rust-analyzer/crates/mbe/src/syntax_bridge.rs

757 lines
25 KiB
Rust
Raw Normal View History

2021-05-22 14:20:22 +00:00
//! Conversions between [`SyntaxNode`] and [`tt::TokenTree`].
use rustc_hash::{FxHashMap, FxHashSet};
2020-08-12 16:26:51 +00:00
use syntax::{
2020-03-08 16:13:04 +00:00
ast::{self, make::tokens::doc_comment},
AstToken, Parse, PreorderWithTokens, SmolStr, SyntaxElement, SyntaxKind,
2020-03-08 16:13:04 +00:00
SyntaxKind::*,
SyntaxNode, SyntaxToken, SyntaxTreeBuilder, TextRange, TextSize, WalkEvent, T,
2019-02-11 18:12:06 +00:00
};
use tt::buffer::{Cursor, TokenBuffer};
2019-04-07 13:42:53 +00:00
use crate::{
2021-12-12 16:06:40 +00:00
to_parser_tokens::to_parser_tokens, tt_iter::TtIter, ExpandError, ParserEntryPoint, TokenMap,
};
2019-11-03 14:46:12 +00:00
/// Convert the syntax node to a `TokenTree` (what macro
/// will consume).
pub fn syntax_node_to_token_tree(node: &SyntaxNode) -> (tt::Subtree, TokenMap) {
syntax_node_to_token_tree_censored(node, &Default::default())
}
/// Convert the syntax node to a `TokenTree` (what macro will consume)
/// with the censored range excluded.
pub fn syntax_node_to_token_tree_censored(
node: &SyntaxNode,
censor: &FxHashSet<SyntaxNode>,
) -> (tt::Subtree, TokenMap) {
2019-11-18 13:08:41 +00:00
let global_offset = node.text_range().start();
let mut c = Convertor::new(node, global_offset, censor);
let subtree = convert_tokens(&mut c);
2021-05-24 16:43:42 +00:00
c.id_alloc.map.shrink_to_fit();
(subtree, c.id_alloc.map)
}
// The following items are what `rustc` macro can be parsed into :
// link: https://github.com/rust-lang/rust/blob/9ebf47851a357faa4cd97f4b1dc7835f6376e639/src/libsyntax/ext/expand.rs#L141
2019-04-18 19:49:56 +00:00
// * Expr(P<ast::Expr>) -> token_tree_to_expr
// * Pat(P<ast::Pat>) -> token_tree_to_pat
// * Ty(P<ast::Ty>) -> token_tree_to_ty
// * Stmts(SmallVec<[ast::Stmt; 1]>) -> token_tree_to_stmts
// * Items(SmallVec<[P<ast::Item>; 1]>) -> token_tree_to_items
//
// * TraitItems(SmallVec<[ast::TraitItem; 1]>)
2020-05-05 15:56:10 +00:00
// * AssocItems(SmallVec<[ast::AssocItem; 1]>)
// * ForeignItems(SmallVec<[ast::ForeignItem; 1]>
2019-05-25 13:55:46 +00:00
pub fn token_tree_to_syntax_node(
2019-09-02 15:51:03 +00:00
tt: &tt::Subtree,
entry_point: ParserEntryPoint,
2019-11-18 13:08:41 +00:00
) -> Result<(Parse<SyntaxNode>, TokenMap), ExpandError> {
2021-01-04 16:22:42 +00:00
let buffer = match tt {
tt::Subtree { delimiter: None, token_trees } => {
TokenBuffer::from_tokens(token_trees.as_slice())
}
2021-01-04 16:22:42 +00:00
_ => TokenBuffer::from_subtree(tt),
};
2021-12-12 16:06:40 +00:00
let parser_tokens = to_parser_tokens(&buffer);
let tree_traversal = parser::parse(&parser_tokens, entry_point);
2019-11-03 19:12:19 +00:00
let mut tree_sink = TtTreeSink::new(buffer.begin());
for event in tree_traversal.iter() {
match event {
parser::TraversalStep::Token { kind, n_raw_tokens } => {
tree_sink.token(kind, n_raw_tokens)
}
parser::TraversalStep::EnterNode { kind } => tree_sink.start_node(kind),
parser::TraversalStep::LeaveNode => tree_sink.finish_node(),
parser::TraversalStep::Error { msg } => tree_sink.error(msg.to_string()),
}
}
2019-09-02 15:51:03 +00:00
if tree_sink.roots.len() != 1 {
return Err(ExpandError::ConversionError);
}
//FIXME: would be cool to report errors
2019-11-03 19:12:19 +00:00
let (parse, range_map) = tree_sink.finish();
2019-11-03 14:46:12 +00:00
Ok((parse, range_map))
2019-09-02 15:51:03 +00:00
}
2020-03-08 16:13:04 +00:00
/// Convert a string to a `TokenTree`
pub fn parse_to_token_tree(text: &str) -> Option<(tt::Subtree, TokenMap)> {
let lexed = parser::LexedStr::new(text);
if lexed.errors().next().is_some() {
2020-03-08 16:13:04 +00:00
return None;
}
let mut conv = RawConvertor {
lexed: lexed,
pos: 0,
2020-03-08 16:13:04 +00:00
id_alloc: TokenIdAlloc {
map: Default::default(),
2020-04-24 21:40:41 +00:00
global_offset: TextSize::default(),
2020-03-08 16:13:04 +00:00
next_id: 0,
},
};
let subtree = convert_tokens(&mut conv);
2020-03-08 16:13:04 +00:00
Some((subtree, conv.id_alloc.map))
}
/// Split token tree with separate expr: $($e:expr)SEP*
pub fn parse_exprs_with_sep(tt: &tt::Subtree, sep: char) -> Vec<tt::Subtree> {
if tt.token_trees.is_empty() {
return Vec::new();
}
let mut iter = TtIter::new(tt);
let mut res = Vec::new();
while iter.peek_n(0).is_some() {
let expanded = iter.expect_fragment(ParserEntryPoint::Expr);
res.push(match expanded.value {
None => break,
Some(tt @ tt::TokenTree::Leaf(_)) => {
tt::Subtree { delimiter: None, token_trees: vec![tt] }
}
Some(tt::TokenTree::Subtree(tt)) => tt,
});
let mut fork = iter.clone();
if fork.expect_char(sep).is_err() {
break;
}
iter = fork;
}
if iter.peek_n(0).is_some() {
res.push(tt::Subtree { delimiter: None, token_trees: iter.into_iter().cloned().collect() });
}
res
}
fn convert_tokens<C: TokenConvertor>(conv: &mut C) -> tt::Subtree {
struct StackEntry {
subtree: tt::Subtree,
idx: usize,
open_range: TextRange,
}
let entry = StackEntry {
subtree: tt::Subtree { delimiter: None, ..Default::default() },
// never used (delimiter is `None`)
idx: !0,
open_range: TextRange::empty(TextSize::of('.')),
};
let mut stack = vec![entry];
loop {
let entry = stack.last_mut().unwrap();
let result = &mut entry.subtree.token_trees;
let (token, range) = match conv.bump() {
None => break,
Some(it) => it,
};
let k: SyntaxKind = token.kind(&conv);
if k == COMMENT {
if let Some(tokens) = conv.convert_doc_comment(&token) {
// FIXME: There has to be a better way to do this
// Add the comments token id to the converted doc string
let id = conv.id_alloc().alloc(range);
result.extend(tokens.into_iter().map(|mut tt| {
if let tt::TokenTree::Subtree(sub) = &mut tt {
if let tt::TokenTree::Leaf(tt::Leaf::Literal(lit)) = &mut sub.token_trees[2]
{
lit.id = id
}
}
tt
}));
}
continue;
}
result.push(if k.is_punct() && k != UNDERSCORE {
assert_eq!(range.len(), TextSize::of('.'));
if let Some(delim) = entry.subtree.delimiter {
let expected = match delim.kind {
tt::DelimiterKind::Parenthesis => T![')'],
tt::DelimiterKind::Brace => T!['}'],
tt::DelimiterKind::Bracket => T![']'],
};
if k == expected {
let entry = stack.pop().unwrap();
conv.id_alloc().close_delim(entry.idx, Some(range));
stack.last_mut().unwrap().subtree.token_trees.push(entry.subtree.into());
continue;
}
}
let delim = match k {
T!['('] => Some(tt::DelimiterKind::Parenthesis),
T!['{'] => Some(tt::DelimiterKind::Brace),
T!['['] => Some(tt::DelimiterKind::Bracket),
_ => None,
};
if let Some(kind) = delim {
let mut subtree = tt::Subtree::default();
let (id, idx) = conv.id_alloc().open_delim(range);
subtree.delimiter = Some(tt::Delimiter { id, kind });
stack.push(StackEntry { subtree, idx, open_range: range });
continue;
} else {
let spacing = match conv.peek() {
Some(next)
if next.kind(&conv).is_trivia()
|| next.kind(&conv) == T!['[']
|| next.kind(&conv) == T!['{']
|| next.kind(&conv) == T!['('] =>
{
tt::Spacing::Alone
}
Some(next) if next.kind(&conv).is_punct() && next.kind(&conv) != UNDERSCORE => {
tt::Spacing::Joint
}
_ => tt::Spacing::Alone,
};
let char = match token.to_char(&conv) {
Some(c) => c,
None => {
panic!("Token from lexer must be single char: token = {:#?}", token);
}
};
tt::Leaf::from(tt::Punct { char, spacing, id: conv.id_alloc().alloc(range) }).into()
}
} else {
macro_rules! make_leaf {
($i:ident) => {
tt::$i { id: conv.id_alloc().alloc(range), text: token.to_text(conv) }.into()
};
}
let leaf: tt::Leaf = match k {
T![true] | T![false] => make_leaf!(Ident),
IDENT => make_leaf!(Ident),
UNDERSCORE => make_leaf!(Ident),
k if k.is_keyword() => make_leaf!(Ident),
k if k.is_literal() => make_leaf!(Literal),
LIFETIME_IDENT => {
let char_unit = TextSize::of('\'');
let r = TextRange::at(range.start(), char_unit);
let apostrophe = tt::Leaf::from(tt::Punct {
char: '\'',
spacing: tt::Spacing::Joint,
id: conv.id_alloc().alloc(r),
});
result.push(apostrophe.into());
let r = TextRange::at(range.start() + char_unit, range.len() - char_unit);
let ident = tt::Leaf::from(tt::Ident {
text: SmolStr::new(&token.to_text(conv)[1..]),
id: conv.id_alloc().alloc(r),
});
result.push(ident.into());
continue;
}
_ => continue,
};
leaf.into()
});
}
// If we get here, we've consumed all input tokens.
// We might have more than one subtree in the stack, if the delimiters are improperly balanced.
// Merge them so we're left with one.
while stack.len() > 1 {
let entry = stack.pop().unwrap();
let parent = stack.last_mut().unwrap();
conv.id_alloc().close_delim(entry.idx, None);
let leaf: tt::Leaf = tt::Punct {
id: conv.id_alloc().alloc(entry.open_range),
char: match entry.subtree.delimiter.unwrap().kind {
tt::DelimiterKind::Parenthesis => '(',
tt::DelimiterKind::Brace => '{',
tt::DelimiterKind::Bracket => '[',
},
spacing: tt::Spacing::Alone,
}
.into();
parent.subtree.token_trees.push(leaf.into());
parent.subtree.token_trees.extend(entry.subtree.token_trees);
}
let subtree = stack.pop().unwrap().subtree;
if subtree.token_trees.len() == 1 {
if let tt::TokenTree::Subtree(first) = &subtree.token_trees[0] {
return first.clone();
}
}
subtree
}
2019-05-04 07:00:16 +00:00
/// Returns the textual content of a doc comment block as a quoted string
/// That is, strips leading `///` (or `/**`, etc)
/// and strips the ending `*/`
/// And then quote the string, which is needed to convert to `tt::Literal`
fn doc_comment_text(comment: &ast::Comment) -> SmolStr {
let prefix_len = comment.prefix().len();
let mut text = &comment.text()[prefix_len..];
// Remove ending "*/"
if comment.kind().shape == ast::CommentShape::Block {
text = &text[0..text.len() - 2];
}
// Quote the string
// Note that `tt::Literal` expect an escaped string
let text = format!("\"{}\"", text.escape_debug());
2019-05-04 07:00:16 +00:00
text.into()
}
2020-08-12 16:26:51 +00:00
fn convert_doc_comment(token: &syntax::SyntaxToken) -> Option<Vec<tt::TokenTree>> {
2021-10-10 08:39:08 +00:00
cov_mark::hit!(test_meta_doc_comments);
2019-07-18 17:09:50 +00:00
let comment = ast::Comment::cast(token.clone())?;
2019-05-04 07:00:16 +00:00
let doc = comment.kind().doc?;
// Make `doc="\" Comments\""
let meta_tkns = vec![mk_ident("doc"), mk_punct('='), mk_doc_literal(&comment)];
2019-05-04 07:00:16 +00:00
// Make `#![]`
let mut token_trees = vec![mk_punct('#')];
2019-05-04 07:00:16 +00:00
if let ast::CommentPlacement::Inner = doc {
token_trees.push(mk_punct('!'));
}
2019-06-03 14:21:08 +00:00
token_trees.push(tt::TokenTree::from(tt::Subtree {
2019-12-12 17:41:44 +00:00
delimiter: Some(tt::Delimiter {
kind: tt::DelimiterKind::Bracket,
id: tt::TokenId::unspecified(),
}),
2019-06-03 14:21:08 +00:00
token_trees: meta_tkns,
}));
2019-05-04 07:00:16 +00:00
return Some(token_trees);
// Helper functions
fn mk_ident(s: &str) -> tt::TokenTree {
tt::TokenTree::from(tt::Leaf::from(tt::Ident {
text: s.into(),
id: tt::TokenId::unspecified(),
}))
}
fn mk_punct(c: char) -> tt::TokenTree {
2019-12-12 13:47:54 +00:00
tt::TokenTree::from(tt::Leaf::from(tt::Punct {
char: c,
spacing: tt::Spacing::Alone,
id: tt::TokenId::unspecified(),
}))
2019-05-04 07:00:16 +00:00
}
fn mk_doc_literal(comment: &ast::Comment) -> tt::TokenTree {
2019-12-12 13:47:54 +00:00
let lit = tt::Literal { text: doc_comment_text(comment), id: tt::TokenId::unspecified() };
2019-05-04 07:00:16 +00:00
tt::TokenTree::from(tt::Leaf::from(lit))
}
}
2020-03-08 16:13:04 +00:00
struct TokenIdAlloc {
2019-11-18 13:08:41 +00:00
map: TokenMap,
2020-04-24 21:40:41 +00:00
global_offset: TextSize,
2019-11-18 13:08:41 +00:00
next_id: u32,
}
2019-04-25 15:12:57 +00:00
2020-03-08 16:13:04 +00:00
impl TokenIdAlloc {
fn alloc(&mut self, absolute_range: TextRange) -> tt::TokenId {
let relative_range = absolute_range - self.global_offset;
let token_id = tt::TokenId(self.next_id);
self.next_id += 1;
self.map.insert(token_id, relative_range);
token_id
}
fn open_delim(&mut self, open_abs_range: TextRange) -> (tt::TokenId, usize) {
2020-03-08 16:13:04 +00:00
let token_id = tt::TokenId(self.next_id);
self.next_id += 1;
let idx = self.map.insert_delim(
2020-03-20 20:22:21 +00:00
token_id,
open_abs_range - self.global_offset,
open_abs_range - self.global_offset,
);
(token_id, idx)
2020-03-08 16:13:04 +00:00
}
fn close_delim(&mut self, idx: usize, close_abs_range: Option<TextRange>) {
2020-03-20 21:20:28 +00:00
match close_abs_range {
None => {
self.map.remove_delim(idx);
2020-03-20 21:20:28 +00:00
}
Some(close) => {
self.map.update_close_delim(idx, close - self.global_offset);
2020-03-20 21:20:28 +00:00
}
}
2020-03-08 16:13:04 +00:00
}
}
/// A Raw Token (straightly from lexer) convertor
struct RawConvertor<'a> {
lexed: parser::LexedStr<'a>,
pos: usize,
2020-03-08 16:13:04 +00:00
id_alloc: TokenIdAlloc,
}
trait SrcToken<Ctx>: std::fmt::Debug {
fn kind(&self, ctx: &Ctx) -> SyntaxKind;
2020-03-20 19:04:11 +00:00
fn to_char(&self, ctx: &Ctx) -> Option<char>;
2020-03-20 19:04:11 +00:00
fn to_text(&self, ctx: &Ctx) -> SmolStr;
2020-03-19 12:40:25 +00:00
}
trait TokenConvertor: Sized {
type Token: SrcToken<Self>;
2020-03-19 12:40:25 +00:00
2020-03-20 19:04:11 +00:00
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;
2020-03-19 12:40:25 +00:00
}
impl<'a> SrcToken<RawConvertor<'a>> for usize {
fn kind(&self, ctx: &RawConvertor<'a>) -> SyntaxKind {
ctx.lexed.kind(*self)
2020-03-08 16:13:04 +00:00
}
fn to_char(&self, ctx: &RawConvertor<'a>) -> Option<char> {
ctx.lexed.text(*self).chars().next()
2020-03-08 16:13:04 +00:00
}
fn to_text(&self, ctx: &RawConvertor<'_>) -> SmolStr {
ctx.lexed.text(*self).into()
2020-03-08 16:13:04 +00:00
}
2020-03-20 19:04:11 +00:00
}
2020-03-08 16:13:04 +00:00
2020-03-20 19:04:11 +00:00
impl<'a> TokenConvertor for RawConvertor<'a> {
type Token = usize;
2020-03-08 16:13:04 +00:00
fn convert_doc_comment(&self, token: &usize) -> Option<Vec<tt::TokenTree>> {
let text = self.lexed.text(*token);
convert_doc_comment(&doc_comment(text))
2020-03-20 19:04:11 +00:00
}
2020-03-08 16:13:04 +00:00
2020-03-20 19:04:11 +00:00
fn bump(&mut self) -> Option<(Self::Token, TextRange)> {
if self.pos == self.lexed.len() {
return None;
}
let token = self.pos;
self.pos += 1;
let range = self.lexed.text_range(token);
let range = TextRange::new(range.start.try_into().unwrap(), range.end.try_into().unwrap());
2020-03-08 16:13:04 +00:00
Some((token, range))
2020-03-20 19:04:11 +00:00
}
2020-03-08 16:13:04 +00:00
2020-03-20 19:04:11 +00:00
fn peek(&self) -> Option<Self::Token> {
if self.pos == self.lexed.len() {
return None;
}
Some(self.pos)
2020-03-20 19:04:11 +00:00
}
fn id_alloc(&mut self) -> &mut TokenIdAlloc {
&mut self.id_alloc
2020-03-08 16:13:04 +00:00
}
}
struct Convertor<'c> {
2020-03-08 16:13:04 +00:00
id_alloc: TokenIdAlloc,
2020-03-20 19:04:11 +00:00
current: Option<SyntaxToken>,
preorder: PreorderWithTokens,
censor: &'c FxHashSet<SyntaxNode>,
2020-03-20 19:04:11 +00:00
range: TextRange,
2020-04-24 21:40:41 +00:00
punct_offset: Option<(SyntaxToken, TextSize)>,
2020-03-08 16:13:04 +00:00
}
impl<'c> Convertor<'c> {
fn new(
node: &SyntaxNode,
global_offset: TextSize,
censor: &'c FxHashSet<SyntaxNode>,
) -> Convertor<'c> {
let range = node.text_range();
let mut preorder = node.preorder_with_tokens();
let first = Self::next_token(&mut preorder, censor);
2020-03-20 19:04:11 +00:00
Convertor {
id_alloc: { TokenIdAlloc { map: TokenMap::default(), global_offset, next_id: 0 } },
current: first,
preorder,
range,
censor,
2020-03-20 19:04:11 +00:00
punct_offset: None,
2019-11-18 13:08:41 +00:00
}
2020-03-20 19:04:11 +00:00
}
fn next_token(
preorder: &mut PreorderWithTokens,
censor: &FxHashSet<SyntaxNode>,
) -> Option<SyntaxToken> {
while let Some(ev) = preorder.next() {
let ele = match ev {
WalkEvent::Enter(ele) => ele,
_ => continue,
};
match ele {
SyntaxElement::Token(t) => return Some(t),
SyntaxElement::Node(node) if censor.contains(&node) => preorder.skip_subtree(),
SyntaxElement::Node(_) => (),
}
}
None
}
2020-03-20 19:04:11 +00:00
}
2020-04-18 11:28:07 +00:00
#[derive(Debug)]
2020-03-20 19:04:11 +00:00
enum SynToken {
2021-01-08 14:41:46 +00:00
Ordinary(SyntaxToken),
2020-04-24 21:40:41 +00:00
Punch(SyntaxToken, TextSize),
2020-03-20 19:04:11 +00:00
}
2019-11-21 15:56:01 +00:00
2020-03-20 19:04:11 +00:00
impl SynToken {
fn token(&self) -> &SyntaxToken {
match self {
2021-01-08 14:41:46 +00:00
SynToken::Ordinary(it) => it,
2020-03-20 19:04:11 +00:00
SynToken::Punch(it, _) => it,
2019-11-21 15:56:01 +00:00
}
2020-03-20 19:04:11 +00:00
}
}
2019-11-21 15:56:01 +00:00
impl<'a> SrcToken<Convertor<'a>> for SynToken {
fn kind(&self, _ctx: &Convertor<'a>) -> SyntaxKind {
2020-03-20 19:04:11 +00:00
self.token().kind()
}
fn to_char(&self, _ctx: &Convertor<'a>) -> Option<char> {
2020-03-20 19:04:11 +00:00
match self {
2021-01-08 14:41:46 +00:00
SynToken::Ordinary(_) => None,
2020-04-24 21:40:41 +00:00
SynToken::Punch(it, i) => it.text().chars().nth((*i).into()),
2020-03-20 19:04:11 +00:00
}
}
fn to_text(&self, _ctx: &Convertor<'a>) -> SmolStr {
2021-01-19 22:56:11 +00:00
self.token().text().into()
2020-03-20 19:04:11 +00:00
}
}
impl TokenConvertor for Convertor<'_> {
2020-03-20 19:04:11 +00:00
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() {
2020-04-24 21:40:41 +00:00
if usize::from(offset) + 1 < punct.text().len() {
2020-04-24 22:57:47 +00:00
let offset = offset + TextSize::of('.');
2020-03-20 19:04:11 +00:00
let range = punct.text_range();
2020-03-20 20:22:21 +00:00
self.punct_offset = Some((punct.clone(), offset));
2020-04-24 21:40:41 +00:00
let range = TextRange::at(range.start() + offset, TextSize::of('.'));
2020-03-20 20:22:21 +00:00
return Some((SynToken::Punch(punct, offset), range));
2019-11-21 15:56:01 +00:00
}
2020-03-20 19:04:11 +00:00
}
2020-03-20 20:22:21 +00:00
let curr = self.current.clone()?;
2020-04-24 21:40:41 +00:00
if !&self.range.contains_range(curr.text_range()) {
2020-03-20 20:22:21 +00:00
return None;
}
self.current = Self::next_token(&mut self.preorder, self.censor);
2020-03-20 19:04:11 +00:00
let token = if curr.kind().is_punct() {
let range = curr.text_range();
2020-04-24 22:57:47 +00:00
let range = TextRange::at(range.start(), TextSize::of('.'));
self.punct_offset = Some((curr.clone(), 0.into()));
(SynToken::Punch(curr, 0.into()), range)
2020-03-20 19:04:11 +00:00
} else {
self.punct_offset = None;
let range = curr.text_range();
2021-01-08 14:41:46 +00:00
(SynToken::Ordinary(curr), range)
2019-11-18 13:08:41 +00:00
};
2019-04-24 15:01:32 +00:00
2020-03-20 19:04:11 +00:00
Some(token)
}
2019-05-04 07:00:16 +00:00
2020-03-20 19:04:11 +00:00
fn peek(&self) -> Option<Self::Token> {
if let Some((punct, mut offset)) = self.punct_offset.clone() {
offset += TextSize::of('.');
2020-04-24 21:40:41 +00:00
if usize::from(offset) < punct.text().len() {
2020-03-20 19:04:11 +00:00
return Some(SynToken::Punch(punct, offset));
}
2019-11-18 13:08:41 +00:00
}
2020-03-20 20:22:21 +00:00
let curr = self.current.clone()?;
2020-04-24 21:40:41 +00:00
if !self.range.contains_range(curr.text_range()) {
2020-03-20 20:22:21 +00:00
return None;
}
2020-03-20 19:04:11 +00:00
let token = if curr.kind().is_punct() {
2020-04-24 22:57:47 +00:00
SynToken::Punch(curr, 0.into())
2020-03-20 19:04:11 +00:00
} else {
2021-01-08 14:41:46 +00:00
SynToken::Ordinary(curr)
2020-03-20 19:04:11 +00:00
};
Some(token)
}
fn id_alloc(&mut self) -> &mut TokenIdAlloc {
&mut self.id_alloc
2019-01-31 18:29:04 +00:00
}
}
2019-02-23 11:18:32 +00:00
2019-05-27 14:56:21 +00:00
struct TtTreeSink<'a> {
2019-02-23 14:21:56 +00:00
buf: String,
2019-05-27 14:56:21 +00:00
cursor: Cursor<'a>,
2020-04-24 21:40:41 +00:00
open_delims: FxHashMap<tt::TokenId, TextSize>,
text_pos: TextSize,
2019-02-23 14:21:56 +00:00
inner: SyntaxTreeBuilder,
2019-11-18 13:08:41 +00:00
token_map: TokenMap,
// Number of roots
// Use for detect ill-form tree which is not single root
roots: smallvec::SmallVec<[usize; 1]>,
2019-02-23 14:21:56 +00:00
}
2019-05-27 14:56:21 +00:00
impl<'a> TtTreeSink<'a> {
2019-11-03 19:12:19 +00:00
fn new(cursor: Cursor<'a>) -> Self {
2019-02-23 14:21:56 +00:00
TtTreeSink {
buf: String::new(),
2019-05-27 14:56:21 +00:00
cursor,
2019-12-12 17:41:44 +00:00
open_delims: FxHashMap::default(),
2019-02-23 14:21:56 +00:00
text_pos: 0.into(),
inner: SyntaxTreeBuilder::default(),
roots: smallvec::SmallVec::new(),
2019-11-18 13:08:41 +00:00
token_map: TokenMap::default(),
2019-02-23 14:21:56 +00:00
}
}
2019-11-03 19:12:19 +00:00
fn finish(mut self) -> (Parse<SyntaxNode>, TokenMap) {
2021-05-24 16:43:42 +00:00
self.token_map.shrink_to_fit();
2019-11-18 13:08:41 +00:00
(self.inner.finish(), self.token_map)
2019-11-03 19:12:19 +00:00
}
2019-02-23 14:21:56 +00:00
}
internal: replace L_DOLLAR/R_DOLLAR with parenthesis hack The general problem we are dealing with here is this: ``` macro_rules! thrice { ($e:expr) => { $e * 3} } fn main() { let x = thrice!(1 + 2); } ``` we really want this to print 9 rather than 7. The way rustc solves this is rather ad-hoc. In rustc, token trees are allowed to include whole AST fragments, so 1+2 is passed through macro expansion as a single unit. This is a significant violation of token tree model. In rust-analyzer, we intended to handle this in a more elegant way, using token trees with "invisible" delimiters. The idea was is that we introduce a new kind of parenthesis, "left $"/"right $", and let the parser intelligently handle this. The idea was inspired by the relevant comment in the proc_macro crate: https://doc.rust-lang.org/stable/proc_macro/enum.Delimiter.html#variant.None > An implicit delimiter, that may, for example, appear around tokens > coming from a “macro variable” $var. It is important to preserve > operator priorities in cases like $var * 3 where $var is 1 + 2. > Implicit delimiters might not survive roundtrip of a token stream > through a string. Now that we are older and wiser, we conclude that the idea doesn't work. _First_, the comment in the proc-macro crate is wishful thinking. Rustc currently completely ignores none delimiters. It solves the (1 + 2) * 3 problem by having magical token trees which can't be duplicated: * https://rust-lang.zulipchat.com/#narrow/stream/185405-t-compiler.2Frust-analyzer/topic/TIL.20that.20token.20streams.20are.20magic * https://rust-lang.zulipchat.com/#narrow/stream/131828-t-compiler/topic/Handling.20of.20Delimiter.3A.3ANone.20by.20the.20parser _Second_, it's not like our implementation in rust-analyzer works. We special-case expressions (as opposed to treating all kinds of $var captures the same) and we don't know how parser error recovery should work with these dollar-parenthesis. So, in this PR we simplify the whole thing away by not pretending that we are doing something proper and instead just explicitly special-casing expressions by wrapping them into real `()`. In the future, to maintain bug-parity with `rustc` what we are going to do is probably adding an explicit `CAPTURED_EXPR` *token* which we can explicitly account for in the parser. If/when rustc starts handling delimiter=none properly, we'll port that logic as well, in addition to special handling.
2021-10-23 17:08:42 +00:00
fn delim_to_str(d: tt::DelimiterKind, closing: bool) -> &'static str {
2019-12-18 03:47:26 +00:00
let texts = match d {
internal: replace L_DOLLAR/R_DOLLAR with parenthesis hack The general problem we are dealing with here is this: ``` macro_rules! thrice { ($e:expr) => { $e * 3} } fn main() { let x = thrice!(1 + 2); } ``` we really want this to print 9 rather than 7. The way rustc solves this is rather ad-hoc. In rustc, token trees are allowed to include whole AST fragments, so 1+2 is passed through macro expansion as a single unit. This is a significant violation of token tree model. In rust-analyzer, we intended to handle this in a more elegant way, using token trees with "invisible" delimiters. The idea was is that we introduce a new kind of parenthesis, "left $"/"right $", and let the parser intelligently handle this. The idea was inspired by the relevant comment in the proc_macro crate: https://doc.rust-lang.org/stable/proc_macro/enum.Delimiter.html#variant.None > An implicit delimiter, that may, for example, appear around tokens > coming from a “macro variable” $var. It is important to preserve > operator priorities in cases like $var * 3 where $var is 1 + 2. > Implicit delimiters might not survive roundtrip of a token stream > through a string. Now that we are older and wiser, we conclude that the idea doesn't work. _First_, the comment in the proc-macro crate is wishful thinking. Rustc currently completely ignores none delimiters. It solves the (1 + 2) * 3 problem by having magical token trees which can't be duplicated: * https://rust-lang.zulipchat.com/#narrow/stream/185405-t-compiler.2Frust-analyzer/topic/TIL.20that.20token.20streams.20are.20magic * https://rust-lang.zulipchat.com/#narrow/stream/131828-t-compiler/topic/Handling.20of.20Delimiter.3A.3ANone.20by.20the.20parser _Second_, it's not like our implementation in rust-analyzer works. We special-case expressions (as opposed to treating all kinds of $var captures the same) and we don't know how parser error recovery should work with these dollar-parenthesis. So, in this PR we simplify the whole thing away by not pretending that we are doing something proper and instead just explicitly special-casing expressions by wrapping them into real `()`. In the future, to maintain bug-parity with `rustc` what we are going to do is probably adding an explicit `CAPTURED_EXPR` *token* which we can explicitly account for in the parser. If/when rustc starts handling delimiter=none properly, we'll port that logic as well, in addition to special handling.
2021-10-23 17:08:42 +00:00
tt::DelimiterKind::Parenthesis => "()",
tt::DelimiterKind::Brace => "{}",
tt::DelimiterKind::Bracket => "[]",
2019-05-27 14:56:21 +00:00
};
let idx = closing as usize;
2021-01-04 16:11:56 +00:00
&texts[idx..texts.len() - (1 - idx)]
}
impl<'a> TtTreeSink<'a> {
2020-04-18 11:28:07 +00:00
fn token(&mut self, kind: SyntaxKind, mut n_tokens: u8) {
2020-12-15 18:23:51 +00:00
if kind == LIFETIME_IDENT {
2020-04-18 11:28:07 +00:00
n_tokens = 2;
}
2020-03-04 15:38:58 +00:00
let mut last = self.cursor;
2019-02-23 14:21:56 +00:00
for _ in 0..n_tokens {
2021-01-04 16:11:56 +00:00
let tmp_str: SmolStr;
2019-05-27 14:56:21 +00:00
if self.cursor.eof() {
break;
}
2020-03-04 15:38:58 +00:00
last = self.cursor;
internal: replace L_DOLLAR/R_DOLLAR with parenthesis hack The general problem we are dealing with here is this: ``` macro_rules! thrice { ($e:expr) => { $e * 3} } fn main() { let x = thrice!(1 + 2); } ``` we really want this to print 9 rather than 7. The way rustc solves this is rather ad-hoc. In rustc, token trees are allowed to include whole AST fragments, so 1+2 is passed through macro expansion as a single unit. This is a significant violation of token tree model. In rust-analyzer, we intended to handle this in a more elegant way, using token trees with "invisible" delimiters. The idea was is that we introduce a new kind of parenthesis, "left $"/"right $", and let the parser intelligently handle this. The idea was inspired by the relevant comment in the proc_macro crate: https://doc.rust-lang.org/stable/proc_macro/enum.Delimiter.html#variant.None > An implicit delimiter, that may, for example, appear around tokens > coming from a “macro variable” $var. It is important to preserve > operator priorities in cases like $var * 3 where $var is 1 + 2. > Implicit delimiters might not survive roundtrip of a token stream > through a string. Now that we are older and wiser, we conclude that the idea doesn't work. _First_, the comment in the proc-macro crate is wishful thinking. Rustc currently completely ignores none delimiters. It solves the (1 + 2) * 3 problem by having magical token trees which can't be duplicated: * https://rust-lang.zulipchat.com/#narrow/stream/185405-t-compiler.2Frust-analyzer/topic/TIL.20that.20token.20streams.20are.20magic * https://rust-lang.zulipchat.com/#narrow/stream/131828-t-compiler/topic/Handling.20of.20Delimiter.3A.3ANone.20by.20the.20parser _Second_, it's not like our implementation in rust-analyzer works. We special-case expressions (as opposed to treating all kinds of $var captures the same) and we don't know how parser error recovery should work with these dollar-parenthesis. So, in this PR we simplify the whole thing away by not pretending that we are doing something proper and instead just explicitly special-casing expressions by wrapping them into real `()`. In the future, to maintain bug-parity with `rustc` what we are going to do is probably adding an explicit `CAPTURED_EXPR` *token* which we can explicitly account for in the parser. If/when rustc starts handling delimiter=none properly, we'll port that logic as well, in addition to special handling.
2021-10-23 17:08:42 +00:00
let text: &str = loop {
break match self.cursor.token_tree() {
Some(tt::buffer::TokenTreeRef::Leaf(leaf, _)) => {
// Mark the range if needed
let (text, id) = match leaf {
tt::Leaf::Ident(ident) => (&ident.text, ident.id),
tt::Leaf::Punct(punct) => {
assert!(punct.char.is_ascii());
let char = &(punct.char as u8);
tmp_str = SmolStr::new_inline(
std::str::from_utf8(std::slice::from_ref(char)).unwrap(),
);
(&tmp_str, punct.id)
}
tt::Leaf::Literal(lit) => (&lit.text, lit.id),
};
let range = TextRange::at(self.text_pos, TextSize::of(text.as_str()));
self.token_map.insert(id, range);
self.cursor = self.cursor.bump();
text
}
Some(tt::buffer::TokenTreeRef::Subtree(subtree, _)) => {
self.cursor = self.cursor.subtree().unwrap();
match subtree.delimiter {
Some(d) => {
self.open_delims.insert(d.id, self.text_pos);
delim_to_str(d.kind, false)
}
None => continue,
}
2019-05-27 14:56:21 +00:00
}
internal: replace L_DOLLAR/R_DOLLAR with parenthesis hack The general problem we are dealing with here is this: ``` macro_rules! thrice { ($e:expr) => { $e * 3} } fn main() { let x = thrice!(1 + 2); } ``` we really want this to print 9 rather than 7. The way rustc solves this is rather ad-hoc. In rustc, token trees are allowed to include whole AST fragments, so 1+2 is passed through macro expansion as a single unit. This is a significant violation of token tree model. In rust-analyzer, we intended to handle this in a more elegant way, using token trees with "invisible" delimiters. The idea was is that we introduce a new kind of parenthesis, "left $"/"right $", and let the parser intelligently handle this. The idea was inspired by the relevant comment in the proc_macro crate: https://doc.rust-lang.org/stable/proc_macro/enum.Delimiter.html#variant.None > An implicit delimiter, that may, for example, appear around tokens > coming from a “macro variable” $var. It is important to preserve > operator priorities in cases like $var * 3 where $var is 1 + 2. > Implicit delimiters might not survive roundtrip of a token stream > through a string. Now that we are older and wiser, we conclude that the idea doesn't work. _First_, the comment in the proc-macro crate is wishful thinking. Rustc currently completely ignores none delimiters. It solves the (1 + 2) * 3 problem by having magical token trees which can't be duplicated: * https://rust-lang.zulipchat.com/#narrow/stream/185405-t-compiler.2Frust-analyzer/topic/TIL.20that.20token.20streams.20are.20magic * https://rust-lang.zulipchat.com/#narrow/stream/131828-t-compiler/topic/Handling.20of.20Delimiter.3A.3ANone.20by.20the.20parser _Second_, it's not like our implementation in rust-analyzer works. We special-case expressions (as opposed to treating all kinds of $var captures the same) and we don't know how parser error recovery should work with these dollar-parenthesis. So, in this PR we simplify the whole thing away by not pretending that we are doing something proper and instead just explicitly special-casing expressions by wrapping them into real `()`. In the future, to maintain bug-parity with `rustc` what we are going to do is probably adding an explicit `CAPTURED_EXPR` *token* which we can explicitly account for in the parser. If/when rustc starts handling delimiter=none properly, we'll port that logic as well, in addition to special handling.
2021-10-23 17:08:42 +00:00
None => {
let parent = self.cursor.end().unwrap();
2019-12-18 03:47:26 +00:00
self.cursor = self.cursor.bump();
internal: replace L_DOLLAR/R_DOLLAR with parenthesis hack The general problem we are dealing with here is this: ``` macro_rules! thrice { ($e:expr) => { $e * 3} } fn main() { let x = thrice!(1 + 2); } ``` we really want this to print 9 rather than 7. The way rustc solves this is rather ad-hoc. In rustc, token trees are allowed to include whole AST fragments, so 1+2 is passed through macro expansion as a single unit. This is a significant violation of token tree model. In rust-analyzer, we intended to handle this in a more elegant way, using token trees with "invisible" delimiters. The idea was is that we introduce a new kind of parenthesis, "left $"/"right $", and let the parser intelligently handle this. The idea was inspired by the relevant comment in the proc_macro crate: https://doc.rust-lang.org/stable/proc_macro/enum.Delimiter.html#variant.None > An implicit delimiter, that may, for example, appear around tokens > coming from a “macro variable” $var. It is important to preserve > operator priorities in cases like $var * 3 where $var is 1 + 2. > Implicit delimiters might not survive roundtrip of a token stream > through a string. Now that we are older and wiser, we conclude that the idea doesn't work. _First_, the comment in the proc-macro crate is wishful thinking. Rustc currently completely ignores none delimiters. It solves the (1 + 2) * 3 problem by having magical token trees which can't be duplicated: * https://rust-lang.zulipchat.com/#narrow/stream/185405-t-compiler.2Frust-analyzer/topic/TIL.20that.20token.20streams.20are.20magic * https://rust-lang.zulipchat.com/#narrow/stream/131828-t-compiler/topic/Handling.20of.20Delimiter.3A.3ANone.20by.20the.20parser _Second_, it's not like our implementation in rust-analyzer works. We special-case expressions (as opposed to treating all kinds of $var captures the same) and we don't know how parser error recovery should work with these dollar-parenthesis. So, in this PR we simplify the whole thing away by not pretending that we are doing something proper and instead just explicitly special-casing expressions by wrapping them into real `()`. In the future, to maintain bug-parity with `rustc` what we are going to do is probably adding an explicit `CAPTURED_EXPR` *token* which we can explicitly account for in the parser. If/when rustc starts handling delimiter=none properly, we'll port that logic as well, in addition to special handling.
2021-10-23 17:08:42 +00:00
match parent.delimiter {
Some(d) => {
if let Some(open_delim) = self.open_delims.get(&d.id) {
let open_range = TextRange::at(*open_delim, TextSize::of('('));
let close_range =
TextRange::at(self.text_pos, TextSize::of('('));
self.token_map.insert_delim(d.id, open_range, close_range);
}
delim_to_str(d.kind, true)
2019-12-18 03:47:26 +00:00
}
internal: replace L_DOLLAR/R_DOLLAR with parenthesis hack The general problem we are dealing with here is this: ``` macro_rules! thrice { ($e:expr) => { $e * 3} } fn main() { let x = thrice!(1 + 2); } ``` we really want this to print 9 rather than 7. The way rustc solves this is rather ad-hoc. In rustc, token trees are allowed to include whole AST fragments, so 1+2 is passed through macro expansion as a single unit. This is a significant violation of token tree model. In rust-analyzer, we intended to handle this in a more elegant way, using token trees with "invisible" delimiters. The idea was is that we introduce a new kind of parenthesis, "left $"/"right $", and let the parser intelligently handle this. The idea was inspired by the relevant comment in the proc_macro crate: https://doc.rust-lang.org/stable/proc_macro/enum.Delimiter.html#variant.None > An implicit delimiter, that may, for example, appear around tokens > coming from a “macro variable” $var. It is important to preserve > operator priorities in cases like $var * 3 where $var is 1 + 2. > Implicit delimiters might not survive roundtrip of a token stream > through a string. Now that we are older and wiser, we conclude that the idea doesn't work. _First_, the comment in the proc-macro crate is wishful thinking. Rustc currently completely ignores none delimiters. It solves the (1 + 2) * 3 problem by having magical token trees which can't be duplicated: * https://rust-lang.zulipchat.com/#narrow/stream/185405-t-compiler.2Frust-analyzer/topic/TIL.20that.20token.20streams.20are.20magic * https://rust-lang.zulipchat.com/#narrow/stream/131828-t-compiler/topic/Handling.20of.20Delimiter.3A.3ANone.20by.20the.20parser _Second_, it's not like our implementation in rust-analyzer works. We special-case expressions (as opposed to treating all kinds of $var captures the same) and we don't know how parser error recovery should work with these dollar-parenthesis. So, in this PR we simplify the whole thing away by not pretending that we are doing something proper and instead just explicitly special-casing expressions by wrapping them into real `()`. In the future, to maintain bug-parity with `rustc` what we are going to do is probably adding an explicit `CAPTURED_EXPR` *token* which we can explicitly account for in the parser. If/when rustc starts handling delimiter=none properly, we'll port that logic as well, in addition to special handling.
2021-10-23 17:08:42 +00:00
None => continue,
2019-12-12 17:41:44 +00:00
}
}
internal: replace L_DOLLAR/R_DOLLAR with parenthesis hack The general problem we are dealing with here is this: ``` macro_rules! thrice { ($e:expr) => { $e * 3} } fn main() { let x = thrice!(1 + 2); } ``` we really want this to print 9 rather than 7. The way rustc solves this is rather ad-hoc. In rustc, token trees are allowed to include whole AST fragments, so 1+2 is passed through macro expansion as a single unit. This is a significant violation of token tree model. In rust-analyzer, we intended to handle this in a more elegant way, using token trees with "invisible" delimiters. The idea was is that we introduce a new kind of parenthesis, "left $"/"right $", and let the parser intelligently handle this. The idea was inspired by the relevant comment in the proc_macro crate: https://doc.rust-lang.org/stable/proc_macro/enum.Delimiter.html#variant.None > An implicit delimiter, that may, for example, appear around tokens > coming from a “macro variable” $var. It is important to preserve > operator priorities in cases like $var * 3 where $var is 1 + 2. > Implicit delimiters might not survive roundtrip of a token stream > through a string. Now that we are older and wiser, we conclude that the idea doesn't work. _First_, the comment in the proc-macro crate is wishful thinking. Rustc currently completely ignores none delimiters. It solves the (1 + 2) * 3 problem by having magical token trees which can't be duplicated: * https://rust-lang.zulipchat.com/#narrow/stream/185405-t-compiler.2Frust-analyzer/topic/TIL.20that.20token.20streams.20are.20magic * https://rust-lang.zulipchat.com/#narrow/stream/131828-t-compiler/topic/Handling.20of.20Delimiter.3A.3ANone.20by.20the.20parser _Second_, it's not like our implementation in rust-analyzer works. We special-case expressions (as opposed to treating all kinds of $var captures the same) and we don't know how parser error recovery should work with these dollar-parenthesis. So, in this PR we simplify the whole thing away by not pretending that we are doing something proper and instead just explicitly special-casing expressions by wrapping them into real `()`. In the future, to maintain bug-parity with `rustc` what we are going to do is probably adding an explicit `CAPTURED_EXPR` *token* which we can explicitly account for in the parser. If/when rustc starts handling delimiter=none properly, we'll port that logic as well, in addition to special handling.
2021-10-23 17:08:42 +00:00
};
2019-05-27 14:56:21 +00:00
};
2021-06-13 03:54:16 +00:00
self.buf += text;
2021-01-04 16:11:56 +00:00
self.text_pos += TextSize::of(text);
2019-02-23 14:21:56 +00:00
}
2019-05-27 14:56:21 +00:00
self.inner.token(kind, self.buf.as_str());
2019-02-23 14:21:56 +00:00
self.buf.clear();
2019-05-27 14:56:21 +00:00
// Add whitespace between adjoint puncts
2020-03-04 15:38:58 +00:00
let next = last.bump();
2019-05-27 14:56:21 +00:00
if let (
2021-01-04 16:22:42 +00:00
Some(tt::buffer::TokenTreeRef::Leaf(tt::Leaf::Punct(curr), _)),
Some(tt::buffer::TokenTreeRef::Leaf(tt::Leaf::Punct(_), _)),
2020-03-04 15:38:58 +00:00
) = (last.token_tree(), next.token_tree())
2019-05-27 14:56:21 +00:00
{
2020-03-05 20:32:08 +00:00
// Note: We always assume the semi-colon would be the last token in
// other parts of RA such that we don't add whitespace here.
if curr.spacing == tt::Spacing::Alone && curr.char != ';' {
self.inner.token(WHITESPACE, " ");
2020-04-24 21:40:41 +00:00
self.text_pos += TextSize::of(' ');
}
}
2019-02-23 14:21:56 +00:00
}
2019-03-30 10:25:53 +00:00
fn start_node(&mut self, kind: SyntaxKind) {
self.inner.start_node(kind);
match self.roots.last_mut() {
None | Some(0) => self.roots.push(1),
Some(ref mut n) => **n += 1,
};
2019-02-23 14:21:56 +00:00
}
2019-03-30 10:25:53 +00:00
fn finish_node(&mut self) {
self.inner.finish_node();
*self.roots.last_mut().unwrap() -= 1;
2019-02-23 14:21:56 +00:00
}
fn error(&mut self, error: String) {
2019-02-23 14:21:56 +00:00
self.inner.error(error, self.text_pos)
}
}