Rollup merge of #110694 - est31:builtin, r=petrochenkov

Implement builtin # syntax and use it for offset_of!(...)

Add `builtin #` syntax to the parser, as well as a generic infrastructure to support both item and expression position builtin syntaxes. The PR also uses this infrastructure for the implementation of the `offset_of!` macro, added by #106934.

cc `@petrochenkov` `@DrMeepster`

cc #110680 `builtin #` tracking issue
cc #106655 `offset_of!` tracking issue

compiler/rustc_ast_passes/src/feature_gate.rs

@@ -603,6 +603,7 @@ pub fn check_crate(krate: &ast::Crate, sess: &Session) {
     gate_all!(yeet_expr, "`do yeet` expression is experimental");
     gate_all!(dyn_star, "`dyn*` trait objects are experimental");
     gate_all!(const_closures, "const closures are experimental");
+    gate_all!(builtin_syntax, "`builtin #` syntax is unstable");
 
     if !visitor.features.negative_bounds {
         for &span in spans.get(&sym::negative_bounds).iter().copied().flatten() {

compiler/rustc_ast_pretty/src/pprust/state/expr.rs

@@ -556,8 +556,7 @@ impl<'a> State<'a> {
                 self.pclose();
             }
             ast::ExprKind::OffsetOf(container, fields) => {
-                // FIXME: This should have its own syntax, distinct from a macro invocation.
-                self.word("offset_of!");
+                self.word("builtin # offset_of");
                 self.popen();
                 self.rbox(0, Inconsistent);
                 self.print_type(container);

compiler/rustc_builtin_macros/messages.ftl

@@ -150,10 +150,6 @@ builtin_macros_format_pos_mismatch = {$n} positional {$n ->
     *[more] arguments
     } in format string, but {$desc}
 
-builtin_macros_offset_of_expected_field = expected field
-
-builtin_macros_offset_of_expected_two_args = expected 2 arguments
-
 builtin_macros_test_case_non_item = `#[test_case]` attribute is only allowed on items
 
 builtin_macros_test_bad_fn = {$kind} functions cannot be used for tests

compiler/rustc_builtin_macros/src/lib.rs

@@ -44,7 +44,6 @@ mod format;
 mod format_foreign;
 mod global_allocator;
 mod log_syntax;
-mod offset_of;
 mod source_util;
 mod test;
 mod trace_macros;
@@ -92,7 +91,6 @@ pub fn register_builtin_macros(resolver: &mut dyn ResolverExpand) {
         line: source_util::expand_line,
         log_syntax: log_syntax::expand_log_syntax,
         module_path: source_util::expand_mod,
-        offset_of: offset_of::expand_offset_of,
         option_env: env::expand_option_env,
         core_panic: edition_panic::expand_panic,
         std_panic: edition_panic::expand_panic,

compiler/rustc_builtin_macros/src/offset_of.rs

@@ -1,99 +0,0 @@
-use rustc_ast as ast;
-use rustc_ast::ptr::P;
-use rustc_ast::token;
-use rustc_ast::tokenstream::TokenStream;
-use rustc_errors::PResult;
-use rustc_expand::base::{self, *};
-use rustc_macros::Diagnostic;
-use rustc_parse::parser::Parser;
-use rustc_span::{symbol::Ident, Span};
-
-#[derive(Diagnostic)]
-#[diag(builtin_macros_offset_of_expected_field)]
-struct ExpectedField {
-    #[primary_span]
-    span: Span,
-}
-
-#[derive(Diagnostic)]
-#[diag(builtin_macros_offset_of_expected_two_args)]
-struct ExpectedTwoArgs {
-    #[primary_span]
-    span: Span,
-}
-
-fn parse_field<'a>(cx: &ExtCtxt<'a>, p: &mut Parser<'a>) -> PResult<'a, Ident> {
-    let token = p.token.uninterpolate();
-    let field = match token.kind {
-        token::Ident(name, _) => Ident::new(name, token.span),
-        token::Literal(token::Lit { kind: token::Integer, symbol, suffix: None }) => {
-            Ident::new(symbol, token.span)
-        }
-        _ => return Err(cx.create_err(ExpectedField { span: p.token.span })),
-    };
-
-    p.bump();
-
-    Ok(field)
-}
-
-fn parse_args<'a>(
-    cx: &mut ExtCtxt<'a>,
-    sp: Span,
-    tts: TokenStream,
-) -> PResult<'a, (P<ast::Ty>, P<[Ident]>)> {
-    let mut p = cx.new_parser_from_tts(tts);
-
-    let container = p.parse_ty()?;
-
-    p.expect(&token::Comma)?;
-
-    if p.eat(&token::Eof) {
-        return Err(cx.create_err(ExpectedTwoArgs { span: sp }));
-    }
-
-    let mut fields = Vec::new();
-
-    loop {
-        let field = parse_field(cx, &mut p)?;
-        fields.push(field);
-
-        if p.eat(&token::Dot) {
-            continue;
-        }
-
-        p.eat(&token::Comma);
-
-        if !p.eat(&token::Eof) {
-            return Err(cx.create_err(ExpectedTwoArgs { span: sp }));
-        }
-
-        break;
-    }
-
-    Ok((container, fields.into()))
-}
-
-pub fn expand_offset_of<'cx>(
-    cx: &'cx mut ExtCtxt<'_>,
-    sp: Span,
-    tts: TokenStream,
-) -> Box<dyn base::MacResult + 'cx> {
-    match parse_args(cx, sp, tts) {
-        Ok((container, fields)) => {
-            let expr = P(ast::Expr {
-                id: ast::DUMMY_NODE_ID,
-                kind: ast::ExprKind::OffsetOf(container, fields),
-                span: sp,
-                attrs: ast::AttrVec::new(),
-                tokens: None,
-            });
-
-            MacEager::expr(expr)
-        }
-        Err(mut err) => {
-            err.emit();
-            DummyResult::any(sp)
-        }
-    }
-}

compiler/rustc_feature/src/active.rs

@@ -313,6 +313,8 @@ declare_features! (
     (active, async_closure, "1.37.0", Some(62290), None),
     /// Allows async functions to be declared, implemented, and used in traits.
     (active, async_fn_in_trait, "1.66.0", Some(91611), None),
+    /// Allows builtin # foo() syntax
+    (active, builtin_syntax, "CURRENT_RUSTC_VERSION", Some(110680), None),
     /// Allows `c"foo"` literals.
     (active, c_str_literals, "CURRENT_RUSTC_VERSION", Some(105723), None),
     /// Treat `extern "C"` function as nounwind.

compiler/rustc_parse/messages.ftl

@@ -257,6 +257,10 @@ parse_invalid_literal_suffix_on_tuple_index = suffixes on a tuple index are inva
     .tuple_exception_line_2 = on proc macros, you'll want to use `syn::Index::from` or `proc_macro::Literal::*_unsuffixed` for code that will desugar to tuple field access
     .tuple_exception_line_3 = see issue #60210 <https://github.com/rust-lang/rust/issues/60210> for more information
 
+parse_expected_builtin_ident = expected identifier after `builtin #`
+
+parse_unknown_builtin_construct = unknown `builtin #` construct `{$name}`
+
 parse_non_string_abi_literal = non-string ABI literal
     .suggestion = specify the ABI with a string literal
 

compiler/rustc_parse/src/errors.rs

@@ -2644,3 +2644,18 @@ pub(crate) struct MalformedCfgAttr {
     pub span: Span,
     pub sugg: &'static str,
 }
+
+#[derive(Diagnostic)]
+#[diag(parse_unknown_builtin_construct)]
+pub(crate) struct UnknownBuiltinConstruct {
+    #[primary_span]
+    pub span: Span,
+    pub name: Symbol,
+}
+
+#[derive(Diagnostic)]
+#[diag(parse_expected_builtin_ident)]
+pub(crate) struct ExpectedBuiltinIdent {
+    #[primary_span]
+    pub span: Span,
+}

compiler/rustc_parse/src/parser/expr.rs

@@ -1300,6 +1300,8 @@ impl<'a> Parser<'a> {
             })
         } else if self.check(&token::OpenDelim(Delimiter::Bracket)) {
             self.parse_expr_array_or_repeat(Delimiter::Bracket)
+        } else if self.is_builtin() {
+            self.parse_expr_builtin()
         } else if self.check_path() {
             self.parse_expr_path_start()
         } else if self.check_keyword(kw::Move)
@@ -1766,6 +1768,61 @@ impl<'a> Parser<'a> {
         self.maybe_recover_from_bad_qpath(expr)
     }
 
+    /// Parse `builtin # ident(args,*)`.
+    fn parse_expr_builtin(&mut self) -> PResult<'a, P<Expr>> {
+        self.parse_builtin(|this, lo, ident| {
+            if ident.name == sym::offset_of {
+                return Ok(Some(this.parse_expr_offset_of(lo)?));
+            }
+
+            Ok(None)
+        })
+    }
+
+    pub(crate) fn parse_builtin<T>(
+        &mut self,
+        parse: impl FnOnce(&mut Parser<'a>, Span, Ident) -> PResult<'a, Option<T>>,
+    ) -> PResult<'a, T> {
+        let lo = self.token.span;
+
+        self.bump(); // `builtin`
+        self.bump(); // `#`
+
+        let Some((ident, false)) = self.token.ident() else {
+            let err = errors::ExpectedBuiltinIdent { span: self.token.span }
+                .into_diagnostic(&self.sess.span_diagnostic);
+            return Err(err);
+        };
+        self.sess.gated_spans.gate(sym::builtin_syntax, ident.span);
+        self.bump();
+
+        self.expect(&TokenKind::OpenDelim(Delimiter::Parenthesis))?;
+        let ret = if let Some(res) = parse(self, lo, ident)? {
+            Ok(res)
+        } else {
+            let err = errors::UnknownBuiltinConstruct { span: lo.to(ident.span), name: ident.name }
+                .into_diagnostic(&self.sess.span_diagnostic);
+            return Err(err);
+        };
+        self.expect(&TokenKind::CloseDelim(Delimiter::Parenthesis))?;
+
+        ret
+    }
+
+    pub(crate) fn parse_expr_offset_of(&mut self, lo: Span) -> PResult<'a, P<Expr>> {
+        let container = self.parse_ty()?;
+        self.expect(&TokenKind::Comma)?;
+
+        let seq_sep = SeqSep { sep: Some(token::Dot), trailing_sep_allowed: false };
+        let (fields, _trailing, _recovered) = self.parse_seq_to_before_end(
+            &TokenKind::CloseDelim(Delimiter::Parenthesis),
+            seq_sep,
+            Parser::parse_field_name,
+        )?;
+        let span = lo.to(self.token.span);
+        Ok(self.mk_expr(span, ExprKind::OffsetOf(container, fields.to_vec().into())))
+    }
+
     /// Returns a string literal if the next token is a string literal.
     /// In case of error returns `Some(lit)` if the next token is a literal with a wrong kind,
     /// and returns `None` if the next token is not literal at all.
@@ -2835,6 +2892,10 @@ impl<'a> Parser<'a> {
         })
     }
 
+    pub(crate) fn is_builtin(&self) -> bool {
+        self.token.is_keyword(kw::Builtin) && self.look_ahead(1, |t| *t == token::Pound)
+    }
+
     /// Parses a `try {...}` expression (`try` token already eaten).
     fn parse_try_block(&mut self, span_lo: Span) -> PResult<'a, P<Expr>> {
         let (attrs, body) = self.parse_inner_attrs_and_block()?;

compiler/rustc_parse/src/parser/item.rs

@@ -265,6 +265,9 @@ impl<'a> Parser<'a> {
             // UNION ITEM
             self.bump(); // `union`
             self.parse_item_union()?
+        } else if self.is_builtin() {
+            // BUILTIN# ITEM
+            return self.parse_item_builtin();
         } else if self.eat_keyword(kw::Macro) {
             // MACROS 2.0 ITEM
             self.parse_item_decl_macro(lo)?
@@ -434,6 +437,11 @@ impl<'a> Parser<'a> {
         }
     }
 
+    fn parse_item_builtin(&mut self) -> PResult<'a, Option<ItemInfo>> {
+        // To be expanded
+        return Ok(None);
+    }
+
     /// Parses an item macro, e.g., `item!();`.
     fn parse_item_macro(&mut self, vis: &Visibility) -> PResult<'a, MacCall> {
         let path = self.parse_path(PathStyle::Mod)?; // `foo::bar`

compiler/rustc_parse/src/parser/stmt.rs

@@ -90,7 +90,11 @@ impl<'a> Parser<'a> {
                 attrs,
                 errors::InvalidVariableDeclarationSub::UseLetNotVar,
             )?
-        } else if self.check_path() && !self.token.is_qpath_start() && !self.is_path_start_item() {
+        } else if self.check_path()
+            && !self.token.is_qpath_start()
+            && !self.is_path_start_item()
+            && !self.is_builtin()
+        {
             // We have avoided contextual keywords like `union`, items with `crate` visibility,
             // or `auto trait` items. We aim to parse an arbitrary path `a::b` but not something
             // that starts like a path (1 token), but it fact not a path.

compiler/rustc_span/src/symbol.rs

@@ -95,6 +95,7 @@ symbols! {
 
         // Weak keywords, have special meaning only in specific contexts.
         Auto:               "auto",
+        Builtin:            "builtin",
         Catch:              "catch",
         Default:            "default",
         MacroRules:         "macro_rules",
@@ -440,6 +441,7 @@ symbols! {
         breakpoint,
         bridge,
         bswap,
+        builtin_syntax,
         c_str,
         c_str_literals,
         c_unwind,