Auto merge of #77255 - Aaron1011:feature/collect-attr-tokens, r=petrochenkov

Unconditionally capture tokens for attributes.

This allows us to avoid synthesizing tokens in `prepend_attr`, since we
have the original tokens available.

We still need to synthesize tokens when expanding `cfg_attr`,
but this is an unavoidable consequence of the syntax of `cfg_attr` -
the user does not supply the `#` and `[]` tokens that a `cfg_attr`
expands to.

This is based on PR https://github.com/rust-lang/rust/pull/77250 - this PR exposes a bug in the current `collect_tokens` implementation, which is fixed by the rewrite.

compiler/rustc_ast/src/ast.rs

@@ -2423,6 +2423,7 @@ pub struct Attribute {
     /// or the construct this attribute is contained within (inner).
     pub style: AttrStyle,
     pub span: Span,
+    pub tokens: Option<LazyTokenStream>,
 }
 
 #[derive(Clone, Encodable, Decodable, Debug)]

compiler/rustc_ast/src/attr/mod.rs

@@ -325,7 +325,7 @@ pub fn mk_attr(style: AttrStyle, path: Path, args: MacArgs, span: Span) -> Attri
 }
 
 pub fn mk_attr_from_item(style: AttrStyle, item: AttrItem, span: Span) -> Attribute {
-    Attribute { kind: AttrKind::Normal(item), id: mk_attr_id(), style, span }
+    Attribute { kind: AttrKind::Normal(item), id: mk_attr_id(), style, span, tokens: None }
 }
 
 /// Returns an inner attribute with the given value and span.
@@ -344,7 +344,13 @@ pub fn mk_doc_comment(
     data: Symbol,
     span: Span,
 ) -> Attribute {
-    Attribute { kind: AttrKind::DocComment(comment_kind, data), id: mk_attr_id(), style, span }
+    Attribute {
+        kind: AttrKind::DocComment(comment_kind, data),
+        id: mk_attr_id(),
+        style,
+        span,
+        tokens: None,
+    }
 }
 
 pub fn list_contains_name(items: &[NestedMetaItem], name: Symbol) -> bool {

compiler/rustc_ast/src/mut_visit.rs

@@ -577,7 +577,7 @@ pub fn noop_visit_local<T: MutVisitor>(local: &mut P<Local>, vis: &mut T) {
 }
 
 pub fn noop_visit_attribute<T: MutVisitor>(attr: &mut Attribute, vis: &mut T) {
-    let Attribute { kind, id: _, style: _, span } = attr;
+    let Attribute { kind, id: _, style: _, span, tokens: _ } = attr;
     match kind {
         AttrKind::Normal(AttrItem { path, args, tokens: _ }) => {
             vis.visit_path(path);

compiler/rustc_ast_lowering/src/expr.rs

@@ -210,9 +210,9 @@ impl<'hir> LoweringContext<'_, 'hir> {
                         ex.span = e.span;
                     }
                     // Merge attributes into the inner expression.
-                    let mut attrs = e.attrs.clone();
+                    let mut attrs: Vec<_> = e.attrs.iter().map(|a| self.lower_attr(a)).collect();
                     attrs.extend::<Vec<_>>(ex.attrs.into());
-                    ex.attrs = attrs;
+                    ex.attrs = attrs.into();
                     return ex;
                 }
 
@@ -1471,13 +1471,15 @@ impl<'hir> LoweringContext<'_, 'hir> {
             hir::MatchSource::ForLoopDesugar,
         ));
 
+        let attrs: Vec<_> = e.attrs.iter().map(|a| self.lower_attr(a)).collect();
+
         // This is effectively `{ let _result = ...; _result }`.
         // The construct was introduced in #21984 and is necessary to make sure that
         // temporaries in the `head` expression are dropped and do not leak to the
         // surrounding scope of the `match` since the `match` is not a terminating scope.
         //
         // Also, add the attributes to the outer returned expr node.
-        self.expr_drop_temps_mut(desugared_span, match_expr, e.attrs.clone())
+        self.expr_drop_temps_mut(desugared_span, match_expr, attrs.into())
     }
 
     /// Desugar `ExprKind::Try` from: `<expr>?` into:

compiler/rustc_ast_lowering/src/lib.rs

@@ -972,7 +972,8 @@ impl<'a, 'hir> LoweringContext<'a, 'hir> {
             AttrKind::DocComment(comment_kind, data) => AttrKind::DocComment(comment_kind, data),
         };
 
-        Attribute { kind, id: attr.id, style: attr.style, span: attr.span }
+        // Tokens aren't needed after macro expansion and parsing
+        Attribute { kind, id: attr.id, style: attr.style, span: attr.span, tokens: None }
     }
 
     fn lower_mac_args(&mut self, args: &MacArgs) -> MacArgs {
@@ -1713,7 +1714,7 @@ impl<'a, 'hir> LoweringContext<'a, 'hir> {
                 pat: self.lower_pat(&l.pat),
                 init,
                 span: l.span,
-                attrs: l.attrs.clone(),
+                attrs: l.attrs.iter().map(|a| self.lower_attr(a)).collect::<Vec<_>>().into(),
                 source: hir::LocalSource::Normal,
             },
             ids,

compiler/rustc_builtin_macros/src/cmdline_attrs.rs

@@ -15,7 +15,7 @@ pub fn inject(mut krate: ast::Crate, parse_sess: &ParseSess, attrs: &[String]) -
         );
 
         let start_span = parser.token.span;
-        let AttrItem { path, args, tokens: _ } = match parser.parse_attr_item() {
+        let AttrItem { path, args, tokens: _ } = match parser.parse_attr_item(false) {
             Ok(ai) => ai,
             Err(mut err) => {
                 err.emit();

compiler/rustc_expand/src/config.rs

@@ -3,10 +3,13 @@
 use rustc_ast::attr::HasAttrs;
 use rustc_ast::mut_visit::*;
 use rustc_ast::ptr::P;
+use rustc_ast::token::{DelimToken, Token, TokenKind};
+use rustc_ast::tokenstream::{DelimSpan, LazyTokenStreamInner, Spacing, TokenStream, TokenTree};
 use rustc_ast::{self as ast, AttrItem, Attribute, MetaItem};
 use rustc_attr as attr;
 use rustc_data_structures::fx::FxHashMap;
 use rustc_data_structures::map_in_place::MapInPlace;
+use rustc_data_structures::sync::Lrc;
 use rustc_errors::{error_code, struct_span_err, Applicability, Handler};
 use rustc_feature::{Feature, Features, State as FeatureState};
 use rustc_feature::{
@@ -289,7 +292,37 @@ impl<'a> StripUnconfigured<'a> {
         expanded_attrs
             .into_iter()
             .flat_map(|(item, span)| {
-                let attr = attr::mk_attr_from_item(attr.style, item, span);
+                let orig_tokens =
+                    attr.tokens.as_ref().unwrap_or_else(|| panic!("Missing tokens for {:?}", attr));
+
+                // We are taking an attribute of the form `#[cfg_attr(pred, attr)]`
+                // and producing an attribute of the form `#[attr]`. We
+                // have captured tokens for `attr` itself, but we need to
+                // synthesize tokens for the wrapper `#` and `[]`, which
+                // we do below.
+
+                // Use the `#` in `#[cfg_attr(pred, attr)]` as the `#` token
+                // for `attr` when we expand it to `#[attr]`
+                let pound_token = orig_tokens.into_token_stream().trees().next().unwrap();
+                if !matches!(pound_token, TokenTree::Token(Token { kind: TokenKind::Pound, .. })) {
+                    panic!("Bad tokens for attribute {:?}", attr);
+                }
+                // We don't really have a good span to use for the syntheized `[]`
+                // in `#[attr]`, so just use the span of the `#` token.
+                let bracket_group = TokenTree::Delimited(
+                    DelimSpan::from_single(pound_token.span()),
+                    DelimToken::Bracket,
+                    item.tokens
+                        .clone()
+                        .unwrap_or_else(|| panic!("Missing tokens for {:?}", item))
+                        .into_token_stream(),
+                );
+
+                let mut attr = attr::mk_attr_from_item(attr.style, item, span);
+                attr.tokens = Some(Lrc::new(LazyTokenStreamInner::Ready(TokenStream::new(vec![
+                    (pound_token, Spacing::Alone),
+                    (bracket_group, Spacing::Alone),
+                ]))));
                 self.process_cfg_attr(attr)
             })
             .collect()

compiler/rustc_expand/src/expand.rs

@@ -1785,6 +1785,7 @@ impl<'a, 'b> MutVisitor for InvocationCollector<'a, 'b> {
                 span: at.span,
                 id: at.id,
                 style: at.style,
+                tokens: None,
             };
         } else {
             noop_visit_attribute(at, self)

compiler/rustc_interface/src/passes.rs

@@ -2,8 +2,9 @@ use crate::interface::{Compiler, Result};
 use crate::proc_macro_decls;
 use crate::util;
 
-use rustc_ast::mut_visit::MutVisitor;
+use rustc_ast::mut_visit::{self, MutVisitor};
-use rustc_ast::{self as ast, visit};
+use rustc_ast::ptr::P;
+use rustc_ast::{self as ast, token, visit};
 use rustc_codegen_ssa::back::link::emit_metadata;
 use rustc_codegen_ssa::traits::CodegenBackend;
 use rustc_data_structures::sync::{par_iter, Lrc, OnceCell, ParallelIterator, WorkerLocal};
@@ -36,6 +37,7 @@ use rustc_span::symbol::Symbol;
 use rustc_span::{FileName, RealFileName};
 use rustc_trait_selection::traits;
 use rustc_typeck as typeck;
+use smallvec::SmallVec;
 use tracing::{info, warn};
 
 use rustc_serialize::json;
@@ -50,6 +52,64 @@ use std::path::PathBuf;
 use std::rc::Rc;
 use std::{env, fs, iter, mem};
 
+/// Remove alls `LazyTokenStreams` from an AST struct
+/// Normally, this is done during AST lowering. However,
+/// printing the AST JSON requires us to serialize
+/// the entire AST, and we don't want to serialize
+/// a `LazyTokenStream`.
+struct TokenStripper;
+impl mut_visit::MutVisitor for TokenStripper {
+    fn flat_map_item(&mut self, mut i: P<ast::Item>) -> SmallVec<[P<ast::Item>; 1]> {
+        i.tokens = None;
+        mut_visit::noop_flat_map_item(i, self)
+    }
+    fn visit_block(&mut self, b: &mut P<ast::Block>) {
+        b.tokens = None;
+        mut_visit::noop_visit_block(b, self);
+    }
+    fn flat_map_stmt(&mut self, mut stmt: ast::Stmt) -> SmallVec<[ast::Stmt; 1]> {
+        stmt.tokens = None;
+        mut_visit::noop_flat_map_stmt(stmt, self)
+    }
+    fn visit_pat(&mut self, p: &mut P<ast::Pat>) {
+        p.tokens = None;
+        mut_visit::noop_visit_pat(p, self);
+    }
+    fn visit_ty(&mut self, ty: &mut P<ast::Ty>) {
+        ty.tokens = None;
+        mut_visit::noop_visit_ty(ty, self);
+    }
+    fn visit_attribute(&mut self, attr: &mut ast::Attribute) {
+        attr.tokens = None;
+        if let ast::AttrKind::Normal(ast::AttrItem { tokens, .. }) = &mut attr.kind {
+            *tokens = None;
+        }
+        mut_visit::noop_visit_attribute(attr, self);
+    }
+
+    fn visit_interpolated(&mut self, nt: &mut token::Nonterminal) {
+        if let token::Nonterminal::NtMeta(meta) = nt {
+            meta.tokens = None;
+        }
+        // Handles all of the other cases
+        mut_visit::noop_visit_interpolated(nt, self);
+    }
+
+    fn visit_path(&mut self, p: &mut ast::Path) {
+        p.tokens = None;
+        mut_visit::noop_visit_path(p, self);
+    }
+    fn visit_vis(&mut self, vis: &mut ast::Visibility) {
+        vis.tokens = None;
+        mut_visit::noop_visit_vis(vis, self);
+    }
+    fn visit_expr(&mut self, e: &mut P<ast::Expr>) {
+        e.tokens = None;
+        mut_visit::noop_visit_expr(e, self);
+    }
+    fn visit_mac(&mut self, _mac: &mut ast::MacCall) {}
+}
+
 pub fn parse<'a>(sess: &'a Session, input: &Input) -> PResult<'a, ast::Crate> {
     let krate = sess.time("parse_crate", || match input {
         Input::File(file) => parse_crate_from_file(file, &sess.parse_sess),
@@ -59,6 +119,10 @@ pub fn parse<'a>(sess: &'a Session, input: &Input) -> PResult<'a, ast::Crate> {
     })?;
 
     if sess.opts.debugging_opts.ast_json_noexpand {
+        // Set any `token` fields to `None` before
+        // we display the AST.
+        let mut krate = krate.clone();
+        TokenStripper.visit_crate(&mut krate);
         println!("{}", json::as_json(&krate));
     }
 
@@ -379,6 +443,10 @@ fn configure_and_expand_inner<'a>(
     }
 
     if sess.opts.debugging_opts.ast_json {
+        // Set any `token` fields to `None` before
+        // we display the AST.
+        let mut krate = krate.clone();
+        TokenStripper.visit_crate(&mut krate);
         println!("{}", json::as_json(&krate));
     }
 

compiler/rustc_middle/src/ich/impls_syntax.rs

@@ -40,11 +40,12 @@ impl<'ctx> rustc_ast::HashStableContext for StableHashingContext<'ctx> {
         debug_assert!(!attr.ident().map_or(false, |ident| self.is_ignored_attr(ident.name)));
         debug_assert!(!attr.is_doc_comment());
 
-        let ast::Attribute { kind, id: _, style, span } = attr;
+        let ast::Attribute { kind, id: _, style, span, tokens } = attr;
         if let ast::AttrKind::Normal(item) = kind {
             item.hash_stable(self, hasher);
             style.hash_stable(self, hasher);
             span.hash_stable(self, hasher);
+            tokens.as_ref().expect_none("Tokens should have been removed during lowering!");
         } else {
             unreachable!();
         }

compiler/rustc_parse/src/lib.rs

@@ -252,9 +252,7 @@ pub fn nt_to_tokenstream(nt: &Nonterminal, sess: &ParseSess, span: Span) -> Toke
     let convert_tokens = |tokens: Option<LazyTokenStream>| tokens.map(|t| t.into_token_stream());
 
     let tokens = match *nt {
-        Nonterminal::NtItem(ref item) => {
+        Nonterminal::NtItem(ref item) => prepend_attrs(&item.attrs, item.tokens.as_ref()),
-            prepend_attrs(sess, &item.attrs, item.tokens.as_ref(), span)
-        }
         Nonterminal::NtBlock(ref block) => convert_tokens(block.tokens.clone()),
         Nonterminal::NtStmt(ref stmt) => {
             // FIXME: We currently only collect tokens for `:stmt`
@@ -279,7 +277,7 @@ pub fn nt_to_tokenstream(nt: &Nonterminal, sess: &ParseSess, span: Span) -> Toke
             if expr.tokens.is_none() {
                 debug!("missing tokens for expr {:?}", expr);
             }
-            prepend_attrs(sess, &expr.attrs, expr.tokens.as_ref(), span)
+            prepend_attrs(&expr.attrs, expr.tokens.as_ref())
         }
     };
 
@@ -603,10 +601,8 @@ fn token_probably_equal_for_proc_macro(first: &Token, other: &Token) -> bool {
 }
 
 fn prepend_attrs(
-    sess: &ParseSess,
     attrs: &[ast::Attribute],
     tokens: Option<&tokenstream::LazyTokenStream>,
-    span: rustc_span::Span,
 ) -> Option<tokenstream::TokenStream> {
     let tokens = tokens?.clone().into_token_stream();
     if attrs.is_empty() {
@@ -619,47 +615,12 @@ fn prepend_attrs(
             ast::AttrStyle::Outer,
             "inner attributes should prevent cached tokens from existing"
         );
-
+        builder.push(
-        let source = pprust::attribute_to_string(attr);
+            attr.tokens
-        let macro_filename = FileName::macro_expansion_source_code(&source);
+                .clone()
-
+                .unwrap_or_else(|| panic!("Attribute {:?} is missing tokens!", attr))
-        let item = match attr.kind {
+                .into_token_stream(),
-            ast::AttrKind::Normal(ref item) => item,
+        );
-            ast::AttrKind::DocComment(..) => {
-                let stream = parse_stream_from_source_str(macro_filename, source, sess, Some(span));
-                builder.push(stream);
-                continue;
-            }
-        };
-
-        // synthesize # [ $path $tokens ] manually here
-        let mut brackets = tokenstream::TokenStreamBuilder::new();
-
-        // For simple paths, push the identifier directly
-        if item.path.segments.len() == 1 && item.path.segments[0].args.is_none() {
-            let ident = item.path.segments[0].ident;
-            let token = token::Ident(ident.name, ident.as_str().starts_with("r#"));
-            brackets.push(tokenstream::TokenTree::token(token, ident.span));
-
-        // ... and for more complicated paths, fall back to a reparse hack that
-        // should eventually be removed.
-        } else {
-            let stream = parse_stream_from_source_str(macro_filename, source, sess, Some(span));
-            brackets.push(stream);
-        }
-
-        brackets.push(item.args.outer_tokens());
-
-        // The span we list here for `#` and for `[ ... ]` are both wrong in
-        // that it encompasses more than each token, but it hopefully is "good
-        // enough" for now at least.
-        builder.push(tokenstream::TokenTree::token(token::Pound, attr.span));
-        let delim_span = tokenstream::DelimSpan::from_single(attr.span);
-        builder.push(tokenstream::TokenTree::Delimited(
-            delim_span,
-            token::DelimToken::Bracket,
-            brackets.build(),
-        ));
     }
     builder.push(tokens.clone());
     Some(builder.build())

compiler/rustc_parse/src/parser/attr.rs

@@ -30,41 +30,53 @@ impl<'a> Parser<'a> {
         let mut just_parsed_doc_comment = false;
         loop {
             debug!("parse_outer_attributes: self.token={:?}", self.token);
-            if self.check(&token::Pound) {
+            let (attr, tokens) = if self.check(&token::Pound) {
-                let inner_error_reason = if just_parsed_doc_comment {
+                self.collect_tokens(|this| {
-                    "an inner attribute is not permitted following an outer doc comment"
+                    let inner_error_reason = if just_parsed_doc_comment {
-                } else if !attrs.is_empty() {
+                        "an inner attribute is not permitted following an outer doc comment"
-                    "an inner attribute is not permitted following an outer attribute"
+                    } else if !attrs.is_empty() {
-                } else {
+                        "an inner attribute is not permitted following an outer attribute"
-                    DEFAULT_UNEXPECTED_INNER_ATTR_ERR_MSG
+                    } else {
-                };
+                        DEFAULT_UNEXPECTED_INNER_ATTR_ERR_MSG
-                let inner_parse_policy = InnerAttrPolicy::Forbidden {
+                    };
-                    reason: inner_error_reason,
+                    let inner_parse_policy = InnerAttrPolicy::Forbidden {
-                    saw_doc_comment: just_parsed_doc_comment,
+                        reason: inner_error_reason,
-                    prev_attr_sp: attrs.last().map(|a| a.span),
+                        saw_doc_comment: just_parsed_doc_comment,
-                };
+                        prev_attr_sp: attrs.last().map(|a| a.span),
-                let attr = self.parse_attribute_with_inner_parse_policy(inner_parse_policy)?;
+                    };
-                attrs.push(attr);
+                    let attr = this.parse_attribute_with_inner_parse_policy(inner_parse_policy)?;
-                just_parsed_doc_comment = false;
+                    just_parsed_doc_comment = false;
+                    Ok(Some(attr))
+                })?
             } else if let token::DocComment(comment_kind, attr_style, data) = self.token.kind {
-                let attr = attr::mk_doc_comment(comment_kind, attr_style, data, self.token.span);
+                self.collect_tokens(|this| {
-                if attr.style != ast::AttrStyle::Outer {
+                    let attr =
-                    self.sess
+                        attr::mk_doc_comment(comment_kind, attr_style, data, this.token.span);
-                        .span_diagnostic
+                    if attr.style != ast::AttrStyle::Outer {
-                        .struct_span_err_with_code(
+                        this.sess
-                            self.token.span,
+                            .span_diagnostic
-                            "expected outer doc comment",
+                            .struct_span_err_with_code(
-                            error_code!(E0753),
+                                this.token.span,
-                        )
+                                "expected outer doc comment",
-                        .note(
+                                error_code!(E0753),
-                            "inner doc comments like this (starting with \
+                            )
-                             `//!` or `/*!`) can only appear before items",
+                            .note(
-                        )
+                                "inner doc comments like this (starting with \
-                        .emit();
+                                 `//!` or `/*!`) can only appear before items",
-                }
+                            )
+                            .emit();
+                    }
+                    this.bump();
+                    just_parsed_doc_comment = true;
+                    Ok(Some(attr))
+                })?
+            } else {
+                (None, None)
+            };
+
+            if let Some(mut attr) = attr {
+                attr.tokens = tokens;
                 attrs.push(attr);
-                self.bump();
-                just_parsed_doc_comment = true;
             } else {
                 break;
             }
@@ -99,7 +111,7 @@ impl<'a> Parser<'a> {
                 if self.eat(&token::Not) { ast::AttrStyle::Inner } else { ast::AttrStyle::Outer };
 
             self.expect(&token::OpenDelim(token::Bracket))?;
-            let item = self.parse_attr_item()?;
+            let item = self.parse_attr_item(false)?;
             self.expect(&token::CloseDelim(token::Bracket))?;
             let attr_sp = lo.to(self.prev_token.span);
 
@@ -148,7 +160,7 @@ impl<'a> Parser<'a> {
     ///     PATH
     ///     PATH `=` UNSUFFIXED_LIT
     /// The delimiters or `=` are still put into the resulting token stream.
-    pub fn parse_attr_item(&mut self) -> PResult<'a, ast::AttrItem> {
+    pub fn parse_attr_item(&mut self, capture_tokens: bool) -> PResult<'a, ast::AttrItem> {
         let item = match self.token.kind {
             token::Interpolated(ref nt) => match **nt {
                 Nonterminal::NtMeta(ref item) => Some(item.clone().into_inner()),
@@ -160,9 +172,18 @@ impl<'a> Parser<'a> {
             self.bump();
             item
         } else {
-            let path = self.parse_path(PathStyle::Mod)?;
+            let do_parse = |this: &mut Self| {
-            let args = self.parse_attr_args()?;
+                let path = this.parse_path(PathStyle::Mod)?;
-            ast::AttrItem { path, args, tokens: None }
+                let args = this.parse_attr_args()?;
+                Ok(ast::AttrItem { path, args, tokens: None })
+            };
+            if capture_tokens {
+                let (mut item, tokens) = self.collect_tokens(do_parse)?;
+                item.tokens = tokens;
+                item
+            } else {
+                do_parse(self)?
+            }
         })
     }
 
@@ -175,19 +196,31 @@ impl<'a> Parser<'a> {
         let mut attrs: Vec<ast::Attribute> = vec![];
         loop {
             // Only try to parse if it is an inner attribute (has `!`).
-            if self.check(&token::Pound) && self.look_ahead(1, |t| t == &token::Not) {
+            let (attr, tokens) =
-                let attr = self.parse_attribute(true)?;
+                if self.check(&token::Pound) && self.look_ahead(1, |t| t == &token::Not) {
-                assert_eq!(attr.style, ast::AttrStyle::Inner);
+                    self.collect_tokens(|this| {
-                attrs.push(attr);
+                        let attr = this.parse_attribute(true)?;
-            } else if let token::DocComment(comment_kind, attr_style, data) = self.token.kind {
+                        assert_eq!(attr.style, ast::AttrStyle::Inner);
-                // We need to get the position of this token before we bump.
+                        Ok(Some(attr))
-                let attr = attr::mk_doc_comment(comment_kind, attr_style, data, self.token.span);
+                    })?
-                if attr.style == ast::AttrStyle::Inner {
+                } else if let token::DocComment(comment_kind, attr_style, data) = self.token.kind {
-                    attrs.push(attr);
+                    self.collect_tokens(|this| {
-                    self.bump();
+                        // We need to get the position of this token before we bump.
+                        let attr =
+                            attr::mk_doc_comment(comment_kind, attr_style, data, this.token.span);
+                        if attr.style == ast::AttrStyle::Inner {
+                            this.bump();
+                            Ok(Some(attr))
+                        } else {
+                            Ok(None)
+                        }
+                    })?
                 } else {
-                    break;
+                    (None, None)
-                }
+                };
+            if let Some(mut attr) = attr {
+                attr.tokens = tokens;
+                attrs.push(attr);
             } else {
                 break;
             }
@@ -220,7 +253,7 @@ impl<'a> Parser<'a> {
         let mut expanded_attrs = Vec::with_capacity(1);
         while self.token.kind != token::Eof {
             let lo = self.token.span;
-            let item = self.parse_attr_item()?;
+            let item = self.parse_attr_item(true)?;
             expanded_attrs.push((item, lo.to(self.prev_token.span)));
             if !self.eat(&token::Comma) {
                 break;

compiler/rustc_parse/src/parser/expr.rs

@@ -1116,7 +1116,7 @@ impl<'a> Parser<'a> {
     ) -> PResult<'a, P<Expr>> {
         if needs_tokens {
             let (mut expr, tokens) = self.collect_tokens(f)?;
-            expr.tokens = Some(tokens);
+            expr.tokens = tokens;
             Ok(expr)
         } else {
             f(self)

compiler/rustc_parse/src/parser/item.rs

@@ -151,7 +151,7 @@ impl<'a> Parser<'a> {
         if let Some(tokens) = tokens {
             if let Some(item) = &mut item {
                 if !item.attrs.iter().any(|attr| attr.style == AttrStyle::Inner) {
-                    item.tokens = Some(tokens);
+                    item.tokens = tokens;
                 }
             }
         }

compiler/rustc_parse/src/parser/mod.rs

@@ -1178,8 +1178,9 @@ impl<'a> Parser<'a> {
 
     /// Records all tokens consumed by the provided callback,
     /// including the current token. These tokens are collected
-    /// into a `TokenStream`, and returned along with the result
+    /// into a `LazyTokenStream`, and returned along with the result
-    /// of the callback.
+    /// of the callback. The returned `LazyTokenStream` will be `None`
+    /// if not tokens were captured.
     ///
     /// Note: If your callback consumes an opening delimiter
     /// (including the case where you call `collect_tokens`
@@ -1195,7 +1196,7 @@ impl<'a> Parser<'a> {
     pub fn collect_tokens<R>(
         &mut self,
         f: impl FnOnce(&mut Self) -> PResult<'a, R>,
-    ) -> PResult<'a, (R, LazyTokenStream)> {
+    ) -> PResult<'a, (R, Option<LazyTokenStream>)> {
         let start_token = (self.token.clone(), self.token_spacing);
         let mut cursor_snapshot = self.token_cursor.clone();
 
@@ -1205,6 +1206,11 @@ impl<'a> Parser<'a> {
         let num_calls = new_calls - cursor_snapshot.num_next_calls;
         let desugar_doc_comments = self.desugar_doc_comments;
 
+        // We didn't capture any tokens
+        if num_calls == 0 {
+            return Ok((ret, None));
+        }
+
         // Produces a `TokenStream` on-demand. Using `cursor_snapshot`
         // and `num_calls`, we can reconstruct the `TokenStream` seen
         // by the callback. This allows us to avoid producing a `TokenStream`
@@ -1233,7 +1239,7 @@ impl<'a> Parser<'a> {
         };
         let stream = LazyTokenStream::new(LazyTokenStreamInner::Lazy(Box::new(lazy_cb)));
 
-        Ok((ret, stream))
+        Ok((ret, Some(stream)))
     }
 
     /// `::{` or `::*`

compiler/rustc_parse/src/parser/nonterminal.rs

@@ -103,7 +103,7 @@ impl<'a> Parser<'a> {
                     // If we captured tokens during parsing (due to outer attributes),
                     // use those.
                     if item.tokens.is_none() {
-                        item.tokens = Some(tokens);
+                        item.tokens = tokens;
                     }
                     token::NtItem(item)
                 }
@@ -115,7 +115,7 @@ impl<'a> Parser<'a> {
                 let (mut block, tokens) = self.collect_tokens(|this| this.parse_block())?;
                 // We have have eaten an NtBlock, which could already have tokens
                 if block.tokens.is_none() {
-                    block.tokens = Some(tokens);
+                    block.tokens = tokens;
                 }
                 token::NtBlock(block)
             }
@@ -124,7 +124,7 @@ impl<'a> Parser<'a> {
                 match stmt {
                     Some(mut s) => {
                         if s.tokens.is_none() {
-                            s.tokens = Some(tokens);
+                            s.tokens = tokens;
                         }
                         token::NtStmt(s)
                     }
@@ -137,7 +137,7 @@ impl<'a> Parser<'a> {
                 let (mut pat, tokens) = self.collect_tokens(|this| this.parse_pat(None))?;
                 // We have have eaten an NtPat, which could already have tokens
                 if pat.tokens.is_none() {
-                    pat.tokens = Some(tokens);
+                    pat.tokens = tokens;
                 }
                 token::NtPat(pat)
             }
@@ -146,7 +146,7 @@ impl<'a> Parser<'a> {
                 // If we captured tokens during parsing (due to outer attributes),
                 // use those.
                 if expr.tokens.is_none() {
-                    expr.tokens = Some(tokens);
+                    expr.tokens = tokens;
                 }
                 token::NtExpr(expr)
             }
@@ -155,7 +155,7 @@ impl<'a> Parser<'a> {
                     self.collect_tokens(|this| this.parse_literal_maybe_minus())?;
                 // We have have eaten a nonterminal, which  could already have tokens
                 if lit.tokens.is_none() {
-                    lit.tokens = Some(tokens);
+                    lit.tokens = tokens;
                 }
                 token::NtLiteral(lit)
             }
@@ -163,7 +163,7 @@ impl<'a> Parser<'a> {
                 let (mut ty, tokens) = self.collect_tokens(|this| this.parse_ty())?;
                 // We have an eaten an NtTy, which could already have tokens
                 if ty.tokens.is_none() {
-                    ty.tokens = Some(tokens);
+                    ty.tokens = tokens;
                 }
                 token::NtTy(ty)
             }
@@ -183,15 +183,15 @@ impl<'a> Parser<'a> {
                     self.collect_tokens(|this| this.parse_path(PathStyle::Type))?;
                 // We have have eaten an NtPath, which could already have tokens
                 if path.tokens.is_none() {
-                    path.tokens = Some(tokens);
+                    path.tokens = tokens;
                 }
                 token::NtPath(path)
             }
             NonterminalKind::Meta => {
-                let (mut attr, tokens) = self.collect_tokens(|this| this.parse_attr_item())?;
+                let (mut attr, tokens) = self.collect_tokens(|this| this.parse_attr_item(false))?;
                 // We may have eaten a nonterminal, which could already have tokens
                 if attr.tokens.is_none() {
-                    attr.tokens = Some(tokens);
+                    attr.tokens = tokens;
                 }
                 token::NtMeta(P(attr))
             }
@@ -201,7 +201,7 @@ impl<'a> Parser<'a> {
                     self.collect_tokens(|this| this.parse_visibility(FollowedByType::Yes))?;
                 // We may have etan an `NtVis`, which could already have tokens
                 if vis.tokens.is_none() {
-                    vis.tokens = Some(tokens);
+                    vis.tokens = tokens;
                 }
                 token::NtVis(vis)
             }

src/test/ui/ast-json/ast-json-noexpand-output.stdout

@@ -1 +1 @@
-{"module":{"inner":{"lo":0,"hi":0},"unsafety":"No","items":[{"attrs":[],"id":0,"span":{"lo":0,"hi":0},"vis":{"kind":"Inherited","span":{"lo":0,"hi":0},"tokens":null},"ident":{"name":"core","span":{"lo":0,"hi":0}},"kind":{"variant":"ExternCrate","fields":[null]},"tokens":null}],"inline":true},"attrs":[{"kind":{"variant":"Normal","fields":[{"path":{"span":{"lo":0,"hi":0},"segments":[{"ident":{"name":"crate_type","span":{"lo":0,"hi":0}},"id":0,"args":null}],"tokens":null},"args":{"variant":"Eq","fields":[{"lo":0,"hi":0},{"0":[[{"variant":"Token","fields":[{"kind":{"variant":"Literal","fields":[{"kind":"Str","symbol":"lib","suffix":null}]},"span":{"lo":0,"hi":0}}]},"Alone"]]}]},"tokens":null}]},"id":null,"style":"Inner","span":{"lo":0,"hi":0}}],"span":{"lo":0,"hi":0},"proc_macros":[]}
+{"module":{"inner":{"lo":0,"hi":0},"unsafety":"No","items":[{"attrs":[],"id":0,"span":{"lo":0,"hi":0},"vis":{"kind":"Inherited","span":{"lo":0,"hi":0},"tokens":null},"ident":{"name":"core","span":{"lo":0,"hi":0}},"kind":{"variant":"ExternCrate","fields":[null]},"tokens":null}],"inline":true},"attrs":[{"kind":{"variant":"Normal","fields":[{"path":{"span":{"lo":0,"hi":0},"segments":[{"ident":{"name":"crate_type","span":{"lo":0,"hi":0}},"id":0,"args":null}],"tokens":null},"args":{"variant":"Eq","fields":[{"lo":0,"hi":0},{"0":[[{"variant":"Token","fields":[{"kind":{"variant":"Literal","fields":[{"kind":"Str","symbol":"lib","suffix":null}]},"span":{"lo":0,"hi":0}}]},"Alone"]]}]},"tokens":null}]},"id":null,"style":"Inner","span":{"lo":0,"hi":0},"tokens":null}],"span":{"lo":0,"hi":0},"proc_macros":[]}

src/test/ui/ast-json/ast-json-output.stdout

@@ -1 +1 @@
-{"module":{"inner":{"lo":0,"hi":0},"unsafety":"No","items":[{"attrs":[{"kind":{"variant":"Normal","fields":[{"path":{"span":{"lo":0,"hi":0},"segments":[{"ident":{"name":"prelude_import","span":{"lo":0,"hi":0}},"id":0,"args":null}],"tokens":null},"args":"Empty","tokens":null}]},"id":null,"style":"Outer","span":{"lo":0,"hi":0}}],"id":0,"span":{"lo":0,"hi":0},"vis":{"kind":"Inherited","span":{"lo":0,"hi":0},"tokens":null},"ident":{"name":"","span":{"lo":0,"hi":0}},"kind":{"variant":"Use","fields":[{"prefix":{"span":{"lo":0,"hi":0},"segments":[{"ident":{"name":"{{root}}","span":{"lo":0,"hi":0}},"id":0,"args":null},{"ident":{"name":"std","span":{"lo":0,"hi":0}},"id":0,"args":null},{"ident":{"name":"prelude","span":{"lo":0,"hi":0}},"id":0,"args":null},{"ident":{"name":"v1","span":{"lo":0,"hi":0}},"id":0,"args":null}],"tokens":null},"kind":"Glob","span":{"lo":0,"hi":0}}]},"tokens":null},{"attrs":[{"kind":{"variant":"Normal","fields":[{"path":{"span":{"lo":0,"hi":0},"segments":[{"ident":{"name":"macro_use","span":{"lo":0,"hi":0}},"id":0,"args":null}],"tokens":null},"args":"Empty","tokens":null}]},"id":null,"style":"Outer","span":{"lo":0,"hi":0}}],"id":0,"span":{"lo":0,"hi":0},"vis":{"kind":"Inherited","span":{"lo":0,"hi":0},"tokens":null},"ident":{"name":"std","span":{"lo":0,"hi":0}},"kind":{"variant":"ExternCrate","fields":[null]},"tokens":null},{"attrs":[],"id":0,"span":{"lo":0,"hi":0},"vis":{"kind":"Inherited","span":{"lo":0,"hi":0},"tokens":null},"ident":{"name":"core","span":{"lo":0,"hi":0}},"kind":{"variant":"ExternCrate","fields":[null]},"tokens":null}],"inline":true},"attrs":[{"kind":{"variant":"Normal","fields":[{"path":{"span":{"lo":0,"hi":0},"segments":[{"ident":{"name":"crate_type","span":{"lo":0,"hi":0}},"id":0,"args":null}],"tokens":null},"args":{"variant":"Eq","fields":[{"lo":0,"hi":0},{"0":[[{"variant":"Token","fields":[{"kind":{"variant":"Literal","fields":[{"kind":"Str","symbol":"lib","suffix":null}]},"span":{"lo":0,"hi":0}}]},"Alone"]]}]},"tokens":null}]},"id":null,"style":"Inner","span":{"lo":0,"hi":0}}],"span":{"lo":0,"hi":0},"proc_macros":[]}
+{"module":{"inner":{"lo":0,"hi":0},"unsafety":"No","items":[{"attrs":[{"kind":{"variant":"Normal","fields":[{"path":{"span":{"lo":0,"hi":0},"segments":[{"ident":{"name":"prelude_import","span":{"lo":0,"hi":0}},"id":0,"args":null}],"tokens":null},"args":"Empty","tokens":null}]},"id":null,"style":"Outer","span":{"lo":0,"hi":0},"tokens":null}],"id":0,"span":{"lo":0,"hi":0},"vis":{"kind":"Inherited","span":{"lo":0,"hi":0},"tokens":null},"ident":{"name":"","span":{"lo":0,"hi":0}},"kind":{"variant":"Use","fields":[{"prefix":{"span":{"lo":0,"hi":0},"segments":[{"ident":{"name":"{{root}}","span":{"lo":0,"hi":0}},"id":0,"args":null},{"ident":{"name":"std","span":{"lo":0,"hi":0}},"id":0,"args":null},{"ident":{"name":"prelude","span":{"lo":0,"hi":0}},"id":0,"args":null},{"ident":{"name":"v1","span":{"lo":0,"hi":0}},"id":0,"args":null}],"tokens":null},"kind":"Glob","span":{"lo":0,"hi":0}}]},"tokens":null},{"attrs":[{"kind":{"variant":"Normal","fields":[{"path":{"span":{"lo":0,"hi":0},"segments":[{"ident":{"name":"macro_use","span":{"lo":0,"hi":0}},"id":0,"args":null}],"tokens":null},"args":"Empty","tokens":null}]},"id":null,"style":"Outer","span":{"lo":0,"hi":0},"tokens":null}],"id":0,"span":{"lo":0,"hi":0},"vis":{"kind":"Inherited","span":{"lo":0,"hi":0},"tokens":null},"ident":{"name":"std","span":{"lo":0,"hi":0}},"kind":{"variant":"ExternCrate","fields":[null]},"tokens":null},{"attrs":[],"id":0,"span":{"lo":0,"hi":0},"vis":{"kind":"Inherited","span":{"lo":0,"hi":0},"tokens":null},"ident":{"name":"core","span":{"lo":0,"hi":0}},"kind":{"variant":"ExternCrate","fields":[null]},"tokens":null}],"inline":true},"attrs":[{"kind":{"variant":"Normal","fields":[{"path":{"span":{"lo":0,"hi":0},"segments":[{"ident":{"name":"crate_type","span":{"lo":0,"hi":0}},"id":0,"args":null}],"tokens":null},"args":{"variant":"Eq","fields":[{"lo":0,"hi":0},{"0":[[{"variant":"Token","fields":[{"kind":{"variant":"Literal","fields":[{"kind":"Str","symbol":"lib","suffix":null}]},"span":{"lo":0,"hi":0}}]},"Alone"]]}]},"tokens":null}]},"id":null,"style":"Inner","span":{"lo":0,"hi":0},"tokens":null}],"span":{"lo":0,"hi":0},"proc_macros":[]}

src/test/ui/proc-macro/issue-75930-derive-cfg.stdout

@@ -26,7 +26,7 @@ PRINT-ATTR INPUT (DEBUG): TokenStream [
     Punct {
         ch: '#',
         spacing: Alone,
-        span: $DIR/issue-75930-derive-cfg.rs:17:24: 17:40 (#0),
+        span: $DIR/issue-75930-derive-cfg.rs:17:1: 17:2 (#0),
     },
     Group {
         delimiter: Bracket,
@@ -46,12 +46,12 @@ PRINT-ATTR INPUT (DEBUG): TokenStream [
                 span: $DIR/issue-75930-derive-cfg.rs:17:29: 17:40 (#0),
             },
         ],
-        span: $DIR/issue-75930-derive-cfg.rs:17:24: 17:40 (#0),
+        span: $DIR/issue-75930-derive-cfg.rs:17:1: 17:2 (#0),
     },
     Punct {
         ch: '#',
         spacing: Alone,
-        span: $DIR/issue-75930-derive-cfg.rs:19:1: 19:17 (#0),
+        span: $DIR/issue-75930-derive-cfg.rs:19:1: 19:2 (#0),
     },
     Group {
         delimiter: Bracket,
@@ -71,12 +71,12 @@ PRINT-ATTR INPUT (DEBUG): TokenStream [
                 span: $DIR/issue-75930-derive-cfg.rs:19:9: 19:16 (#0),
             },
         ],
-        span: $DIR/issue-75930-derive-cfg.rs:19:1: 19:17 (#0),
+        span: $DIR/issue-75930-derive-cfg.rs:19:2: 19:17 (#0),
     },
     Punct {
         ch: '#',
         spacing: Alone,
-        span: $DIR/issue-75930-derive-cfg.rs:20:1: 20:19 (#0),
+        span: $DIR/issue-75930-derive-cfg.rs:20:1: 20:2 (#0),
     },
     Group {
         delimiter: Bracket,
@@ -96,12 +96,12 @@ PRINT-ATTR INPUT (DEBUG): TokenStream [
                 span: $DIR/issue-75930-derive-cfg.rs:20:15: 20:18 (#0),
             },
         ],
-        span: $DIR/issue-75930-derive-cfg.rs:20:1: 20:19 (#0),
+        span: $DIR/issue-75930-derive-cfg.rs:20:2: 20:19 (#0),
     },
     Punct {
         ch: '#',
         spacing: Alone,
-        span: $DIR/issue-75930-derive-cfg.rs:16:1: 16:19 (#0),
+        span: $DIR/issue-75930-derive-cfg.rs:16:1: 16:2 (#0),
     },
     Group {
         delimiter: Bracket,
@@ -121,7 +121,7 @@ PRINT-ATTR INPUT (DEBUG): TokenStream [
                 span: $DIR/issue-75930-derive-cfg.rs:16:15: 16:18 (#0),
             },
         ],
-        span: $DIR/issue-75930-derive-cfg.rs:16:1: 16:19 (#0),
+        span: $DIR/issue-75930-derive-cfg.rs:16:2: 16:19 (#0),
     },
     Ident {
         ident: "struct",