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/nightly-rustc/",
9 html_playground_url = "https://play.rust-lang.org/",
10 test(attr(deny(warnings)))
13 #![cfg_attr(bootstrap, feature(or_patterns))]
14 #![feature(bool_to_option)]
26 use rustc_span::{InnerSpan, Symbol};
28 /// The type of format string that we are parsing.
29 #[derive(Copy, Clone, Debug, Eq, PartialEq)]
31 /// A normal format string as per `format_args!`.
33 /// An inline assembly template string for `asm!`.
37 #[derive(Copy, Clone)]
38 struct InnerOffset(usize);
41 fn to(self, end: InnerOffset) -> InnerSpan {
42 InnerSpan::new(self.0, end.0)
46 /// A piece is a portion of the format string which represents the next part
47 /// to emit. These are emitted as a stream by the `Parser` class.
48 #[derive(Copy, Clone, Debug, PartialEq)]
50 /// A literal string which should directly be emitted
52 /// This describes that formatting should process the next argument (as
53 /// specified inside) for emission.
54 NextArgument(Argument<'a>),
57 /// Representation of an argument specification.
58 #[derive(Copy, Clone, Debug, PartialEq)]
59 pub struct Argument<'a> {
60 /// Where to find this argument
61 pub position: Position,
62 /// How to format the argument
63 pub format: FormatSpec<'a>,
66 /// Specification for the formatting of an argument in the format string.
67 #[derive(Copy, Clone, Debug, PartialEq)]
68 pub struct FormatSpec<'a> {
69 /// Optionally specified character to fill alignment with.
70 pub fill: Option<char>,
71 /// Optionally specified alignment.
73 /// Packed version of various flags provided.
75 /// The integer precision to use.
77 /// The span of the precision formatting flag (for diagnostics).
78 pub precision_span: Option<InnerSpan>,
79 /// The string width requested for the resulting format.
81 /// The span of the width formatting flag (for diagnostics).
82 pub width_span: Option<InnerSpan>,
83 /// The descriptor string representing the name of the format desired for
84 /// this argument, this can be empty or any number of characters, although
85 /// it is required to be one word.
87 /// The span of the descriptor string (for diagnostics).
88 pub ty_span: Option<InnerSpan>,
91 /// Enum describing where an argument for a format can be located.
92 #[derive(Copy, Clone, Debug, PartialEq)]
94 /// The argument is implied to be located at an index
95 ArgumentImplicitlyIs(usize),
96 /// The argument is located at a specific index given in the format
98 /// The argument has a name.
99 ArgumentNamed(Symbol),
103 pub fn index(&self) -> Option<usize> {
105 ArgumentIs(i) | ArgumentImplicitlyIs(i) => Some(*i),
111 /// Enum of alignments which are supported.
112 #[derive(Copy, Clone, Debug, PartialEq)]
114 /// The value will be aligned to the left.
116 /// The value will be aligned to the right.
118 /// The value will be aligned in the center.
120 /// The value will take on a default alignment.
124 /// Various flags which can be applied to format strings. The meaning of these
125 /// flags is defined by the formatters themselves.
126 #[derive(Copy, Clone, Debug, PartialEq)]
128 /// A `+` will be used to denote positive numbers.
130 /// A `-` will be used to denote negative numbers. This is the default.
132 /// An alternate form will be used for the value. In the case of numbers,
133 /// this means that the number will be prefixed with the supplied string.
135 /// For numbers, this means that the number will be padded with zeroes,
136 /// and the sign (`+` or `-`) will precede them.
137 FlagSignAwareZeroPad,
138 /// For Debug / `?`, format integers in lower-case hexadecimal.
140 /// For Debug / `?`, format integers in upper-case hexadecimal.
144 /// A count is used for the precision and width parameters of an integer, and
145 /// can reference either an argument or a literal integer.
146 #[derive(Copy, Clone, Debug, PartialEq)]
148 /// The count is specified explicitly.
150 /// The count is specified by the argument with the given name.
152 /// The count is specified by the argument at the given index.
154 /// The count is implied and cannot be explicitly specified.
158 pub struct ParseError {
159 pub description: string::String,
160 pub note: Option<string::String>,
161 pub label: string::String,
163 pub secondary_label: Option<(string::String, InnerSpan)>,
166 /// The parser structure for interpreting the input format string. This is
167 /// modeled as an iterator over `Piece` structures to form a stream of tokens
170 /// This is a recursive-descent parser for the sake of simplicity, and if
171 /// necessary there's probably lots of room for improvement performance-wise.
172 pub struct Parser<'a> {
175 cur: iter::Peekable<str::CharIndices<'a>>,
176 /// Error messages accumulated during parsing
177 pub errors: Vec<ParseError>,
178 /// Current position of implicit positional argument pointer
180 /// `Some(raw count)` when the string is "raw", used to position spans correctly
181 style: Option<usize>,
182 /// Start and end byte offset of every successfully parsed argument
183 pub arg_places: Vec<InnerSpan>,
184 /// Characters that need to be shifted
186 /// Span of the last opening brace seen, used for error reporting
187 last_opening_brace: Option<InnerSpan>,
188 /// Whether the source string is comes from `println!` as opposed to `format!` or `print!`
189 append_newline: bool,
190 /// Whether this formatting string is a literal or it comes from a macro.
191 pub is_literal: bool,
192 /// Start position of the current line.
193 cur_line_start: usize,
194 /// Start and end byte offset of every line of the format string. Excludes
195 /// newline characters and leading whitespace.
196 pub line_spans: Vec<InnerSpan>,
199 impl<'a> Iterator for Parser<'a> {
200 type Item = Piece<'a>;
202 fn next(&mut self) -> Option<Piece<'a>> {
203 if let Some(&(pos, c)) = self.cur.peek() {
206 let curr_last_brace = self.last_opening_brace;
207 let byte_pos = self.to_span_index(pos);
208 self.last_opening_brace = Some(byte_pos.to(InnerOffset(byte_pos.0 + 1)));
210 if self.consume('{') {
211 self.last_opening_brace = curr_last_brace;
213 Some(String(self.string(pos + 1)))
215 let arg = self.argument();
216 if let Some(end) = self.must_consume('}') {
217 let start = self.to_span_index(pos);
218 let end = self.to_span_index(end + 1);
220 self.arg_places.push(start.to(end));
223 Some(NextArgument(arg))
228 if self.consume('}') {
229 Some(String(self.string(pos + 1)))
231 let err_pos = self.to_span_index(pos);
233 "unmatched `}` found",
235 "if you intended to print `}`, you can escape it using `}}`",
241 _ => Some(String(self.string(pos))),
245 let start = self.to_span_index(self.cur_line_start);
246 let end = self.to_span_index(self.input.len());
247 let span = start.to(end);
248 if self.line_spans.last() != Some(&span) {
249 self.line_spans.push(span);
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_or(0, |raw| raw + 1);
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 if let Some((_, '>' | '<' | '^')) = self.cur.clone().nth(1) {
534 if self.consume('<') {
535 spec.align = AlignLeft;
536 } else if self.consume('>') {
537 spec.align = AlignRight;
538 } else if self.consume('^') {
539 spec.align = AlignCenter;
542 if self.consume('+') {
543 spec.flags |= 1 << (FlagSignPlus as u32);
544 } else if self.consume('-') {
545 spec.flags |= 1 << (FlagSignMinus as u32);
548 if self.consume('#') {
549 spec.flags |= 1 << (FlagAlternate as u32);
551 // Width and precision
552 let mut havewidth = false;
554 if self.consume('0') {
555 // small ambiguity with '0$' as a format string. In theory this is a
556 // '0' flag and then an ill-formatted format string with just a '$'
557 // and no count, but this is better if we instead interpret this as
558 // no '0' flag and '0$' as the width instead.
559 if self.consume('$') {
560 spec.width = CountIsParam(0);
563 spec.flags |= 1 << (FlagSignAwareZeroPad as u32);
567 let width_span_start = if let Some((pos, _)) = self.cur.peek() { *pos } else { 0 };
568 let (w, sp) = self.count(width_span_start);
570 spec.width_span = sp;
572 if let Some(start) = self.consume_pos('.') {
573 if let Some(end) = self.consume_pos('*') {
574 // Resolve `CountIsNextParam`.
575 // We can do this immediately as `position` is resolved later.
578 spec.precision = CountIsParam(i);
579 spec.precision_span =
580 Some(self.to_span_index(start).to(self.to_span_index(end + 1)));
582 let (p, sp) = self.count(start);
584 spec.precision_span = sp;
587 let ty_span_start = self.cur.peek().map(|(pos, _)| *pos);
588 // Optional radix followed by the actual format specifier
589 if self.consume('x') {
590 if self.consume('?') {
591 spec.flags |= 1 << (FlagDebugLowerHex as u32);
596 } else if self.consume('X') {
597 if self.consume('?') {
598 spec.flags |= 1 << (FlagDebugUpperHex as u32);
603 } else if self.consume('?') {
606 spec.ty = self.word();
607 let ty_span_end = self.cur.peek().map(|(pos, _)| *pos);
608 if !spec.ty.is_empty() {
609 spec.ty_span = ty_span_start
610 .and_then(|s| ty_span_end.map(|e| (s, e)))
611 .map(|(start, end)| self.to_span_index(start).to(self.to_span_index(end)));
617 /// Parses an inline assembly template modifier at the current position, returning the modifier
618 /// in the `ty` field of the `FormatSpec` struct.
619 fn inline_asm(&mut self) -> FormatSpec<'a> {
620 let mut spec = FormatSpec {
624 precision: CountImplied,
625 precision_span: None,
628 ty: &self.input[..0],
631 if !self.consume(':') {
635 let ty_span_start = self.cur.peek().map(|(pos, _)| *pos);
636 spec.ty = self.word();
637 let ty_span_end = self.cur.peek().map(|(pos, _)| *pos);
638 if !spec.ty.is_empty() {
639 spec.ty_span = ty_span_start
640 .and_then(|s| ty_span_end.map(|e| (s, e)))
641 .map(|(start, end)| self.to_span_index(start).to(self.to_span_index(end)));
647 /// Parses a `Count` parameter at the current position. This does not check
648 /// for 'CountIsNextParam' because that is only used in precision, not
650 fn count(&mut self, start: usize) -> (Count, Option<InnerSpan>) {
651 if let Some(i) = self.integer() {
652 if let Some(end) = self.consume_pos('$') {
653 let span = self.to_span_index(start).to(self.to_span_index(end + 1));
654 (CountIsParam(i), Some(span))
659 let tmp = self.cur.clone();
660 let word = self.word();
664 } else if self.consume('$') {
665 (CountIsName(Symbol::intern(word)), None)
673 /// Parses a word starting at the current position. A word is the same as
674 /// Rust identifier, except that it can't start with `_` character.
675 fn word(&mut self) -> &'a str {
676 let start = match self.cur.peek() {
677 Some(&(pos, c)) if rustc_lexer::is_id_start(c) => {
686 while let Some(&(pos, c)) = self.cur.peek() {
687 if rustc_lexer::is_id_continue(c) {
694 let end = end.unwrap_or(self.input.len());
695 let word = &self.input[start..end];
698 "invalid argument name `_`",
699 "invalid argument name",
700 "argument name cannot be a single underscore",
701 self.to_span_index(start).to(self.to_span_index(end)),
707 /// Optionally parses an integer at the current position. This doesn't deal
708 /// with overflow at all, it's just accumulating digits.
709 fn integer(&mut self) -> Option<usize> {
711 let mut found = false;
712 while let Some(&(_, c)) = self.cur.peek() {
713 if let Some(i) = c.to_digit(10) {
714 cur = cur * 10 + i as usize;
725 /// Finds the indices of all characters that have been processed and differ between the actual
726 /// written code (code snippet) and the `InternedString` that gets processed in the `Parser`
727 /// in order to properly synthethise the intra-string `Span`s for error diagnostics.
728 fn find_skips_from_snippet(
729 snippet: Option<string::String>,
730 str_style: Option<usize>,
731 ) -> (Vec<usize>, bool) {
732 let snippet = match snippet {
733 Some(ref s) if s.starts_with('"') || s.starts_with("r\"") || s.starts_with("r#") => s,
734 _ => return (vec![], false),
737 fn find_skips(snippet: &str, is_raw: bool) -> Vec<usize> {
738 let mut eat_ws = false;
739 let mut s = snippet.char_indices().peekable();
740 let mut skips = vec![];
741 while let Some((pos, c)) = s.next() {
742 match (c, s.peek()) {
743 // skip whitespace and empty lines ending in '\\'
744 ('\\', Some((next_pos, '\n'))) if !is_raw => {
747 skips.push(*next_pos);
750 ('\\', Some((next_pos, '\n' | 'n' | 't'))) if eat_ws => {
752 skips.push(*next_pos);
755 (' ' | '\n' | '\t', _) if eat_ws => {
758 ('\\', Some((next_pos, 'n' | 't' | 'r' | '0' | '\\' | '\'' | '\"'))) => {
759 skips.push(*next_pos);
762 ('\\', Some((_, 'x'))) if !is_raw => {
764 // consume `\xAB` literal
765 if let Some((pos, _)) = s.next() {
772 ('\\', Some((_, 'u'))) if !is_raw => {
773 if let Some((pos, _)) = s.next() {
776 if let Some((next_pos, next_c)) = s.next() {
778 skips.push(next_pos);
779 let mut i = 0; // consume up to 6 hexanumeric chars + closing `}`
780 while let (Some((next_pos, c)), true) = (s.next(), i < 7) {
782 skips.push(next_pos);
784 skips.push(next_pos);
791 } else if next_c.is_digit(16) {
792 skips.push(next_pos);
793 // We suggest adding `{` and `}` when appropriate, accept it here as if
795 let mut i = 0; // consume up to 6 hexanumeric chars
796 while let (Some((next_pos, c)), _) = (s.next(), i < 6) {
798 skips.push(next_pos);
808 // `take_while(|c| c.is_whitespace())`
817 let r_start = str_style.map_or(0, |r| r + 1);
818 let r_end = str_style.unwrap_or(0);
819 let s = &snippet[r_start + 1..snippet.len() - r_end - 1];
820 (find_skips(s, str_style.is_some()), true)