl, body);
         };
         self.lower_body(|this| {
             let (parameters, expr) = this.lower_coroutine_body_with_moved_arguments(

compiler/rustc_ast_passes/src/ast_validation.rs

@@ -920,7 +920,9 @@ impl<'a> Visitor<'a> for AstValidator<'a> {
             ItemKind::Fn(box Fn { defaultness, sig, generics, body }) => {
                 self.check_defaultness(item.span, *defaultness);
 
-                if body.is_none() {
+                let is_intrinsic =
+                    item.attrs.iter().any(|a| a.name_or_empty() == sym::rustc_intrinsic);
+                if body.is_none() && !is_intrinsic {
                     self.dcx().emit_err(errors::FnWithoutBody {
                         span: item.span,
                         replace_span: self.ending_semi_or_hi(item.span),