Don't mutate syntax trees when preparing proc-macro input

Author: Veykril

Cargo.lock

@@ -14,6 +14,7 @@ log = "0.4.8"
 either = "1.5.3"
 rustc-hash = "1.0.0"
 la-arena = { version = "0.2.0", path = "../../lib/arena" }
+itertools = "0.10.0"
 
 base_db = { path = "../base_db", version = "0.0.0" }
 cfg = { path = "../cfg", version = "0.0.0" }

crates/hir_expand/Cargo.toml

@@ -3,19 +3,20 @@
 use std::sync::Arc;
 
 use base_db::{salsa, SourceDatabase};
+use itertools::Itertools;
 use limit::Limit;
 use mbe::{ExpandError, ExpandResult};
 use parser::{FragmentKind, T};
 use syntax::{
     algo::diff,
-    ast::{self, NameOwner},
-    AstNode, GreenNode, Parse, SyntaxNode, SyntaxToken,
+    ast::{self, AttrsOwner, NameOwner},
+    AstNode, GreenNode, Parse, SyntaxNode, SyntaxToken, TextRange,
 };
 
 use crate::{
-    ast_id_map::AstIdMap, hygiene::HygieneFrame, input::process_macro_input, BuiltinAttrExpander,
-    BuiltinDeriveExpander, BuiltinFnLikeExpander, HirFileId, HirFileIdRepr, MacroCallId,
-    MacroCallKind, MacroCallLoc, MacroDefId, MacroDefKind, MacroFile, ProcMacroExpander,
+    ast_id_map::AstIdMap, hygiene::HygieneFrame, BuiltinAttrExpander, BuiltinDeriveExpander,
+    BuiltinFnLikeExpander, HirFileId, HirFileIdRepr, MacroCallId, MacroCallKind, MacroCallLoc,
+    MacroDefId, MacroDefKind, MacroFile, ProcMacroExpander,
 };
 
 /// Total limit on the number of tokens produced by any macro invocation.
@@ -257,9 +258,28 @@ fn parse_macro_expansion(
 
 fn macro_arg(db: &dyn AstDatabase, id: MacroCallId) -> Option<Arc<(tt::Subtree, mbe::TokenMap)>> {
     let arg = db.macro_arg_text(id)?;
-    let (mut tt, tmap) = mbe::syntax_node_to_token_tree(&SyntaxNode::new_root(arg));
+    let loc = db.lookup_intern_macro(id);
+
+    let node = SyntaxNode::new_root(arg);
+    let censor = match loc.kind {
+        MacroCallKind::FnLike { .. } => None,
+        MacroCallKind::Derive { derive_attr_index, .. } => match ast::Item::cast(node.clone()) {
+            Some(item) => item
+                .attrs()
+                .map(|attr| attr.syntax().text_range())
+                .take(derive_attr_index as usize + 1)
+                .fold1(TextRange::cover),
+            None => None,
+        },
+        MacroCallKind::Attr { invoc_attr_index, .. } => match ast::Item::cast(node.clone()) {
+            Some(item) => {
+                item.attrs().nth(invoc_attr_index as usize).map(|attr| attr.syntax().text_range())
+            }
+            None => None,
+        },
+    };
+    let (mut tt, tmap) = mbe::syntax_node_to_token_tree_censored(&node, censor);
 
-    let loc: MacroCallLoc = db.lookup_intern_macro(id);
     if loc.def.is_proc_macro() {
         // proc macros expect their inputs without parentheses, MBEs expect it with them included
         tt.delimiter = None;
@@ -271,7 +291,6 @@ fn macro_arg(db: &dyn AstDatabase, id: MacroCallId) -> Option<Arc<(tt::Subtree,
 fn macro_arg_text(db: &dyn AstDatabase, id: MacroCallId) -> Option<GreenNode> {
     let loc = db.lookup_intern_macro(id);
     let arg = loc.kind.arg(db)?;
-    let arg = process_macro_input(&loc.kind, arg);
     if matches!(loc.kind, MacroCallKind::FnLike { .. }) {
         let first = arg.first_child_or_token().map_or(T![.], |it| it.kind());
         let last = arg.last_child_or_token().map_or(T![.], |it| it.kind());

crates/hir_expand/src/db.rs

@@ -14,7 +14,6 @@ pub mod builtin_macro;
 pub mod proc_macro;
 pub mod quote;
 pub mod eager;
-mod input;
 
 use base_db::ProcMacroKind;
 use either::Either;

crates/hir_expand/src/input.rs

@@ -67,7 +67,7 @@ impl fmt::Display for ExpandError {
 pub use crate::{
     syntax_bridge::{
         parse_exprs_with_sep, parse_to_token_tree, syntax_node_to_token_tree,
-        token_tree_to_syntax_node,
+        syntax_node_to_token_tree_censored, token_tree_to_syntax_node,
     },
     token_map::TokenMap,
 };

crates/hir_expand/src/lib.rs

@@ -1,5 +1,7 @@
 //! Conversions between [`SyntaxNode`] and [`tt::TokenTree`].
 
+use std::iter;
+
 use parser::{FragmentKind, ParseError, TreeSink};
 use rustc_hash::FxHashMap;
 use syntax::{
@@ -13,11 +15,20 @@ use tt::buffer::{Cursor, TokenBuffer};
 use crate::{subtree_source::SubtreeTokenSource, tt_iter::TtIter};
 use crate::{ExpandError, TokenMap};
 
-/// Convert the syntax node to a `TokenTree` (what macro
+/// Convert the syntax node to a `TokenTree` with the censored nodes excluded (what macro
 /// will consume).
 pub fn syntax_node_to_token_tree(node: &SyntaxNode) -> (tt::Subtree, TokenMap) {
+    syntax_node_to_token_tree_censored(node, None)
+}
+
+/// Convert the syntax node to a `TokenTree` with the censored nodes excluded (what macro
+/// will consume).
+pub fn syntax_node_to_token_tree_censored(
+    node: &SyntaxNode,
+    censor: Option<TextRange>,
+) -> (tt::Subtree, TokenMap) {
     let global_offset = node.text_range().start();
-    let mut c = Convertor::new(node, global_offset);
+    let mut c = Convertor::new(node, global_offset, censor);
     let subtree = convert_tokens(&mut c);
     c.id_alloc.map.shrink_to_fit();
     (subtree, c.id_alloc.map)
@@ -446,16 +457,24 @@ impl<'a> TokenConvertor for RawConvertor<'a> {
 struct Convertor {
     id_alloc: TokenIdAlloc,
     current: Option<SyntaxToken>,
+    censor: Option<TextRange>,
     range: TextRange,
     punct_offset: Option<(SyntaxToken, TextSize)>,
 }
 
 impl Convertor {
-    fn new(node: &SyntaxNode, global_offset: TextSize) -> Convertor {
+    fn new(node: &SyntaxNode, global_offset: TextSize, censor: Option<TextRange>) -> Convertor {
+        let first = node.first_token();
+        let current = match censor {
+            Some(censor) => iter::successors(first, |token| token.next_token())
+                .find(|token| !censor.contains_range(token.text_range())),
+            None => first,
+        };
         Convertor {
             id_alloc: { TokenIdAlloc { map: TokenMap::default(), global_offset, next_id: 0 } },
-            current: node.first_token(),
+            current,
             range: node.text_range(),
+            censor,
             punct_offset: None,
         }
     }
@@ -512,8 +531,11 @@ impl TokenConvertor for Convertor {
         if !&self.range.contains_range(curr.text_range()) {
             return None;
         }
-        self.current = curr.next_token();
-
+        self.current = match self.censor {
+            Some(censor) => iter::successors(curr.next_token(), |token| token.next_token())
+                .find(|token| !censor.contains_range(token.text_range())),
+            None => curr.next_token(),
+        };
         let token = if curr.kind().is_punct() {
             let range = curr.text_range();
             let range = TextRange::at(range.start(), TextSize::of('.'));