Merge #3664

3664: Introduce TokenConverter Trait r=matklad a=edwin0cheng

This PR add a `TokenConverter` Trait to share the conversion logic between raw `lexer` token and Syntax Node Token.

Related #2158.

Co-authored-by: Edwin Cheng <edwin0cheng@gmail.com>

crates/ra_mbe/src/syntax_bridge.rs

@@ -3,12 +3,11 @@
 use ra_parser::{FragmentKind, ParseError, TreeSink};
 use ra_syntax::{
     ast::{self, make::tokens::doc_comment},
-    tokenize, AstToken, NodeOrToken, Parse, SmolStr, SyntaxKind,
+    tokenize, AstToken, Parse, SmolStr, SyntaxKind,
     SyntaxKind::*,
-    SyntaxNode, SyntaxTreeBuilder, TextRange, TextUnit, Token, T,
+    SyntaxNode, SyntaxToken, SyntaxTreeBuilder, TextRange, TextUnit, Token as RawToken, T,
 };
 use rustc_hash::FxHashMap;
-use std::iter::successors;
 use tt::buffer::{Cursor, TokenBuffer};
 
 use crate::subtree_source::SubtreeTokenSource;
@@ -50,10 +49,8 @@ pub fn ast_to_token_tree(ast: &impl ast::AstNode) -> Option<(tt::Subtree, TokenM
 /// will consume).
 pub fn syntax_node_to_token_tree(node: &SyntaxNode) -> Option<(tt::Subtree, TokenMap)> {
     let global_offset = node.text_range().start();
-    let mut c = Convertor {
-        id_alloc: { TokenIdAlloc { map: TokenMap::default(), global_offset, next_id: 0 } },
-    };
-    let subtree = c.go(node)?;
+    let mut c = Convertor::new(node, global_offset);
+    let subtree = c.go()?;
     Some((subtree, c.id_alloc.map))
 }
 
@@ -152,6 +149,10 @@ impl TokenMap {
             }
         }
     }
+
+    fn remove_delim(&mut self, token_id: tt::TokenId) {
+        self.entries.retain(|(tid, _)| *tid != token_id);
+    }
 }
 
 /// Returns the textual content of a doc comment block as a quoted string
@@ -237,25 +238,26 @@ impl TokenIdAlloc {
         token_id
     }
 
-    fn delim(&mut self, open_abs_range: TextRange, close_abs_range: TextRange) -> tt::TokenId {
-        let open_relative_range = open_abs_range - self.global_offset;
-        let close_relative_range = close_abs_range - self.global_offset;
-        let token_id = tt::TokenId(self.next_id);
-        self.next_id += 1;
-
-        self.map.insert_delim(token_id, open_relative_range, close_relative_range);
-        token_id
-    }
-
     fn open_delim(&mut self, open_abs_range: TextRange) -> tt::TokenId {
         let token_id = tt::TokenId(self.next_id);
         self.next_id += 1;
-        self.map.insert_delim(token_id, open_abs_range, open_abs_range);
+        self.map.insert_delim(
+            token_id,
+            open_abs_range - self.global_offset,
+            open_abs_range - self.global_offset,
+        );
         token_id
     }
 
-    fn close_delim(&mut self, id: tt::TokenId, close_abs_range: TextRange) {
-        self.map.update_close_delim(id, close_abs_range);
+    fn close_delim(&mut self, id: tt::TokenId, close_abs_range: Option<TextRange>) {
+        match close_abs_range {
+            None => {
+                self.map.remove_delim(id);
+            }
+            Some(close) => {
+                self.map.update_close_delim(id, close - self.global_offset);
+            }
+        }
     }
 }
 
@@ -264,10 +266,20 @@ struct RawConvertor<'a> {
     text: &'a str,
     offset: TextUnit,
     id_alloc: TokenIdAlloc,
-    inner: std::slice::Iter<'a, Token>,
+    inner: std::slice::Iter<'a, RawToken>,
 }
 
-impl RawConvertor<'_> {
+trait SrcToken {
+    fn kind(&self) -> SyntaxKind;
+
+    fn to_char(&self) -> Option<char>;
+
+    fn to_text(&self) -> SmolStr;
+}
+
+trait TokenConvertor {
+    type Token: SrcToken;
+
     fn go(&mut self) -> Option<tt::Subtree> {
         let mut subtree = tt::Subtree::default();
         subtree.delimiter = None;
@@ -285,33 +297,22 @@ impl RawConvertor<'_> {
         Some(subtree)
     }
 
-    fn bump(&mut self) -> Option<(Token, TextRange)> {
-        let token = self.inner.next()?;
-        let range = TextRange::offset_len(self.offset, token.len);
-        self.offset += token.len;
-        Some((*token, range))
-    }
-
-    fn peek(&self) -> Option<Token> {
-        self.inner.as_slice().get(0).cloned()
-    }
-
     fn collect_leaf(&mut self, result: &mut Vec<tt::TokenTree>) {
         let (token, range) = match self.bump() {
             None => return,
             Some(it) => it,
         };
 
-        let k: SyntaxKind = token.kind;
+        let k: SyntaxKind = token.kind();
         if k == COMMENT {
-            let node = doc_comment(&self.text[range]);
-            if let Some(tokens) = convert_doc_comment(&node) {
+            if let Some(tokens) = self.convert_doc_comment(&token) {
                 result.extend(tokens);
             }
             return;
         }
 
         result.push(if k.is_punct() {
+            assert_eq!(range.len().to_usize(), 1);
             let delim = match k {
                 T!['('] => Some((tt::DelimiterKind::Parenthesis, T![')'])),
                 T!['{'] => Some((tt::DelimiterKind::Brace, T!['}'])),
@@ -321,40 +322,51 @@ impl RawConvertor<'_> {
 
             if let Some((kind, closed)) = delim {
                 let mut subtree = tt::Subtree::default();
-                let id = self.id_alloc.open_delim(range);
+                let id = self.id_alloc().open_delim(range);
                 subtree.delimiter = Some(tt::Delimiter { kind, id });
 
-                while self.peek().map(|it| it.kind != closed).unwrap_or(false) {
+                while self.peek().map(|it| it.kind() != closed).unwrap_or(false) {
                     self.collect_leaf(&mut subtree.token_trees);
                 }
                 let last_range = match self.bump() {
-                    None => return,
+                    None => {
+                        // For error resilience, we insert an char punct for the opening delim here
+                        self.id_alloc().close_delim(id, None);
+                        let leaf: tt::Leaf = tt::Punct {
+                            id: self.id_alloc().alloc(range),
+                            char: token.to_char().unwrap(),
+                            spacing: tt::Spacing::Alone,
+                        }
+                        .into();
+                        result.push(leaf.into());
+                        result.extend(subtree.token_trees);
+                        return;
+                    }
                     Some(it) => it.1,
                 };
-                self.id_alloc.close_delim(id, last_range);
+                self.id_alloc().close_delim(id, Some(last_range));
                 subtree.into()
             } else {
                 let spacing = match self.peek() {
                     Some(next)
-                        if next.kind.is_trivia()
-                            || next.kind == T!['[']
-                            || next.kind == T!['{']
-                            || next.kind == T!['('] =>
+                        if next.kind().is_trivia()
+                            || next.kind() == T!['[']
+                            || next.kind() == T!['{']
+                            || next.kind() == T!['('] =>
                     {
                         tt::Spacing::Alone
                     }
-                    Some(next) if next.kind.is_punct() => tt::Spacing::Joint,
+                    Some(next) if next.kind().is_punct() => tt::Spacing::Joint,
                     _ => tt::Spacing::Alone,
                 };
-                let char =
-                    self.text[range].chars().next().expect("Token from lexer must be single char");
+                let char = token.to_char().expect("Token from lexer must be single char");
 
-                tt::Leaf::from(tt::Punct { char, spacing, id: self.id_alloc.alloc(range) }).into()
+                tt::Leaf::from(tt::Punct { char, spacing, id: self.id_alloc().alloc(range) }).into()
             }
         } else {
             macro_rules! make_leaf {
                 ($i:ident) => {
-                    tt::$i { id: self.id_alloc.alloc(range), text: self.text[range].into() }.into()
+                    tt::$i { id: self.id_alloc().alloc(range), text: token.to_text() }.into()
                 };
             }
             let leaf: tt::Leaf = match k {
@@ -368,133 +380,168 @@ impl RawConvertor<'_> {
             leaf.into()
         });
     }
+
+    fn convert_doc_comment(&self, token: &Self::Token) -> Option<Vec<tt::TokenTree>>;
+
+    fn bump(&mut self) -> Option<(Self::Token, TextRange)>;
+
+    fn peek(&self) -> Option<Self::Token>;
+
+    fn id_alloc(&mut self) -> &mut TokenIdAlloc;
+}
+
+impl<'a> SrcToken for (RawToken, &'a str) {
+    fn kind(&self) -> SyntaxKind {
+        self.0.kind
+    }
+
+    fn to_char(&self) -> Option<char> {
+        self.1.chars().next()
+    }
+
+    fn to_text(&self) -> SmolStr {
+        self.1.into()
+    }
+}
+
+impl RawConvertor<'_> {}
+
+impl<'a> TokenConvertor for RawConvertor<'a> {
+    type Token = (RawToken, &'a str);
+
+    fn convert_doc_comment(&self, token: &Self::Token) -> Option<Vec<tt::TokenTree>> {
+        convert_doc_comment(&doc_comment(token.1))
+    }
+
+    fn bump(&mut self) -> Option<(Self::Token, TextRange)> {
+        let token = self.inner.next()?;
+        let range = TextRange::offset_len(self.offset, token.len);
+        self.offset += token.len;
+
+        Some(((*token, &self.text[range]), range))
+    }
+
+    fn peek(&self) -> Option<Self::Token> {
+        let token = self.inner.as_slice().get(0).cloned();
+
+        token.map(|it| {
+            let range = TextRange::offset_len(self.offset, it.len);
+            (it, &self.text[range])
+        })
+    }
+
+    fn id_alloc(&mut self) -> &mut TokenIdAlloc {
+        &mut self.id_alloc
+    }
 }
 
-// FIXME: There are some duplicate logic between RawConvertor and Convertor
-// It would be nice to refactor to converting SyntaxNode to ra_parser::Token and thus
-// use RawConvertor directly. But performance-wise it may not be a good idea ?
 struct Convertor {
     id_alloc: TokenIdAlloc,
+    current: Option<SyntaxToken>,
+    range: TextRange,
+    punct_offset: Option<(SyntaxToken, TextUnit)>,
 }
 
 impl Convertor {
-    fn go(&mut self, tt: &SyntaxNode) -> Option<tt::Subtree> {
-        // This tree is empty
-        if tt.first_child_or_token().is_none() {
-            return Some(tt::Subtree { token_trees: vec![], delimiter: None });
+    fn new(node: &SyntaxNode, global_offset: TextUnit) -> Convertor {
+        Convertor {
+            id_alloc: { TokenIdAlloc { map: TokenMap::default(), global_offset, next_id: 0 } },
+            current: node.first_token(),
+            range: node.text_range(),
+            punct_offset: None,
+        }
+    }
+}
+
+enum SynToken {
+    Ordiniary(SyntaxToken),
+    Punch(SyntaxToken, TextUnit),
+}
+
+impl SynToken {
+    fn token(&self) -> &SyntaxToken {
+        match self {
+            SynToken::Ordiniary(it) => it,
+            SynToken::Punch(it, _) => it,
+        }
+    }
+}
+
+impl SrcToken for SynToken {
+    fn kind(&self) -> SyntaxKind {
+        self.token().kind()
+    }
+    fn to_char(&self) -> Option<char> {
+        match self {
+            SynToken::Ordiniary(_) => None,
+            SynToken::Punch(it, i) => it.text().chars().nth(i.to_usize()),
+        }
+    }
+    fn to_text(&self) -> SmolStr {
+        self.token().text().clone()
+    }
+}
+
+impl TokenConvertor for Convertor {
+    type Token = SynToken;
+    fn convert_doc_comment(&self, token: &Self::Token) -> Option<Vec<tt::TokenTree>> {
+        convert_doc_comment(token.token())
+    }
+
+    fn bump(&mut self) -> Option<(Self::Token, TextRange)> {
+        if let Some((punct, offset)) = self.punct_offset.clone() {
+            if offset.to_usize() + 1 < punct.text().len() {
+                let offset = offset + TextUnit::from_usize(1);
+                let range = punct.text_range();
+                self.punct_offset = Some((punct.clone(), offset));
+                let range = TextRange::offset_len(range.start() + offset, TextUnit::from_usize(1));
+                return Some((SynToken::Punch(punct, offset), range));
+            }
         }
 
-        let first_child = tt.first_child_or_token()?;
-        let last_child = tt.last_child_or_token()?;
-
-        // ignore trivial first_child and last_child
-        let first_child = successors(Some(first_child), |it| {
-            if it.kind().is_trivia() {
-                it.next_sibling_or_token()
-            } else {
-                None
-            }
-        })
-        .last()
-        .unwrap();
-        if first_child.kind().is_trivia() {
-            return Some(tt::Subtree { token_trees: vec![], delimiter: None });
+        let curr = self.current.clone()?;
+        if !curr.text_range().is_subrange(&self.range) {
+            return None;
         }
+        self.current = curr.next_token();
 
-        let last_child = successors(Some(last_child), |it| {
-            if it.kind().is_trivia() {
-                it.prev_sibling_or_token()
-            } else {
-                None
-            }
-        })
-        .last()
-        .unwrap();
-
-        let (delimiter_kind, skip_first) = match (first_child.kind(), last_child.kind()) {
-            (T!['('], T![')']) => (Some(tt::DelimiterKind::Parenthesis), true),
-            (T!['{'], T!['}']) => (Some(tt::DelimiterKind::Brace), true),
-            (T!['['], T![']']) => (Some(tt::DelimiterKind::Bracket), true),
-            _ => (None, false),
+        let token = if curr.kind().is_punct() {
+            let range = curr.text_range();
+            let range = TextRange::offset_len(range.start(), TextUnit::from_usize(1));
+            self.punct_offset = Some((curr.clone(), TextUnit::from_usize(0)));
+            (SynToken::Punch(curr, TextUnit::from_usize(0)), range)
+        } else {
+            self.punct_offset = None;
+            let range = curr.text_range();
+            (SynToken::Ordiniary(curr), range)
         };
-        let delimiter = delimiter_kind.map(|kind| tt::Delimiter {
-            kind,
-            id: self.id_alloc.delim(first_child.text_range(), last_child.text_range()),
-        });
 
-        let mut token_trees = Vec::new();
-        let mut child_iter = tt.children_with_tokens().skip(skip_first as usize).peekable();
+        Some(token)
+    }
 
-        while let Some(child) = child_iter.next() {
-            if skip_first && (child == first_child || child == last_child) {
-                continue;
+    fn peek(&self) -> Option<Self::Token> {
+        if let Some((punct, mut offset)) = self.punct_offset.clone() {
+            offset = offset + TextUnit::from_usize(1);
+            if offset.to_usize() < punct.text().len() {
+                return Some(SynToken::Punch(punct, offset));
             }
-
-            match child {
-                NodeOrToken::Token(token) => {
-                    if let Some(doc_tokens) = convert_doc_comment(&token) {
-                        token_trees.extend(doc_tokens);
-                    } else if token.kind().is_trivia() {
-                        continue;
-                    } else if token.kind().is_punct() {
-                        // we need to pull apart joined punctuation tokens
-                        let last_spacing = match child_iter.peek() {
-                            Some(NodeOrToken::Token(token)) => {
-                                if token.kind().is_punct() {
-                                    tt::Spacing::Joint
-                                } else {
-                                    tt::Spacing::Alone
-                                }
-                            }
-                            _ => tt::Spacing::Alone,
-                        };
-                        let spacing_iter = std::iter::repeat(tt::Spacing::Joint)
-                            .take(token.text().len() - 1)
-                            .chain(std::iter::once(last_spacing));
-                        for (char, spacing) in token.text().chars().zip(spacing_iter) {
-                            token_trees.push(
-                                tt::Leaf::from(tt::Punct {
-                                    char,
-                                    spacing,
-                                    id: self.id_alloc.alloc(token.text_range()),
-                                })
-                                .into(),
-                            );
-                        }
-                    } else {
-                        macro_rules! make_leaf {
-                            ($i:ident) => {
-                                tt::$i {
-                                    id: self.id_alloc.alloc(token.text_range()),
-                                    text: token.text().clone(),
-                                }
-                                .into()
-                            };
-                        }
-
-                        let child: tt::Leaf = match token.kind() {
-                            T![true] | T![false] => make_leaf!(Literal),
-                            IDENT | LIFETIME => make_leaf!(Ident),
-                            k if k.is_keyword() => make_leaf!(Ident),
-                            k if k.is_literal() => make_leaf!(Literal),
-                            _ => return None,
-                        };
-                        token_trees.push(child.into());
-                    }
-                }
-                NodeOrToken::Node(node) => {
-                    let child_subtree = self.go(&node)?;
-                    if child_subtree.delimiter.is_none() && node.kind() != SyntaxKind::TOKEN_TREE {
-                        token_trees.extend(child_subtree.token_trees);
-                    } else {
-                        token_trees.push(child_subtree.into());
-                    }
-                }
-            };
         }
 
-        let res = tt::Subtree { delimiter, token_trees };
-        Some(res)
+        let curr = self.current.clone()?;
+        if !curr.text_range().is_subrange(&self.range) {
+            return None;
+        }
+
+        let token = if curr.kind().is_punct() {
+            SynToken::Punch(curr, TextUnit::from_usize(0))
+        } else {
+            SynToken::Ordiniary(curr)
+        };
+        Some(token)
+    }
+
+    fn id_alloc(&mut self) -> &mut TokenIdAlloc {
+        &mut self.id_alloc
     }
 }
 

crates/ra_mbe/src/tests.rs

@@ -427,22 +427,28 @@ MACRO_ITEMS@[0; 40)
     );
 }
 
+fn to_subtree(tt: &tt::TokenTree) -> &tt::Subtree {
+    if let tt::TokenTree::Subtree(subtree) = tt {
+        return &subtree;
+    }
+    unreachable!("It is not a subtree");
+}
+fn to_literal(tt: &tt::TokenTree) -> &tt::Literal {
+    if let tt::TokenTree::Leaf(tt::Leaf::Literal(lit)) = tt {
+        return lit;
+    }
+    unreachable!("It is not a literal");
+}
+
+fn to_punct(tt: &tt::TokenTree) -> &tt::Punct {
+    if let tt::TokenTree::Leaf(tt::Leaf::Punct(lit)) = tt {
+        return lit;
+    }
+    unreachable!("It is not a Punct");
+}
+
 #[test]
 fn test_expand_literals_to_token_tree() {
-    fn to_subtree(tt: &tt::TokenTree) -> &tt::Subtree {
-        if let tt::TokenTree::Subtree(subtree) = tt {
-            return &subtree;
-        }
-        unreachable!("It is not a subtree");
-    }
-
-    fn to_literal(tt: &tt::TokenTree) -> &tt::Literal {
-        if let tt::TokenTree::Leaf(tt::Leaf::Literal(lit)) = tt {
-            return lit;
-        }
-        unreachable!("It is not a literal");
-    }
-
     let expansion = parse_macro(
         r#"
             macro_rules! literals {
@@ -470,6 +476,22 @@ fn test_expand_literals_to_token_tree() {
     assert_eq!(to_literal(&stm_tokens[15 + 3]).text, "\"rust1\"");
 }
 
+#[test]
+fn test_attr_to_token_tree() {
+    let expansion = parse_to_token_tree_by_syntax(
+        r#"
+            #[derive(Copy)]
+            struct Foo;
+            "#,
+    );
+
+    assert_eq!(to_punct(&expansion.token_trees[0]).char, '#');
+    assert_eq!(
+        to_subtree(&expansion.token_trees[1]).delimiter_kind(),
+        Some(tt::DelimiterKind::Bracket)
+    );
+}
+
 #[test]
 fn test_two_idents() {
     parse_macro(
@@ -1427,8 +1449,8 @@ impl MacroFixture {
         let macro_invocation =
             source_file.syntax().descendants().find_map(ast::MacroCall::cast).unwrap();
 
-        let (invocation_tt, _) =
-            ast_to_token_tree(&macro_invocation.token_tree().unwrap()).unwrap();
+        let (invocation_tt, _) = ast_to_token_tree(&macro_invocation.token_tree().unwrap())
+            .ok_or_else(|| ExpandError::ConversionError)?;
 
         self.rules.expand(&invocation_tt).result()
     }
@@ -1517,6 +1539,16 @@ pub(crate) fn parse_macro(ra_fixture: &str) -> MacroFixture {
     MacroFixture { rules }
 }
 
+pub(crate) fn parse_to_token_tree_by_syntax(ra_fixture: &str) -> tt::Subtree {
+    let source_file = ast::SourceFile::parse(ra_fixture).ok().unwrap();
+    let tt = syntax_node_to_token_tree(source_file.syntax()).unwrap().0;
+
+    let parsed = parse_to_token_tree(ra_fixture).unwrap().0;
+    assert_eq!(tt, parsed);
+
+    parsed
+}
+
 fn debug_dump_ignore_spaces(node: &ra_syntax::SyntaxNode) -> String {
     let mut level = 0;
     let mut buf = String::new();
@@ -1662,5 +1694,5 @@ fn test_expand_bad_literal() {
         macro_rules! foo { ($i:literal) => {}; }
     "#,
     )
-    .assert_expand_err(r#"foo!(&k");"#, &ExpandError::BindingError("".to_string()));
+    .assert_expand_err(r#"foo!(&k");"#, &ExpandError::BindingError("".into()));
 }