1 //! Macro support for format strings
3 //! These structures are used when parsing format strings for the compiler.
4 //! Parsing does not happen at runtime: structures of `std::fmt::rt` are
8 html_root_url = "https://doc.rust-lang.org/nightly/",
9 html_playground_url = "https://play.rust-lang.org/",
10 test(attr(deny(warnings)))
13 #![feature(or_patterns)]
14 #![feature(rustc_private)]
15 #![feature(unicode_internals)]
16 #![feature(bool_to_option)]
28 use rustc_span::{InnerSpan, Symbol};
30 /// The type of format string that we are parsing.
31 #[derive(Copy, Clone, Debug, Eq, PartialEq)]
33 /// A normal format string as per `format_args!`.
35 /// An inline assembly template string for `asm!`.
39 #[derive(Copy, Clone)]
40 struct InnerOffset(usize);
43 fn to(self, end: InnerOffset) -> InnerSpan {
44 InnerSpan::new(self.0, end.0)
48 /// A piece is a portion of the format string which represents the next part
49 /// to emit. These are emitted as a stream by the `Parser` class.
50 #[derive(Copy, Clone, Debug, PartialEq)]
52 /// A literal string which should directly be emitted
54 /// This describes that formatting should process the next argument (as
55 /// specified inside) for emission.
56 NextArgument(Argument<'a>),
59 /// Representation of an argument specification.
60 #[derive(Copy, Clone, Debug, PartialEq)]
61 pub struct Argument<'a> {
62 /// Where to find this argument
63 pub position: Position,
64 /// How to format the argument
65 pub format: FormatSpec<'a>,
68 /// Specification for the formatting of an argument in the format string.
69 #[derive(Copy, Clone, Debug, PartialEq)]
70 pub struct FormatSpec<'a> {
71 /// Optionally specified character to fill alignment with.
72 pub fill: Option<char>,
73 /// Optionally specified alignment.
75 /// Packed version of various flags provided.
77 /// The integer precision to use.
79 /// The span of the precision formatting flag (for diagnostics).
80 pub precision_span: Option<InnerSpan>,
81 /// The string width requested for the resulting format.
83 /// The span of the width formatting flag (for diagnostics).
84 pub width_span: Option<InnerSpan>,
85 /// The descriptor string representing the name of the format desired for
86 /// this argument, this can be empty or any number of characters, although
87 /// it is required to be one word.
89 /// The span of the descriptor string (for diagnostics).
90 pub ty_span: Option<InnerSpan>,
93 /// Enum describing where an argument for a format can be located.
94 #[derive(Copy, Clone, Debug, PartialEq)]
96 /// The argument is implied to be located at an index
97 ArgumentImplicitlyIs(usize),
98 /// The argument is located at a specific index given in the format
100 /// The argument has a name.
101 ArgumentNamed(Symbol),
105 pub fn index(&self) -> Option<usize> {
107 ArgumentIs(i) | ArgumentImplicitlyIs(i) => Some(*i),
113 /// Enum of alignments which are supported.
114 #[derive(Copy, Clone, Debug, PartialEq)]
116 /// The value will be aligned to the left.
118 /// The value will be aligned to the right.
120 /// The value will be aligned in the center.
122 /// The value will take on a default alignment.
126 /// Various flags which can be applied to format strings. The meaning of these
127 /// flags is defined by the formatters themselves.
128 #[derive(Copy, Clone, Debug, PartialEq)]
130 /// A `+` will be used to denote positive numbers.
132 /// A `-` will be used to denote negative numbers. This is the default.
134 /// An alternate form will be used for the value. In the case of numbers,
135 /// this means that the number will be prefixed with the supplied string.
137 /// For numbers, this means that the number will be padded with zeroes,
138 /// and the sign (`+` or `-`) will precede them.
139 FlagSignAwareZeroPad,
140 /// For Debug / `?`, format integers in lower-case hexadecimal.
142 /// For Debug / `?`, format integers in upper-case hexadecimal.
146 /// A count is used for the precision and width parameters of an integer, and
147 /// can reference either an argument or a literal integer.
148 #[derive(Copy, Clone, Debug, PartialEq)]
150 /// The count is specified explicitly.
152 /// The count is specified by the argument with the given name.
154 /// The count is specified by the argument at the given index.
156 /// The count is implied and cannot be explicitly specified.
160 pub struct ParseError {
161 pub description: string::String,
162 pub note: Option<string::String>,
163 pub label: string::String,
165 pub secondary_label: Option<(string::String, InnerSpan)>,
168 /// The parser structure for interpreting the input format string. This is
169 /// modeled as an iterator over `Piece` structures to form a stream of tokens
172 /// This is a recursive-descent parser for the sake of simplicity, and if
173 /// necessary there's probably lots of room for improvement performance-wise.
174 pub struct Parser<'a> {
177 cur: iter::Peekable<str::CharIndices<'a>>,
178 /// Error messages accumulated during parsing
179 pub errors: Vec<ParseError>,
180 /// Current position of implicit positional argument pointer
182 /// `Some(raw count)` when the string is "raw", used to position spans correctly
183 style: Option<usize>,
184 /// Start and end byte offset of every successfully parsed argument
185 pub arg_places: Vec<InnerSpan>,
186 /// Characters that need to be shifted
188 /// Span of the last opening brace seen, used for error reporting
189 last_opening_brace: Option<InnerSpan>,
190 /// Whether the source string is comes from `println!` as opposed to `format!` or `print!`
191 append_newline: bool,
192 /// Whether this formatting string is a literal or it comes from a macro.
194 /// Start position of the current line.
195 cur_line_start: usize,
196 /// Start and end byte offset of every line of the format string. Excludes
197 /// newline characters and leading whitespace.
198 pub line_spans: Vec<InnerSpan>,
201 impl<'a> Iterator for Parser<'a> {
202 type Item = Piece<'a>;
204 fn next(&mut self) -> Option<Piece<'a>> {
205 if let Some(&(pos, c)) = self.cur.peek() {
208 let curr_last_brace = self.last_opening_brace;
209 let byte_pos = self.to_span_index(pos);
210 self.last_opening_brace = Some(byte_pos.to(InnerOffset(byte_pos.0 + 1)));
212 if self.consume('{') {
213 self.last_opening_brace = curr_last_brace;
215 Some(String(self.string(pos + 1)))
217 let arg = self.argument();
218 if let Some(end) = self.must_consume('}') {
219 let start = self.to_span_index(pos);
220 let end = self.to_span_index(end + 1);
222 self.arg_places.push(start.to(end));
225 Some(NextArgument(arg))
230 if self.consume('}') {
231 Some(String(self.string(pos + 1)))
233 let err_pos = self.to_span_index(pos);
235 "unmatched `}` found",
237 "if you intended to print `}`, you can escape it using `}}`",
243 _ => Some(String(self.string(pos))),
246 if self.is_literal && self.cur_line_start != self.input.len() {
247 let start = self.to_span_index(self.cur_line_start);
248 let end = self.to_span_index(self.input.len());
249 self.line_spans.push(start.to(end));
250 self.cur_line_start = self.input.len();
257 impl<'a> Parser<'a> {
258 /// Creates a new parser for the given format string
261 style: Option<usize>,
262 snippet: Option<string::String>,
263 append_newline: bool,
266 let (skips, is_literal) = find_skips_from_snippet(snippet, style);
270 cur: s.char_indices().peekable(),
276 last_opening_brace: None,
284 /// Notifies of an error. The message doesn't actually need to be of type
285 /// String, but I think it does when this eventually uses conditions so it
286 /// might as well start using it now.
287 fn err<S1: Into<string::String>, S2: Into<string::String>>(
293 self.errors.push(ParseError {
294 description: description.into(),
298 secondary_label: None,
302 /// Notifies of an error. The message doesn't actually need to be of type
303 /// String, but I think it does when this eventually uses conditions so it
304 /// might as well start using it now.
306 S1: Into<string::String>,
307 S2: Into<string::String>,
308 S3: Into<string::String>,
316 self.errors.push(ParseError {
317 description: description.into(),
318 note: Some(note.into()),
321 secondary_label: None,
325 /// Optionally consumes the specified character. If the character is not at
326 /// the current position, then the current iterator isn't moved and `false` is
327 /// returned, otherwise the character is consumed and `true` is returned.
328 fn consume(&mut self, c: char) -> bool {
329 self.consume_pos(c).is_some()
332 /// Optionally consumes the specified character. If the character is not at
333 /// the current position, then the current iterator isn't moved and `None` is
334 /// returned, otherwise the character is consumed and the current position is
336 fn consume_pos(&mut self, c: char) -> Option<usize> {
337 if let Some(&(pos, maybe)) = self.cur.peek() {
346 fn to_span_index(&self, pos: usize) -> InnerOffset {
348 // This handles the raw string case, the raw argument is the number of #
349 // in r###"..."### (we need to add one because of the `r`).
350 let raw = self.style.map(|raw| raw + 1).unwrap_or(0);
351 for skip in &self.skips {
354 } else if pos == *skip && raw == 0 {
360 InnerOffset(raw + pos + 1)
363 /// Forces consumption of the specified character. If the character is not
364 /// found, an error is emitted.
365 fn must_consume(&mut self, c: char) -> Option<usize> {
368 if let Some(&(pos, maybe)) = self.cur.peek() {
373 let pos = self.to_span_index(pos);
374 let description = format!("expected `'}}'`, found `{:?}`", maybe);
375 let label = "expected `}`".to_owned();
376 let (note, secondary_label) = if c == '}' {
379 "if you intended to print `{`, you can escape it using `{{`".to_owned(),
381 self.last_opening_brace
382 .map(|sp| ("because of this opening brace".to_owned(), sp)),
387 self.errors.push(ParseError {
397 let description = format!("expected `{:?}` but string was terminated", c);
398 // point at closing `"`
399 let pos = self.input.len() - if self.append_newline { 1 } else { 0 };
400 let pos = self.to_span_index(pos);
402 let label = format!("expected `{:?}`", c);
403 let (note, secondary_label) = if c == '}' {
406 "if you intended to print `{`, you can escape it using `{{`".to_owned(),
408 self.last_opening_brace
409 .map(|sp| ("because of this opening brace".to_owned(), sp)),
414 self.errors.push(ParseError {
422 self.err(description, format!("expected `{:?}`", c), pos.to(pos));
428 /// Consumes all whitespace characters until the first non-whitespace character
430 while let Some(&(_, c)) = self.cur.peek() {
431 if c.is_whitespace() {
439 /// Parses all of a string which is to be considered a "raw literal" in a
440 /// format string. This is everything outside of the braces.
441 fn string(&mut self, start: usize) -> &'a str {
442 // we may not consume the character, peek the iterator
443 while let Some(&(pos, c)) = self.cur.peek() {
446 return &self.input[start..pos];
448 '\n' if self.is_literal => {
449 let start = self.to_span_index(self.cur_line_start);
450 let end = self.to_span_index(pos);
451 self.line_spans.push(start.to(end));
452 self.cur_line_start = pos + 1;
456 if self.is_literal && pos == self.cur_line_start && c.is_whitespace() {
457 self.cur_line_start = pos + c.len_utf8();
463 &self.input[start..self.input.len()]
466 /// Parses an `Argument` structure, or what's contained within braces inside the format string.
467 fn argument(&mut self) -> Argument<'a> {
468 let pos = self.position();
469 let format = match self.mode {
470 ParseMode::Format => self.format(),
471 ParseMode::InlineAsm => self.inline_asm(),
474 // Resolve position after parsing format spec.
475 let pos = match pos {
476 Some(position) => position,
480 ArgumentImplicitlyIs(i)
484 Argument { position: pos, format }
487 /// Parses a positional argument for a format. This could either be an
488 /// integer index of an argument, a named argument, or a blank string.
489 /// Returns `Some(parsed_position)` if the position is not implicitly
490 /// consuming a macro argument, `None` if it's the case.
491 fn position(&mut self) -> Option<Position> {
492 if let Some(i) = self.integer() {
495 match self.cur.peek() {
496 Some(&(_, c)) if rustc_lexer::is_id_start(c) => {
497 Some(ArgumentNamed(Symbol::intern(self.word())))
500 // This is an `ArgumentNext`.
501 // Record the fact and do the resolution after parsing the
502 // format spec, to make things like `{:.*}` work.
508 /// Parses a format specifier at the current position, returning all of the
509 /// relevant information in the `FormatSpec` struct.
510 fn format(&mut self) -> FormatSpec<'a> {
511 let mut spec = FormatSpec {
515 precision: CountImplied,
516 precision_span: None,
519 ty: &self.input[..0],
522 if !self.consume(':') {
527 if let Some(&(_, c)) = self.cur.peek() {
528 match self.cur.clone().nth(1) {
529 Some((_, '>' | '<' | '^')) => {
537 if self.consume('<') {
538 spec.align = AlignLeft;
539 } else if self.consume('>') {
540 spec.align = AlignRight;
541 } else if self.consume('^') {
542 spec.align = AlignCenter;
545 if self.consume('+') {
546 spec.flags |= 1 << (FlagSignPlus as u32);
547 } else if self.consume('-') {
548 spec.flags |= 1 << (FlagSignMinus as u32);
551 if self.consume('#') {
552 spec.flags |= 1 << (FlagAlternate as u32);
554 // Width and precision
555 let mut havewidth = false;
557 if self.consume('0') {
558 // small ambiguity with '0$' as a format string. In theory this is a
559 // '0' flag and then an ill-formatted format string with just a '$'
560 // and no count, but this is better if we instead interpret this as
561 // no '0' flag and '0$' as the width instead.
562 if self.consume('$') {
563 spec.width = CountIsParam(0);
566 spec.flags |= 1 << (FlagSignAwareZeroPad as u32);
570 let width_span_start = if let Some((pos, _)) = self.cur.peek() { *pos } else { 0 };
571 let (w, sp) = self.count(width_span_start);
573 spec.width_span = sp;
575 if let Some(start) = self.consume_pos('.') {
576 if let Some(end) = self.consume_pos('*') {
577 // Resolve `CountIsNextParam`.
578 // We can do this immediately as `position` is resolved later.
581 spec.precision = CountIsParam(i);
582 spec.precision_span =
583 Some(self.to_span_index(start).to(self.to_span_index(end + 1)));
585 let (p, sp) = self.count(start);
587 spec.precision_span = sp;
590 let ty_span_start = self.cur.peek().map(|(pos, _)| *pos);
591 // Optional radix followed by the actual format specifier
592 if self.consume('x') {
593 if self.consume('?') {
594 spec.flags |= 1 << (FlagDebugLowerHex as u32);
599 } else if self.consume('X') {
600 if self.consume('?') {
601 spec.flags |= 1 << (FlagDebugUpperHex as u32);
606 } else if self.consume('?') {
609 spec.ty = self.word();
610 let ty_span_end = self.cur.peek().map(|(pos, _)| *pos);
611 if !spec.ty.is_empty() {
612 spec.ty_span = ty_span_start
613 .and_then(|s| ty_span_end.map(|e| (s, e)))
614 .map(|(start, end)| self.to_span_index(start).to(self.to_span_index(end)));
620 /// Parses an inline assembly template modifier at the current position, returning the modifier
621 /// in the `ty` field of the `FormatSpec` struct.
622 fn inline_asm(&mut self) -> FormatSpec<'a> {
623 let mut spec = FormatSpec {
627 precision: CountImplied,
628 precision_span: None,
631 ty: &self.input[..0],
634 if !self.consume(':') {
638 let ty_span_start = self.cur.peek().map(|(pos, _)| *pos);
639 spec.ty = self.word();
640 let ty_span_end = self.cur.peek().map(|(pos, _)| *pos);
641 if !spec.ty.is_empty() {
642 spec.ty_span = ty_span_start
643 .and_then(|s| ty_span_end.map(|e| (s, e)))
644 .map(|(start, end)| self.to_span_index(start).to(self.to_span_index(end)));
650 /// Parses a `Count` parameter at the current position. This does not check
651 /// for 'CountIsNextParam' because that is only used in precision, not
653 fn count(&mut self, start: usize) -> (Count, Option<InnerSpan>) {
654 if let Some(i) = self.integer() {
655 if let Some(end) = self.consume_pos('$') {
656 let span = self.to_span_index(start).to(self.to_span_index(end + 1));
657 (CountIsParam(i), Some(span))
662 let tmp = self.cur.clone();
663 let word = self.word();
667 } else if self.consume('$') {
668 (CountIsName(Symbol::intern(word)), None)
676 /// Parses a word starting at the current position. A word is the same as
677 /// Rust identifier, except that it can't start with `_` character.
678 fn word(&mut self) -> &'a str {
679 let start = match self.cur.peek() {
680 Some(&(pos, c)) if rustc_lexer::is_id_start(c) => {
689 while let Some(&(pos, c)) = self.cur.peek() {
690 if rustc_lexer::is_id_continue(c) {
697 let end = end.unwrap_or(self.input.len());
698 let word = &self.input[start..end];
701 "invalid argument name `_`",
702 "invalid argument name",
703 "argument name cannot be a single underscore",
704 self.to_span_index(start).to(self.to_span_index(end)),
710 /// Optionally parses an integer at the current position. This doesn't deal
711 /// with overflow at all, it's just accumulating digits.
712 fn integer(&mut self) -> Option<usize> {
714 let mut found = false;
715 while let Some(&(_, c)) = self.cur.peek() {
716 if let Some(i) = c.to_digit(10) {
717 cur = cur * 10 + i as usize;
728 /// Finds the indices of all characters that have been processed and differ between the actual
729 /// written code (code snippet) and the `InternedString` that gets processed in the `Parser`
730 /// in order to properly synthethise the intra-string `Span`s for error diagnostics.
731 fn find_skips_from_snippet(
732 snippet: Option<string::String>,
733 str_style: Option<usize>,
734 ) -> (Vec<usize>, bool) {
735 let snippet = match snippet {
736 Some(ref s) if s.starts_with('"') || s.starts_with("r#") => s,
737 _ => return (vec![], false),
740 fn find_skips(snippet: &str, is_raw: bool) -> Vec<usize> {
741 let mut eat_ws = false;
742 let mut s = snippet.chars().enumerate().peekable();
743 let mut skips = vec![];
744 while let Some((pos, c)) = s.next() {
745 match (c, s.peek()) {
746 // skip whitespace and empty lines ending in '\\'
747 ('\\', Some((next_pos, '\n'))) if !is_raw => {
750 skips.push(*next_pos);
753 ('\\', Some((next_pos, '\n' | 'n' | 't'))) if eat_ws => {
755 skips.push(*next_pos);
758 (' ' | '\n' | '\t', _) if eat_ws => {
761 ('\\', Some((next_pos, 'n' | 't' | '0' | '\\' | '\'' | '\"'))) => {
762 skips.push(*next_pos);
765 ('\\', Some((_, 'x'))) if !is_raw => {
767 // consume `\xAB` literal
768 if let Some((pos, _)) = s.next() {
775 ('\\', Some((_, 'u'))) if !is_raw => {
776 if let Some((pos, _)) = s.next() {
779 if let Some((next_pos, next_c)) = s.next() {
781 skips.push(next_pos);
782 let mut i = 0; // consume up to 6 hexanumeric chars + closing `}`
783 while let (Some((next_pos, c)), true) = (s.next(), i < 7) {
785 skips.push(next_pos);
787 skips.push(next_pos);
794 } else if next_c.is_digit(16) {
795 skips.push(next_pos);
796 // We suggest adding `{` and `}` when appropriate, accept it here as if
798 let mut i = 0; // consume up to 6 hexanumeric chars
799 while let (Some((next_pos, c)), _) = (s.next(), i < 6) {
801 skips.push(next_pos);
811 // `take_while(|c| c.is_whitespace())`
820 let r_start = str_style.map(|r| r + 1).unwrap_or(0);
821 let r_end = str_style.map(|r| r).unwrap_or(0);
822 let s = &snippet[r_start + 1..snippet.len() - r_end - 1];
823 (find_skips(s, str_style.is_some()), true)