]> git.proxmox.com Git - rustc.git/blob - src/libfmt_macros/lib.rs
New upstream version 1.42.0+dfsg1
[rustc.git] / src / libfmt_macros / lib.rs
1 //! Macro support for format strings
2 //!
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
5 //! generated instead.
6
7 #![doc(
8 html_root_url = "https://doc.rust-lang.org/nightly/",
9 html_playground_url = "https://play.rust-lang.org/",
10 test(attr(deny(warnings)))
11 )]
12 #![feature(nll)]
13 #![feature(rustc_private)]
14 #![feature(unicode_internals)]
15 #![feature(bool_to_option)]
16
17 pub use Alignment::*;
18 pub use Count::*;
19 pub use Flag::*;
20 pub use Piece::*;
21 pub use Position::*;
22
23 use std::iter;
24 use std::str;
25 use std::string;
26
27 use rustc_span::{InnerSpan, Symbol};
28
29 #[derive(Copy, Clone)]
30 struct InnerOffset(usize);
31
32 impl InnerOffset {
33 fn to(self, end: InnerOffset) -> InnerSpan {
34 InnerSpan::new(self.0, end.0)
35 }
36 }
37
38 /// A piece is a portion of the format string which represents the next part
39 /// to emit. These are emitted as a stream by the `Parser` class.
40 #[derive(Copy, Clone, Debug, PartialEq)]
41 pub enum Piece<'a> {
42 /// A literal string which should directly be emitted
43 String(&'a str),
44 /// This describes that formatting should process the next argument (as
45 /// specified inside) for emission.
46 NextArgument(Argument<'a>),
47 }
48
49 /// Representation of an argument specification.
50 #[derive(Copy, Clone, Debug, PartialEq)]
51 pub struct Argument<'a> {
52 /// Where to find this argument
53 pub position: Position,
54 /// How to format the argument
55 pub format: FormatSpec<'a>,
56 }
57
58 /// Specification for the formatting of an argument in the format string.
59 #[derive(Copy, Clone, Debug, PartialEq)]
60 pub struct FormatSpec<'a> {
61 /// Optionally specified character to fill alignment with.
62 pub fill: Option<char>,
63 /// Optionally specified alignment.
64 pub align: Alignment,
65 /// Packed version of various flags provided.
66 pub flags: u32,
67 /// The integer precision to use.
68 pub precision: Count,
69 /// The span of the precision formatting flag (for diagnostics).
70 pub precision_span: Option<InnerSpan>,
71 /// The string width requested for the resulting format.
72 pub width: Count,
73 /// The span of the width formatting flag (for diagnostics).
74 pub width_span: Option<InnerSpan>,
75 /// The descriptor string representing the name of the format desired for
76 /// this argument, this can be empty or any number of characters, although
77 /// it is required to be one word.
78 pub ty: &'a str,
79 /// The span of the descriptor string (for diagnostics).
80 pub ty_span: Option<InnerSpan>,
81 }
82
83 /// Enum describing where an argument for a format can be located.
84 #[derive(Copy, Clone, Debug, PartialEq)]
85 pub enum Position {
86 /// The argument is implied to be located at an index
87 ArgumentImplicitlyIs(usize),
88 /// The argument is located at a specific index given in the format
89 ArgumentIs(usize),
90 /// The argument has a name.
91 ArgumentNamed(Symbol),
92 }
93
94 impl Position {
95 pub fn index(&self) -> Option<usize> {
96 match self {
97 ArgumentIs(i) | ArgumentImplicitlyIs(i) => Some(*i),
98 _ => None,
99 }
100 }
101 }
102
103 /// Enum of alignments which are supported.
104 #[derive(Copy, Clone, Debug, PartialEq)]
105 pub enum Alignment {
106 /// The value will be aligned to the left.
107 AlignLeft,
108 /// The value will be aligned to the right.
109 AlignRight,
110 /// The value will be aligned in the center.
111 AlignCenter,
112 /// The value will take on a default alignment.
113 AlignUnknown,
114 }
115
116 /// Various flags which can be applied to format strings. The meaning of these
117 /// flags is defined by the formatters themselves.
118 #[derive(Copy, Clone, Debug, PartialEq)]
119 pub enum Flag {
120 /// A `+` will be used to denote positive numbers.
121 FlagSignPlus,
122 /// A `-` will be used to denote negative numbers. This is the default.
123 FlagSignMinus,
124 /// An alternate form will be used for the value. In the case of numbers,
125 /// this means that the number will be prefixed with the supplied string.
126 FlagAlternate,
127 /// For numbers, this means that the number will be padded with zeroes,
128 /// and the sign (`+` or `-`) will precede them.
129 FlagSignAwareZeroPad,
130 /// For Debug / `?`, format integers in lower-case hexadecimal.
131 FlagDebugLowerHex,
132 /// For Debug / `?`, format integers in upper-case hexadecimal.
133 FlagDebugUpperHex,
134 }
135
136 /// A count is used for the precision and width parameters of an integer, and
137 /// can reference either an argument or a literal integer.
138 #[derive(Copy, Clone, Debug, PartialEq)]
139 pub enum Count {
140 /// The count is specified explicitly.
141 CountIs(usize),
142 /// The count is specified by the argument with the given name.
143 CountIsName(Symbol),
144 /// The count is specified by the argument at the given index.
145 CountIsParam(usize),
146 /// The count is implied and cannot be explicitly specified.
147 CountImplied,
148 }
149
150 pub struct ParseError {
151 pub description: string::String,
152 pub note: Option<string::String>,
153 pub label: string::String,
154 pub span: InnerSpan,
155 pub secondary_label: Option<(string::String, InnerSpan)>,
156 }
157
158 /// The parser structure for interpreting the input format string. This is
159 /// modeled as an iterator over `Piece` structures to form a stream of tokens
160 /// being output.
161 ///
162 /// This is a recursive-descent parser for the sake of simplicity, and if
163 /// necessary there's probably lots of room for improvement performance-wise.
164 pub struct Parser<'a> {
165 input: &'a str,
166 cur: iter::Peekable<str::CharIndices<'a>>,
167 /// Error messages accumulated during parsing
168 pub errors: Vec<ParseError>,
169 /// Current position of implicit positional argument pointer
170 curarg: usize,
171 /// `Some(raw count)` when the string is "raw", used to position spans correctly
172 style: Option<usize>,
173 /// Start and end byte offset of every successfully parsed argument
174 pub arg_places: Vec<InnerSpan>,
175 /// Characters that need to be shifted
176 skips: Vec<usize>,
177 /// Span of the last opening brace seen, used for error reporting
178 last_opening_brace: Option<InnerSpan>,
179 /// Wether the source string is comes from `println!` as opposed to `format!` or `print!`
180 append_newline: bool,
181 }
182
183 impl<'a> Iterator for Parser<'a> {
184 type Item = Piece<'a>;
185
186 fn next(&mut self) -> Option<Piece<'a>> {
187 if let Some(&(pos, c)) = self.cur.peek() {
188 match c {
189 '{' => {
190 let curr_last_brace = self.last_opening_brace;
191 let byte_pos = self.to_span_index(pos);
192 self.last_opening_brace = Some(byte_pos.to(InnerOffset(byte_pos.0 + 1)));
193 self.cur.next();
194 if self.consume('{') {
195 self.last_opening_brace = curr_last_brace;
196
197 Some(String(self.string(pos + 1)))
198 } else {
199 let arg = self.argument();
200 if let Some(end) = self.must_consume('}') {
201 let start = self.to_span_index(pos);
202 let end = self.to_span_index(end + 1);
203 self.arg_places.push(start.to(end));
204 }
205 Some(NextArgument(arg))
206 }
207 }
208 '}' => {
209 self.cur.next();
210 if self.consume('}') {
211 Some(String(self.string(pos + 1)))
212 } else {
213 let err_pos = self.to_span_index(pos);
214 self.err_with_note(
215 "unmatched `}` found",
216 "unmatched `}`",
217 "if you intended to print `}`, you can escape it using `}}`",
218 err_pos.to(err_pos),
219 );
220 None
221 }
222 }
223 '\n' => Some(String(self.string(pos))),
224 _ => Some(String(self.string(pos))),
225 }
226 } else {
227 None
228 }
229 }
230 }
231
232 impl<'a> Parser<'a> {
233 /// Creates a new parser for the given format string
234 pub fn new(
235 s: &'a str,
236 style: Option<usize>,
237 skips: Vec<usize>,
238 append_newline: bool,
239 ) -> Parser<'a> {
240 Parser {
241 input: s,
242 cur: s.char_indices().peekable(),
243 errors: vec![],
244 curarg: 0,
245 style,
246 arg_places: vec![],
247 skips,
248 last_opening_brace: None,
249 append_newline,
250 }
251 }
252
253 /// Notifies of an error. The message doesn't actually need to be of type
254 /// String, but I think it does when this eventually uses conditions so it
255 /// might as well start using it now.
256 fn err<S1: Into<string::String>, S2: Into<string::String>>(
257 &mut self,
258 description: S1,
259 label: S2,
260 span: InnerSpan,
261 ) {
262 self.errors.push(ParseError {
263 description: description.into(),
264 note: None,
265 label: label.into(),
266 span,
267 secondary_label: None,
268 });
269 }
270
271 /// Notifies of an error. The message doesn't actually need to be of type
272 /// String, but I think it does when this eventually uses conditions so it
273 /// might as well start using it now.
274 fn err_with_note<
275 S1: Into<string::String>,
276 S2: Into<string::String>,
277 S3: Into<string::String>,
278 >(
279 &mut self,
280 description: S1,
281 label: S2,
282 note: S3,
283 span: InnerSpan,
284 ) {
285 self.errors.push(ParseError {
286 description: description.into(),
287 note: Some(note.into()),
288 label: label.into(),
289 span,
290 secondary_label: None,
291 });
292 }
293
294 /// Optionally consumes the specified character. If the character is not at
295 /// the current position, then the current iterator isn't moved and `false` is
296 /// returned, otherwise the character is consumed and `true` is returned.
297 fn consume(&mut self, c: char) -> bool {
298 self.consume_pos(c).is_some()
299 }
300
301 /// Optionally consumes the specified character. If the character is not at
302 /// the current position, then the current iterator isn't moved and `None` is
303 /// returned, otherwise the character is consumed and the current position is
304 /// returned.
305 fn consume_pos(&mut self, c: char) -> Option<usize> {
306 if let Some(&(pos, maybe)) = self.cur.peek() {
307 if c == maybe {
308 self.cur.next();
309 return Some(pos);
310 }
311 }
312 None
313 }
314
315 fn to_span_index(&self, pos: usize) -> InnerOffset {
316 let mut pos = pos;
317 // This handles the raw string case, the raw argument is the number of #
318 // in r###"..."### (we need to add one because of the `r`).
319 let raw = self.style.map(|raw| raw + 1).unwrap_or(0);
320 for skip in &self.skips {
321 if pos > *skip {
322 pos += 1;
323 } else if pos == *skip && raw == 0 {
324 pos += 1;
325 } else {
326 break;
327 }
328 }
329 InnerOffset(raw + pos + 1)
330 }
331
332 /// Forces consumption of the specified character. If the character is not
333 /// found, an error is emitted.
334 fn must_consume(&mut self, c: char) -> Option<usize> {
335 self.ws();
336
337 if let Some(&(pos, maybe)) = self.cur.peek() {
338 if c == maybe {
339 self.cur.next();
340 Some(pos)
341 } else {
342 let pos = self.to_span_index(pos);
343 let description = format!("expected `'}}'`, found `{:?}`", maybe);
344 let label = "expected `}`".to_owned();
345 let (note, secondary_label) = if c == '}' {
346 (
347 Some(
348 "if you intended to print `{`, you can escape it using `{{`".to_owned(),
349 ),
350 self.last_opening_brace
351 .map(|sp| ("because of this opening brace".to_owned(), sp)),
352 )
353 } else {
354 (None, None)
355 };
356 self.errors.push(ParseError {
357 description,
358 note,
359 label,
360 span: pos.to(pos),
361 secondary_label,
362 });
363 None
364 }
365 } else {
366 let description = format!("expected `{:?}` but string was terminated", c);
367 // point at closing `"`
368 let pos = self.input.len() - if self.append_newline { 1 } else { 0 };
369 let pos = self.to_span_index(pos);
370 if c == '}' {
371 let label = format!("expected `{:?}`", c);
372 let (note, secondary_label) = if c == '}' {
373 (
374 Some(
375 "if you intended to print `{`, you can escape it using `{{`".to_owned(),
376 ),
377 self.last_opening_brace
378 .map(|sp| ("because of this opening brace".to_owned(), sp)),
379 )
380 } else {
381 (None, None)
382 };
383 self.errors.push(ParseError {
384 description,
385 note,
386 label,
387 span: pos.to(pos),
388 secondary_label,
389 });
390 } else {
391 self.err(description, format!("expected `{:?}`", c), pos.to(pos));
392 }
393 None
394 }
395 }
396
397 /// Consumes all whitespace characters until the first non-whitespace character
398 fn ws(&mut self) {
399 while let Some(&(_, c)) = self.cur.peek() {
400 if c.is_whitespace() {
401 self.cur.next();
402 } else {
403 break;
404 }
405 }
406 }
407
408 /// Parses all of a string which is to be considered a "raw literal" in a
409 /// format string. This is everything outside of the braces.
410 fn string(&mut self, start: usize) -> &'a str {
411 // we may not consume the character, peek the iterator
412 while let Some(&(pos, c)) = self.cur.peek() {
413 match c {
414 '{' | '}' => {
415 return &self.input[start..pos];
416 }
417 _ => {
418 self.cur.next();
419 }
420 }
421 }
422 &self.input[start..self.input.len()]
423 }
424
425 /// Parses an `Argument` structure, or what's contained within braces inside the format string.
426 fn argument(&mut self) -> Argument<'a> {
427 let pos = self.position();
428 let format = self.format();
429
430 // Resolve position after parsing format spec.
431 let pos = match pos {
432 Some(position) => position,
433 None => {
434 let i = self.curarg;
435 self.curarg += 1;
436 ArgumentImplicitlyIs(i)
437 }
438 };
439
440 Argument { position: pos, format }
441 }
442
443 /// Parses a positional argument for a format. This could either be an
444 /// integer index of an argument, a named argument, or a blank string.
445 /// Returns `Some(parsed_position)` if the position is not implicitly
446 /// consuming a macro argument, `None` if it's the case.
447 fn position(&mut self) -> Option<Position> {
448 if let Some(i) = self.integer() {
449 Some(ArgumentIs(i))
450 } else {
451 match self.cur.peek() {
452 Some(&(_, c)) if rustc_lexer::is_id_start(c) => {
453 Some(ArgumentNamed(Symbol::intern(self.word())))
454 }
455
456 // This is an `ArgumentNext`.
457 // Record the fact and do the resolution after parsing the
458 // format spec, to make things like `{:.*}` work.
459 _ => None,
460 }
461 }
462 }
463
464 /// Parses a format specifier at the current position, returning all of the
465 /// relevant information in the `FormatSpec` struct.
466 fn format(&mut self) -> FormatSpec<'a> {
467 let mut spec = FormatSpec {
468 fill: None,
469 align: AlignUnknown,
470 flags: 0,
471 precision: CountImplied,
472 precision_span: None,
473 width: CountImplied,
474 width_span: None,
475 ty: &self.input[..0],
476 ty_span: None,
477 };
478 if !self.consume(':') {
479 return spec;
480 }
481
482 // fill character
483 if let Some(&(_, c)) = self.cur.peek() {
484 match self.cur.clone().nth(1) {
485 Some((_, '>')) | Some((_, '<')) | Some((_, '^')) => {
486 spec.fill = Some(c);
487 self.cur.next();
488 }
489 _ => {}
490 }
491 }
492 // Alignment
493 if self.consume('<') {
494 spec.align = AlignLeft;
495 } else if self.consume('>') {
496 spec.align = AlignRight;
497 } else if self.consume('^') {
498 spec.align = AlignCenter;
499 }
500 // Sign flags
501 if self.consume('+') {
502 spec.flags |= 1 << (FlagSignPlus as u32);
503 } else if self.consume('-') {
504 spec.flags |= 1 << (FlagSignMinus as u32);
505 }
506 // Alternate marker
507 if self.consume('#') {
508 spec.flags |= 1 << (FlagAlternate as u32);
509 }
510 // Width and precision
511 let mut havewidth = false;
512
513 if self.consume('0') {
514 // small ambiguity with '0$' as a format string. In theory this is a
515 // '0' flag and then an ill-formatted format string with just a '$'
516 // and no count, but this is better if we instead interpret this as
517 // no '0' flag and '0$' as the width instead.
518 if self.consume('$') {
519 spec.width = CountIsParam(0);
520 havewidth = true;
521 } else {
522 spec.flags |= 1 << (FlagSignAwareZeroPad as u32);
523 }
524 }
525 if !havewidth {
526 let width_span_start = if let Some((pos, _)) = self.cur.peek() { *pos } else { 0 };
527 let (w, sp) = self.count(width_span_start);
528 spec.width = w;
529 spec.width_span = sp;
530 }
531 if let Some(start) = self.consume_pos('.') {
532 if let Some(end) = self.consume_pos('*') {
533 // Resolve `CountIsNextParam`.
534 // We can do this immediately as `position` is resolved later.
535 let i = self.curarg;
536 self.curarg += 1;
537 spec.precision = CountIsParam(i);
538 spec.precision_span =
539 Some(self.to_span_index(start).to(self.to_span_index(end + 1)));
540 } else {
541 let (p, sp) = self.count(start);
542 spec.precision = p;
543 spec.precision_span = sp;
544 }
545 }
546 let ty_span_start = self.cur.peek().map(|(pos, _)| *pos);
547 // Optional radix followed by the actual format specifier
548 if self.consume('x') {
549 if self.consume('?') {
550 spec.flags |= 1 << (FlagDebugLowerHex as u32);
551 spec.ty = "?";
552 } else {
553 spec.ty = "x";
554 }
555 } else if self.consume('X') {
556 if self.consume('?') {
557 spec.flags |= 1 << (FlagDebugUpperHex as u32);
558 spec.ty = "?";
559 } else {
560 spec.ty = "X";
561 }
562 } else if self.consume('?') {
563 spec.ty = "?";
564 } else {
565 spec.ty = self.word();
566 let ty_span_end = self.cur.peek().map(|(pos, _)| *pos);
567 if !spec.ty.is_empty() {
568 spec.ty_span = ty_span_start
569 .and_then(|s| ty_span_end.map(|e| (s, e)))
570 .map(|(start, end)| self.to_span_index(start).to(self.to_span_index(end)));
571 }
572 }
573 spec
574 }
575
576 /// Parses a `Count` parameter at the current position. This does not check
577 /// for 'CountIsNextParam' because that is only used in precision, not
578 /// width.
579 fn count(&mut self, start: usize) -> (Count, Option<InnerSpan>) {
580 if let Some(i) = self.integer() {
581 if let Some(end) = self.consume_pos('$') {
582 let span = self.to_span_index(start).to(self.to_span_index(end + 1));
583 (CountIsParam(i), Some(span))
584 } else {
585 (CountIs(i), None)
586 }
587 } else {
588 let tmp = self.cur.clone();
589 let word = self.word();
590 if word.is_empty() {
591 self.cur = tmp;
592 (CountImplied, None)
593 } else if self.consume('$') {
594 (CountIsName(Symbol::intern(word)), None)
595 } else {
596 self.cur = tmp;
597 (CountImplied, None)
598 }
599 }
600 }
601
602 /// Parses a word starting at the current position. A word is the same as
603 /// Rust identifier, except that it can't start with `_` character.
604 fn word(&mut self) -> &'a str {
605 let start = match self.cur.peek() {
606 Some(&(pos, c)) if rustc_lexer::is_id_start(c) => {
607 self.cur.next();
608 pos
609 }
610 _ => {
611 return "";
612 }
613 };
614 let mut end = None;
615 while let Some(&(pos, c)) = self.cur.peek() {
616 if rustc_lexer::is_id_continue(c) {
617 self.cur.next();
618 } else {
619 end = Some(pos);
620 break;
621 }
622 }
623 let end = end.unwrap_or(self.input.len());
624 let word = &self.input[start..end];
625 if word == "_" {
626 self.err_with_note(
627 "invalid argument name `_`",
628 "invalid argument name",
629 "argument name cannot be a single underscore",
630 self.to_span_index(start).to(self.to_span_index(end)),
631 );
632 }
633 word
634 }
635
636 /// Optionally parses an integer at the current position. This doesn't deal
637 /// with overflow at all, it's just accumulating digits.
638 fn integer(&mut self) -> Option<usize> {
639 let mut cur = 0;
640 let mut found = false;
641 while let Some(&(_, c)) = self.cur.peek() {
642 if let Some(i) = c.to_digit(10) {
643 cur = cur * 10 + i as usize;
644 found = true;
645 self.cur.next();
646 } else {
647 break;
648 }
649 }
650 found.then_some(cur)
651 }
652 }
653
654 #[cfg(test)]
655 mod tests;