1 use rustc_ast::ast::AttrStyle;
2 use rustc_ast::token::{self, CommentKind, Token, TokenKind};
3 use rustc_ast::tokenstream::{Spacing, TokenStream};
4 use rustc_errors::{error_code, Applicability, DiagnosticBuilder, FatalError, PResult};
5 use rustc_lexer::unescape::{self, Mode};
6 use rustc_lexer::{Base, DocStyle, RawStrError};
7 use rustc_session::parse::ParseSess;
8 use rustc_span::symbol::{sym, Symbol};
9 use rustc_span::{BytePos, Pos, Span};
14 mod unescape_error_reporting;
17 use unescape_error_reporting::{emit_unescape_error, escaped_char};
19 #[derive(Clone, Debug)]
20 pub struct UnmatchedBrace {
21 pub expected_delim: token::DelimToken,
22 pub found_delim: Option<token::DelimToken>,
24 pub unclosed_span: Option<Span>,
25 pub candidate_span: Option<Span>,
28 crate fn parse_token_trees<'a>(
32 override_span: Option<Span>,
33 ) -> (PResult<'a, TokenStream>, Vec<UnmatchedBrace>) {
34 StringReader { sess, start_pos, pos: start_pos, end_src_index: src.len(), src, override_span }
38 struct StringReader<'a> {
40 /// Initial position, read-only.
42 /// The absolute offset within the source_map of the current character.
44 /// Stop reading src at this index.
46 /// Source text to tokenize.
48 override_span: Option<Span>,
51 impl<'a> StringReader<'a> {
52 fn mk_sp(&self, lo: BytePos, hi: BytePos) -> Span {
53 self.override_span.unwrap_or_else(|| Span::with_root_ctxt(lo, hi))
56 /// Returns the next token, and info about preceding whitespace, if any.
57 fn next_token(&mut self) -> (Spacing, Token) {
58 let mut spacing = Spacing::Joint;
60 // Skip `#!` at the start of the file
61 let start_src_index = self.src_index(self.pos);
62 let text: &str = &self.src[start_src_index..self.end_src_index];
63 let is_beginning_of_file = self.pos == self.start_pos;
64 if is_beginning_of_file {
65 if let Some(shebang_len) = rustc_lexer::strip_shebang(text) {
66 self.pos = self.pos + BytePos::from_usize(shebang_len);
67 spacing = Spacing::Alone;
71 // Skip trivial (whitespace & comments) tokens
73 let start_src_index = self.src_index(self.pos);
74 let text: &str = &self.src[start_src_index..self.end_src_index];
77 let span = self.mk_sp(self.pos, self.pos);
78 return (spacing, Token::new(token::Eof, span));
81 let token = rustc_lexer::first_token(text);
84 self.pos = self.pos + BytePos::from_usize(token.len);
86 debug!("next_token: {:?}({:?})", token.kind, self.str_from(start));
88 match self.cook_lexer_token(token.kind, start) {
90 let span = self.mk_sp(start, self.pos);
91 return (spacing, Token::new(kind, span));
93 None => spacing = Spacing::Alone,
98 /// Report a fatal lexical error with a given span.
99 fn fatal_span(&self, sp: Span, m: &str) -> FatalError {
100 self.sess.span_diagnostic.span_fatal(sp, m)
103 /// Report a lexical error with a given span.
104 fn err_span(&self, sp: Span, m: &str) {
105 self.sess.span_diagnostic.struct_span_err(sp, m).emit();
108 /// Report a fatal error spanning [`from_pos`, `to_pos`).
109 fn fatal_span_(&self, from_pos: BytePos, to_pos: BytePos, m: &str) -> FatalError {
110 self.fatal_span(self.mk_sp(from_pos, to_pos), m)
113 /// Report a lexical error spanning [`from_pos`, `to_pos`).
114 fn err_span_(&self, from_pos: BytePos, to_pos: BytePos, m: &str) {
115 self.err_span(self.mk_sp(from_pos, to_pos), m)
118 fn struct_fatal_span_char(
124 ) -> DiagnosticBuilder<'a> {
127 .struct_span_fatal(self.mk_sp(from_pos, to_pos), &format!("{}: {}", m, escaped_char(c)))
130 /// Turns simple `rustc_lexer::TokenKind` enum into a rich
131 /// `rustc_ast::TokenKind`. This turns strings into interned
132 /// symbols and runs additional validation.
133 fn cook_lexer_token(&self, token: rustc_lexer::TokenKind, start: BytePos) -> Option<TokenKind> {
135 rustc_lexer::TokenKind::LineComment { doc_style } => {
136 // Skip non-doc comments
137 let doc_style = doc_style?;
139 // Opening delimiter of the length 3 is not included into the symbol.
140 let content_start = start + BytePos(3);
141 let content = self.str_from(content_start);
142 self.cook_doc_comment(content_start, content, CommentKind::Line, doc_style)
144 rustc_lexer::TokenKind::BlockComment { doc_style, terminated } => {
146 let msg = match doc_style {
147 Some(_) => "unterminated block doc-comment",
148 None => "unterminated block comment",
150 let last_bpos = self.pos;
151 self.sess.span_diagnostic.span_fatal_with_code(
152 self.mk_sp(start, last_bpos),
158 // Skip non-doc comments
159 let doc_style = doc_style?;
161 // Opening delimiter of the length 3 and closing delimiter of the length 2
162 // are not included into the symbol.
163 let content_start = start + BytePos(3);
164 let content_end = self.pos - BytePos(if terminated { 2 } else { 0 });
165 let content = self.str_from_to(content_start, content_end);
166 self.cook_doc_comment(content_start, content, CommentKind::Block, doc_style)
168 rustc_lexer::TokenKind::Whitespace => return None,
169 rustc_lexer::TokenKind::Ident | rustc_lexer::TokenKind::RawIdent => {
170 let is_raw_ident = token == rustc_lexer::TokenKind::RawIdent;
171 let mut ident_start = start;
173 ident_start = ident_start + BytePos(2);
175 let sym = nfc_normalize(self.str_from(ident_start));
176 let span = self.mk_sp(start, self.pos);
177 self.sess.symbol_gallery.insert(sym, span);
179 if !sym.can_be_raw() {
180 self.err_span(span, &format!("`{}` cannot be a raw identifier", sym));
182 self.sess.raw_identifier_spans.borrow_mut().push(span);
184 token::Ident(sym, is_raw_ident)
186 rustc_lexer::TokenKind::Literal { kind, suffix_start } => {
187 let suffix_start = start + BytePos(suffix_start as u32);
188 let (kind, symbol) = self.cook_lexer_literal(start, suffix_start, kind);
189 let suffix = if suffix_start < self.pos {
190 let string = self.str_from(suffix_start);
195 self.mk_sp(suffix_start, self.pos),
196 "underscore literal suffix is not allowed",
199 "this was previously accepted by the compiler but is \
200 being phased out; it will become a hard error in \
205 <https://github.com/rust-lang/rust/issues/42326> \
206 for more information",
211 Some(Symbol::intern(string))
216 token::Literal(token::Lit { kind, symbol, suffix })
218 rustc_lexer::TokenKind::Lifetime { starts_with_number } => {
219 // Include the leading `'` in the real identifier, for macro
220 // expansion purposes. See #12512 for the gory details of why
221 // this is necessary.
222 let lifetime_name = self.str_from(start);
223 if starts_with_number {
224 self.err_span_(start, self.pos, "lifetimes cannot start with a number");
226 let ident = Symbol::intern(lifetime_name);
227 token::Lifetime(ident)
229 rustc_lexer::TokenKind::Semi => token::Semi,
230 rustc_lexer::TokenKind::Comma => token::Comma,
231 rustc_lexer::TokenKind::Dot => token::Dot,
232 rustc_lexer::TokenKind::OpenParen => token::OpenDelim(token::Paren),
233 rustc_lexer::TokenKind::CloseParen => token::CloseDelim(token::Paren),
234 rustc_lexer::TokenKind::OpenBrace => token::OpenDelim(token::Brace),
235 rustc_lexer::TokenKind::CloseBrace => token::CloseDelim(token::Brace),
236 rustc_lexer::TokenKind::OpenBracket => token::OpenDelim(token::Bracket),
237 rustc_lexer::TokenKind::CloseBracket => token::CloseDelim(token::Bracket),
238 rustc_lexer::TokenKind::At => token::At,
239 rustc_lexer::TokenKind::Pound => token::Pound,
240 rustc_lexer::TokenKind::Tilde => token::Tilde,
241 rustc_lexer::TokenKind::Question => token::Question,
242 rustc_lexer::TokenKind::Colon => token::Colon,
243 rustc_lexer::TokenKind::Dollar => token::Dollar,
244 rustc_lexer::TokenKind::Eq => token::Eq,
245 rustc_lexer::TokenKind::Bang => token::Not,
246 rustc_lexer::TokenKind::Lt => token::Lt,
247 rustc_lexer::TokenKind::Gt => token::Gt,
248 rustc_lexer::TokenKind::Minus => token::BinOp(token::Minus),
249 rustc_lexer::TokenKind::And => token::BinOp(token::And),
250 rustc_lexer::TokenKind::Or => token::BinOp(token::Or),
251 rustc_lexer::TokenKind::Plus => token::BinOp(token::Plus),
252 rustc_lexer::TokenKind::Star => token::BinOp(token::Star),
253 rustc_lexer::TokenKind::Slash => token::BinOp(token::Slash),
254 rustc_lexer::TokenKind::Caret => token::BinOp(token::Caret),
255 rustc_lexer::TokenKind::Percent => token::BinOp(token::Percent),
257 rustc_lexer::TokenKind::Unknown => {
258 let c = self.str_from(start).chars().next().unwrap();
260 self.struct_fatal_span_char(start, self.pos, "unknown start of token", c);
261 // FIXME: the lexer could be used to turn the ASCII version of unicode homoglyphs,
262 // instead of keeping a table in `check_for_substitution`into the token. Ideally,
263 // this should be inside `rustc_lexer`. However, we should first remove compound
264 // tokens like `<<` from `rustc_lexer`, and then add fancier error recovery to it,
265 // as there will be less overall work to do this way.
266 let token = unicode_chars::check_for_substitution(self, start, c, &mut err);
268 err.help("source files must contain UTF-8 encoded text, unexpected null bytes might occur when a different encoding is used");
278 content_start: BytePos,
280 comment_kind: CommentKind,
283 if content.contains('\r') {
284 for (idx, _) in content.char_indices().filter(|&(_, c)| c == '\r') {
286 content_start + BytePos(idx as u32),
287 content_start + BytePos(idx as u32 + 1),
289 CommentKind::Line => "bare CR not allowed in doc-comment",
290 CommentKind::Block => "bare CR not allowed in block doc-comment",
296 let attr_style = match doc_style {
297 DocStyle::Outer => AttrStyle::Outer,
298 DocStyle::Inner => AttrStyle::Inner,
301 token::DocComment(comment_kind, attr_style, Symbol::intern(content))
304 fn cook_lexer_literal(
307 suffix_start: BytePos,
308 kind: rustc_lexer::LiteralKind,
309 ) -> (token::LitKind, Symbol) {
310 // prefix means `"` or `br"` or `r###"`, ...
311 let (lit_kind, mode, prefix_len, postfix_len) = match kind {
312 rustc_lexer::LiteralKind::Char { terminated } => {
314 self.sess.span_diagnostic.span_fatal_with_code(
315 self.mk_sp(start, suffix_start),
316 "unterminated character literal",
320 (token::Char, Mode::Char, 1, 1) // ' '
322 rustc_lexer::LiteralKind::Byte { terminated } => {
324 self.sess.span_diagnostic.span_fatal_with_code(
325 self.mk_sp(start + BytePos(1), suffix_start),
326 "unterminated byte constant",
330 (token::Byte, Mode::Byte, 2, 1) // b' '
332 rustc_lexer::LiteralKind::Str { terminated } => {
334 self.sess.span_diagnostic.span_fatal_with_code(
335 self.mk_sp(start, suffix_start),
336 "unterminated double quote string",
340 (token::Str, Mode::Str, 1, 1) // " "
342 rustc_lexer::LiteralKind::ByteStr { terminated } => {
344 self.sess.span_diagnostic.span_fatal_with_code(
345 self.mk_sp(start + BytePos(1), suffix_start),
346 "unterminated double quote byte string",
350 (token::ByteStr, Mode::ByteStr, 2, 1) // b" "
352 rustc_lexer::LiteralKind::RawStr { n_hashes, err } => {
353 self.report_raw_str_error(start, err);
354 let n = u32::from(n_hashes);
355 (token::StrRaw(n_hashes), Mode::RawStr, 2 + n, 1 + n) // r##" "##
357 rustc_lexer::LiteralKind::RawByteStr { n_hashes, err } => {
358 self.report_raw_str_error(start, err);
359 let n = u32::from(n_hashes);
360 (token::ByteStrRaw(n_hashes), Mode::RawByteStr, 3 + n, 1 + n) // br##" "##
362 rustc_lexer::LiteralKind::Int { base, empty_int } => {
363 return if empty_int {
366 .struct_span_err_with_code(
367 self.mk_sp(start, suffix_start),
368 "no valid digits found for number",
372 (token::Integer, sym::integer(0))
374 self.validate_int_literal(base, start, suffix_start);
375 (token::Integer, self.symbol_from_to(start, suffix_start))
378 rustc_lexer::LiteralKind::Float { base, empty_exponent } => {
380 self.err_span_(start, self.pos, "expected at least one digit in exponent");
384 Base::Hexadecimal => self.err_span_(
387 "hexadecimal float literal is not supported",
390 self.err_span_(start, suffix_start, "octal float literal is not supported")
393 self.err_span_(start, suffix_start, "binary float literal is not supported")
398 let id = self.symbol_from_to(start, suffix_start);
399 return (token::Float, id);
402 let content_start = start + BytePos(prefix_len);
403 let content_end = suffix_start - BytePos(postfix_len);
404 let id = self.symbol_from_to(content_start, content_end);
405 self.validate_literal_escape(mode, content_start, content_end, prefix_len, postfix_len);
410 fn src_index(&self, pos: BytePos) -> usize {
411 (pos - self.start_pos).to_usize()
414 /// Slice of the source text from `start` up to but excluding `self.pos`,
415 /// meaning the slice does not include the character `self.ch`.
416 fn str_from(&self, start: BytePos) -> &str {
417 self.str_from_to(start, self.pos)
420 /// As symbol_from, with an explicit endpoint.
421 fn symbol_from_to(&self, start: BytePos, end: BytePos) -> Symbol {
422 debug!("taking an ident from {:?} to {:?}", start, end);
423 Symbol::intern(self.str_from_to(start, end))
426 /// Slice of the source text spanning from `start` up to but excluding `end`.
427 fn str_from_to(&self, start: BytePos, end: BytePos) -> &str {
428 &self.src[self.src_index(start)..self.src_index(end)]
431 fn report_raw_str_error(&self, start: BytePos, opt_err: Option<RawStrError>) {
433 Some(RawStrError::InvalidStarter { bad_char }) => {
434 self.report_non_started_raw_string(start, bad_char)
436 Some(RawStrError::NoTerminator { expected, found, possible_terminator_offset }) => self
437 .report_unterminated_raw_string(start, expected, possible_terminator_offset, found),
438 Some(RawStrError::TooManyDelimiters { found }) => {
439 self.report_too_many_hashes(start, found)
445 fn report_non_started_raw_string(&self, start: BytePos, bad_char: char) -> ! {
446 self.struct_fatal_span_char(
449 "found invalid character; only `#` is allowed in raw string delimitation",
456 fn report_unterminated_raw_string(
460 possible_offset: Option<usize>,
461 found_terminators: usize,
463 let mut err = self.sess.span_diagnostic.struct_span_fatal_with_code(
464 self.mk_sp(start, start),
465 "unterminated raw string",
469 err.span_label(self.mk_sp(start, start), "unterminated raw string");
473 "this raw string should be terminated with `\"{}`",
478 if let Some(possible_offset) = possible_offset {
479 let lo = start + BytePos(possible_offset as u32);
480 let hi = lo + BytePos(found_terminators as u32);
481 let span = self.mk_sp(lo, hi);
484 "consider terminating the string here",
485 "#".repeat(n_hashes),
486 Applicability::MaybeIncorrect,
494 /// Note: It was decided to not add a test case, because it would be too big.
495 /// <https://github.com/rust-lang/rust/pull/50296#issuecomment-392135180>
496 fn report_too_many_hashes(&self, start: BytePos, found: usize) -> ! {
501 "too many `#` symbols: raw strings may be delimited \
502 by up to 65535 `#` symbols, but found {}",
509 fn validate_literal_escape(
512 content_start: BytePos,
513 content_end: BytePos,
517 let lit_content = self.str_from_to(content_start, content_end);
518 unescape::unescape_literal(lit_content, mode, &mut |range, result| {
519 // Here we only check for errors. The actual unescaping is done later.
520 if let Err(err) = result {
521 let span_with_quotes = self
522 .mk_sp(content_start - BytePos(prefix_len), content_end + BytePos(postfix_len));
523 let (start, end) = (range.start as u32, range.end as u32);
524 let lo = content_start + BytePos(start);
525 let hi = lo + BytePos(end - start);
526 let span = self.mk_sp(lo, hi);
528 &self.sess.span_diagnostic,
540 fn validate_int_literal(&self, base: Base, content_start: BytePos, content_end: BytePos) {
541 let base = match base {
546 let s = self.str_from_to(content_start + BytePos(2), content_end);
547 for (idx, c) in s.char_indices() {
548 let idx = idx as u32;
549 if c != '_' && c.to_digit(base).is_none() {
550 let lo = content_start + BytePos(2 + idx);
551 let hi = content_start + BytePos(2 + idx + c.len_utf8() as u32);
552 self.err_span_(lo, hi, &format!("invalid digit for a base {} literal", base));
558 pub fn nfc_normalize(string: &str) -> Symbol {
559 use unicode_normalization::{is_nfc_quick, IsNormalized, UnicodeNormalization};
560 match is_nfc_quick(string.chars()) {
561 IsNormalized::Yes => Symbol::intern(string),
563 let normalized_str: String = string.chars().nfc().collect();
564 Symbol::intern(&normalized_str)