Merge from rustc

8 files changed, 132 insertions, 63 deletions

diff --git a/compiler/rustc_parse/src/errors.rs b/compiler/rustc_parse/src/errors.rs
index 40502158469..fdd500e90f8 100644
--- a/compiler/rustc_parse/src/errors.rs
+++ b/compiler/rustc_parse/src/errors.rs
@@ -1756,6 +1756,14 @@ pub(crate) struct MissingFnParams {
 }
 
 #[derive(Diagnostic)]
+#[diag(parse_invalid_path_sep_in_fn_definition)]
+pub(crate) struct InvalidPathSepInFnDefinition {
+    #[primary_span]
+    #[suggestion(code = "", applicability = "machine-applicable", style = "verbose")]
+    pub span: Span,
+}
+
+#[derive(Diagnostic)]
 #[diag(parse_missing_trait_in_trait_impl)]
 pub(crate) struct MissingTraitInTraitImpl {
     #[primary_span]
@@ -2111,6 +2119,24 @@ pub(crate) enum UnknownPrefixSugg {
 }
 
 #[derive(Diagnostic)]
+#[diag(parse_reserved_string)]
+#[note]
+pub(crate) struct ReservedString {
+    #[primary_span]
+    pub span: Span,
+    #[subdiagnostic]
+    pub sugg: Option<GuardedStringSugg>,
+}
+#[derive(Subdiagnostic)]
+#[suggestion(
+    parse_suggestion_whitespace,
+    code = " ",
+    applicability = "maybe-incorrect",
+    style = "verbose"
+)]
+pub(crate) struct GuardedStringSugg(#[primary_span] pub Span);
+
+#[derive(Diagnostic)]
 #[diag(parse_too_many_hashes)]
 pub(crate) struct TooManyHashes {
     #[primary_span]
@@ -2608,6 +2634,7 @@ pub(crate) struct ExpectedCommaAfterPatternField {
 
 #[derive(Diagnostic)]
 #[diag(parse_unexpected_expr_in_pat)]
+#[note]
 pub(crate) struct UnexpectedExpressionInPattern {
     /// The unexpected expr's span.
     #[primary_span]
diff --git a/compiler/rustc_parse/src/lexer/mod.rs b/compiler/rustc_parse/src/lexer/mod.rs
index 3e46fc93fa4..d627ef3d2cb 100644
--- a/compiler/rustc_parse/src/lexer/mod.rs
+++ b/compiler/rustc_parse/src/lexer/mod.rs
@@ -10,7 +10,8 @@ use rustc_lexer::unescape::{self, EscapeError, Mode};
 use rustc_lexer::{Base, Cursor, DocStyle, LiteralKind, RawStrError};
 use rustc_session::lint::BuiltinLintDiag;
 use rustc_session::lint::builtin::{
-    RUST_2021_PREFIXES_INCOMPATIBLE_SYNTAX, TEXT_DIRECTION_CODEPOINT_IN_COMMENT,
+    RUST_2021_PREFIXES_INCOMPATIBLE_SYNTAX, RUST_2024_GUARDED_STRING_INCOMPATIBLE_SYNTAX,
+    TEXT_DIRECTION_CODEPOINT_IN_COMMENT,
 };
 use rustc_session::parse::ParseSess;
 use rustc_span::symbol::Symbol;
@@ -251,6 +252,7 @@ impl<'psess, 'src> StringReader<'psess, 'src> {
                     let prefix_span = self.mk_sp(start, lit_start);
                     return (Token::new(self.ident(start), prefix_span), preceded_by_whitespace);
                 }
+                rustc_lexer::TokenKind::GuardedStrPrefix => self.maybe_report_guarded_str(start, str_before),
                 rustc_lexer::TokenKind::Literal { kind, suffix_start } => {
                     let suffix_start = start + BytePos(suffix_start);
                     let (kind, symbol) = self.cook_lexer_literal(start, suffix_start, kind);
@@ -781,6 +783,86 @@ impl<'psess, 'src> StringReader<'psess, 'src> {
         }
     }
 
+    /// Detect guarded string literal syntax
+    ///
+    /// RFC 3598 reserved this syntax for future use. As of Rust 2024,
+    /// using this syntax produces an error. In earlier editions, however, it
+    /// only results in an (allowed by default) lint, and is treated as
+    /// separate tokens.
+    fn maybe_report_guarded_str(&mut self, start: BytePos, str_before: &'src str) -> TokenKind {
+        let span = self.mk_sp(start, self.pos);
+        let edition2024 = span.edition().at_least_rust_2024();
+
+        let space_pos = start + BytePos(1);
+        let space_span = self.mk_sp(space_pos, space_pos);
+
+        let mut cursor = Cursor::new(str_before);
+
+        let (span, unterminated) = match cursor.guarded_double_quoted_string() {
+            Some(rustc_lexer::GuardedStr { n_hashes, terminated, token_len }) => {
+                let end = start + BytePos(token_len);
+                let span = self.mk_sp(start, end);
+                let str_start = start + BytePos(n_hashes);
+
+                if edition2024 {
+                    self.cursor = cursor;
+                    self.pos = end;
+                }
+
+                let unterminated = if terminated { None } else { Some(str_start) };
+
+                (span, unterminated)
+            }
+            _ => {
+                // We should only get here in the `##+` case.
+                debug_assert_eq!(self.str_from_to(start, start + BytePos(2)), "##");
+
+                (span, None)
+            }
+        };
+        if edition2024 {
+            if let Some(str_start) = unterminated {
+                // Only a fatal error if string is unterminated.
+                self.dcx()
+                    .struct_span_fatal(
+                        self.mk_sp(str_start, self.pos),
+                        "unterminated double quote string",
+                    )
+                    .with_code(E0765)
+                    .emit()
+            }
+
+            let sugg = if span.from_expansion() {
+                None
+            } else {
+                Some(errors::GuardedStringSugg(space_span))
+            };
+
+            // In Edition 2024 and later, emit a hard error.
+            let err = self.dcx().emit_err(errors::ReservedString { span, sugg });
+
+            token::Literal(token::Lit {
+                kind: token::Err(err),
+                symbol: self.symbol_from_to(start, self.pos),
+                suffix: None,
+            })
+        } else {
+            // Before Rust 2024, only emit a lint for migration.
+            self.psess.buffer_lint(
+                RUST_2024_GUARDED_STRING_INCOMPATIBLE_SYNTAX,
+                span,
+                ast::CRATE_NODE_ID,
+                BuiltinLintDiag::ReservedString(space_span),
+            );
+
+            // For backwards compatibility, roll back to after just the first `#`
+            // and return the `Pound` token.
+            self.pos = start + BytePos(1);
+            self.cursor = Cursor::new(&str_before[1..]);
+            token::Pound
+        }
+    }
+
     fn report_too_many_hashes(&self, start: BytePos, num: u32) -> ! {
         self.dcx().emit_fatal(errors::TooManyHashes { span: self.mk_sp(start, self.pos), num });
     }
diff --git a/compiler/rustc_parse/src/lib.rs b/compiler/rustc_parse/src/lib.rs
index f7a8b8780ed..2792050a0b3 100644
--- a/compiler/rustc_parse/src/lib.rs
+++ b/compiler/rustc_parse/src/lib.rs
@@ -18,7 +18,7 @@ use std::path::Path;
 
 use rustc_ast as ast;
 use rustc_ast::tokenstream::TokenStream;
-use rustc_ast::{AttrItem, Attribute, MetaItem, token};
+use rustc_ast::{AttrItem, Attribute, MetaItemInner, token};
 use rustc_ast_pretty::pprust;
 use rustc_data_structures::sync::Lrc;
 use rustc_errors::{Diag, FatalError, PResult};
@@ -160,7 +160,7 @@ pub fn fake_token_stream_for_crate(psess: &ParseSess, krate: &ast::Crate) -> Tok
 pub fn parse_cfg_attr(
     cfg_attr: &Attribute,
     psess: &ParseSess,
-) -> Option<(MetaItem, Vec<(AttrItem, Span)>)> {
+) -> Option<(MetaItemInner, Vec<(AttrItem, Span)>)> {
     const CFG_ATTR_GRAMMAR_HELP: &str = "#[cfg_attr(condition, attribute, other_attribute, ...)]";
     const CFG_ATTR_NOTE_REF: &str = "for more information, visit \
         <https://doc.rust-lang.org/reference/conditional-compilation.html#the-cfg_attr-attribute>";
diff --git a/compiler/rustc_parse/src/parser/attr.rs b/compiler/rustc_parse/src/parser/attr.rs
index c65cf3f40f6..b6fa2099588 100644
--- a/compiler/rustc_parse/src/parser/attr.rs
+++ b/compiler/rustc_parse/src/parser/attr.rs
@@ -356,8 +356,10 @@ impl<'a> Parser<'a> {
     }
 
     /// Parses `cfg_attr(pred, attr_item_list)` where `attr_item_list` is comma-delimited.
-    pub fn parse_cfg_attr(&mut self) -> PResult<'a, (ast::MetaItem, Vec<(ast::AttrItem, Span)>)> {
-        let cfg_predicate = self.parse_meta_item(AllowLeadingUnsafe::No)?;
+    pub fn parse_cfg_attr(
+        &mut self,
+    ) -> PResult<'a, (ast::MetaItemInner, Vec<(ast::AttrItem, Span)>)> {
+        let cfg_predicate = self.parse_meta_item_inner()?;
         self.expect(&token::Comma)?;
 
         // Presumably, the majority of the time there will only be one attr.
@@ -375,7 +377,7 @@ impl<'a> Parser<'a> {
     }
 
     /// Matches `COMMASEP(meta_item_inner)`.
-    pub(crate) fn parse_meta_seq_top(&mut self) -> PResult<'a, ThinVec<ast::NestedMetaItem>> {
+    pub(crate) fn parse_meta_seq_top(&mut self) -> PResult<'a, ThinVec<ast::MetaItemInner>> {
         // Presumably, the majority of the time there will only be one attr.
         let mut nmis = ThinVec::with_capacity(1);
         while self.token != token::Eof {
@@ -452,14 +454,14 @@ impl<'a> Parser<'a> {
     /// ```ebnf
     /// MetaItemInner = UNSUFFIXED_LIT | MetaItem ;
     /// ```
-    fn parse_meta_item_inner(&mut self) -> PResult<'a, ast::NestedMetaItem> {
+    pub fn parse_meta_item_inner(&mut self) -> PResult<'a, ast::MetaItemInner> {
         match self.parse_unsuffixed_meta_item_lit() {
-            Ok(lit) => return Ok(ast::NestedMetaItem::Lit(lit)),
+            Ok(lit) => return Ok(ast::MetaItemInner::Lit(lit)),
             Err(err) => err.cancel(), // we provide a better error below
         }
 
         match self.parse_meta_item(AllowLeadingUnsafe::No) {
-            Ok(mi) => return Ok(ast::NestedMetaItem::MetaItem(mi)),
+            Ok(mi) => return Ok(ast::MetaItemInner::MetaItem(mi)),
             Err(err) => err.cancel(), // we provide a better error below
         }
 
diff --git a/compiler/rustc_parse/src/parser/generics.rs b/compiler/rustc_parse/src/parser/generics.rs
index b9256daa725..5aebe716b0a 100644
--- a/compiler/rustc_parse/src/parser/generics.rs
+++ b/compiler/rustc_parse/src/parser/generics.rs
@@ -269,6 +269,13 @@ impl<'a> Parser<'a> {
     ///                  | ( < lifetimes , typaramseq ( , )? > )
     /// where   typaramseq = ( typaram ) | ( typaram , typaramseq )
     pub(super) fn parse_generics(&mut self) -> PResult<'a, ast::Generics> {
+        // invalid path separator `::` in function definition
+        // for example `fn invalid_path_separator::<T>() {}`
+        if self.eat_noexpect(&token::PathSep) {
+            self.dcx()
+                .emit_err(errors::InvalidPathSepInFnDefinition { span: self.prev_token.span });
+        }
+
         let span_lo = self.token.span;
         let (params, span) = if self.eat_lt() {
             let params = self.parse_generic_params()?;
diff --git a/compiler/rustc_parse/src/parser/item.rs b/compiler/rustc_parse/src/parser/item.rs
index 9fc82d84225..36733726564 100644
--- a/compiler/rustc_parse/src/parser/item.rs
+++ b/compiler/rustc_parse/src/parser/item.rs
@@ -1984,7 +1984,7 @@ impl<'a> Parser<'a> {
             }
         }
         self.expect_field_ty_separator()?;
-        let ty = self.parse_ty_for_field_def()?;
+        let ty = self.parse_ty()?;
         if self.token == token::Colon && self.look_ahead(1, |t| *t != token::Colon) {
             self.dcx().emit_err(errors::SingleColonStructType { span: self.token.span });
         }
@@ -2009,9 +2009,7 @@ impl<'a> Parser<'a> {
     /// for better diagnostics and suggestions.
     fn parse_field_ident(&mut self, adt_ty: &str, lo: Span) -> PResult<'a, Ident> {
         let (ident, is_raw) = self.ident_or_err(true)?;
-        if ident.name == kw::Underscore {
-            self.psess.gated_spans.gate(sym::unnamed_fields, lo);
-        } else if matches!(is_raw, IdentIsRaw::No) && ident.is_reserved() {
+        if matches!(is_raw, IdentIsRaw::No) && ident.is_reserved() {
             let snapshot = self.create_snapshot_for_diagnostic();
             let err = if self.check_fn_front_matter(false, Case::Sensitive) {
                 let inherited_vis =
diff --git a/compiler/rustc_parse/src/parser/ty.rs b/compiler/rustc_parse/src/parser/ty.rs
index a8ed8b5df9c..625a4cabdf2 100644
--- a/compiler/rustc_parse/src/parser/ty.rs
+++ b/compiler/rustc_parse/src/parser/ty.rs
@@ -128,17 +128,6 @@ impl<'a> Parser<'a> {
         )
     }
 
-    /// Parse a type suitable for a field definition.
-    /// The difference from `parse_ty` is that this version
-    /// allows anonymous structs and unions.
-    pub(super) fn parse_ty_for_field_def(&mut self) -> PResult<'a, P<Ty>> {
-        if self.can_begin_anon_struct_or_union() {
-            self.parse_anon_struct_or_union()
-        } else {
-            self.parse_ty()
-        }
-    }
-
     /// Parse a type suitable for a function or function pointer parameter.
     /// The difference from `parse_ty` is that this version allows `...`
     /// (`CVarArgs`) at the top level of the type.
@@ -382,37 +371,6 @@ impl<'a> Parser<'a> {
         if allow_qpath_recovery { self.maybe_recover_from_bad_qpath(ty) } else { Ok(ty) }
     }
 
-    /// Parse an anonymous struct or union (only for field definitions):
-    /// ```ignore (feature-not-ready)
-    /// #[repr(C)]
-    /// struct Foo {
-    ///     _: struct { // anonymous struct
-    ///         x: u32,
-    ///         y: f64,
-    ///     }
-    ///     _: union { // anonymous union
-    ///         z: u32,
-    ///         w: f64,
-    ///     }
-    /// }
-    /// ```
-    fn parse_anon_struct_or_union(&mut self) -> PResult<'a, P<Ty>> {
-        assert!(self.token.is_keyword(kw::Union) || self.token.is_keyword(kw::Struct));
-        let is_union = self.token.is_keyword(kw::Union);
-
-        let lo = self.token.span;
-        self.bump();
-
-        let (fields, _recovered) =
-            self.parse_record_struct_body(if is_union { "union" } else { "struct" }, lo, false)?;
-        let span = lo.to(self.prev_token.span);
-        self.psess.gated_spans.gate(sym::unnamed_fields, span);
-        let id = ast::DUMMY_NODE_ID;
-        let kind =
-            if is_union { TyKind::AnonUnion(id, fields) } else { TyKind::AnonStruct(id, fields) };
-        Ok(self.mk_ty(span, kind))
-    }
-
     /// Parses either:
     /// - `(TYPE)`, a parenthesized type.
     /// - `(TYPE,)`, a tuple with a single field of type TYPE.
@@ -813,11 +771,6 @@ impl<'a> Parser<'a> {
         Ok(bounds)
     }
 
-    pub(super) fn can_begin_anon_struct_or_union(&mut self) -> bool {
-        (self.token.is_keyword(kw::Struct) || self.token.is_keyword(kw::Union))
-            && self.look_ahead(1, |t| t == &token::OpenDelim(Delimiter::Brace))
-    }
-
     /// Can the current token begin a bound?
     fn can_begin_bound(&mut self) -> bool {
         self.check_path()
diff --git a/compiler/rustc_parse/src/validate_attr.rs b/compiler/rustc_parse/src/validate_attr.rs
index b15d1edf79c..f3174e7dea2 100644
--- a/compiler/rustc_parse/src/validate_attr.rs
+++ b/compiler/rustc_parse/src/validate_attr.rs
@@ -3,8 +3,8 @@
 use rustc_ast::token::Delimiter;
 use rustc_ast::tokenstream::DelimSpan;
 use rustc_ast::{
-    self as ast, AttrArgs, AttrArgsEq, Attribute, DelimArgs, MetaItem, MetaItemKind,
-    NestedMetaItem, Safety,
+    self as ast, AttrArgs, AttrArgsEq, Attribute, DelimArgs, MetaItem, MetaItemInner, MetaItemKind,
+    Safety,
 };
 use rustc_errors::{Applicability, FatalError, PResult};
 use rustc_feature::{AttributeSafety, AttributeTemplate, BUILTIN_ATTRIBUTE_MAP, BuiltinAttribute};
@@ -143,7 +143,7 @@ pub(super) fn check_cfg_attr_bad_delim(psess: &ParseSess, span: DelimSpan, delim
 
 /// Checks that the given meta-item is compatible with this `AttributeTemplate`.
 fn is_attr_template_compatible(template: &AttributeTemplate, meta: &ast::MetaItemKind) -> bool {
-    let is_one_allowed_subword = |items: &[NestedMetaItem]| match items {
+    let is_one_allowed_subword = |items: &[MetaItemInner]| match items {
         [item] => item.is_word() && template.one_of.iter().any(|&word| item.has_name(word)),
         _ => false,
     };