]> git.lizzy.rs Git - rust.git/blob - src/comment.rs
0f2409be1c6b0331387a6536cc05c998d1a96717
[rust.git] / src / comment.rs
1 // Copyright 2015 The Rust Project Developers. See the COPYRIGHT
2 // file at the top-level directory of this distribution and at
3 // http://rust-lang.org/COPYRIGHT.
4 //
5 // Licensed under the Apache License, Version 2.0 <LICENSE-APACHE or
6 // http://www.apache.org/licenses/LICENSE-2.0> or the MIT license
7 // <LICENSE-MIT or http://opensource.org/licenses/MIT>, at your
8 // option. This file may not be copied, modified, or distributed
9 // except according to those terms.
10
11 // Formatting and tools for comments.
12
13 use std::{self, borrow::Cow, iter};
14
15 use itertools::{multipeek, MultiPeek};
16 use syntax::source_map::Span;
17
18 use config::Config;
19 use rewrite::RewriteContext;
20 use shape::{Indent, Shape};
21 use string::{rewrite_string, StringFormat};
22 use utils::{count_newlines, first_line_width, last_line_width};
23 use {ErrorKind, FormattingError};
24
25 fn is_custom_comment(comment: &str) -> bool {
26     if !comment.starts_with("//") {
27         false
28     } else if let Some(c) = comment.chars().nth(2) {
29         !c.is_alphanumeric() && !c.is_whitespace()
30     } else {
31         false
32     }
33 }
34
35 #[derive(Copy, Clone, PartialEq, Eq)]
36 pub enum CommentStyle<'a> {
37     DoubleSlash,
38     TripleSlash,
39     Doc,
40     SingleBullet,
41     DoubleBullet,
42     Exclamation,
43     Custom(&'a str),
44 }
45
46 fn custom_opener(s: &str) -> &str {
47     s.lines().next().map_or("", |first_line| {
48         first_line
49             .find(' ')
50             .map_or(first_line, |space_index| &first_line[0..=space_index])
51     })
52 }
53
54 impl<'a> CommentStyle<'a> {
55     /// Returns true if the commenting style covers a line only.
56     pub fn is_line_comment(&self) -> bool {
57         match *self {
58             CommentStyle::DoubleSlash
59             | CommentStyle::TripleSlash
60             | CommentStyle::Doc
61             | CommentStyle::Custom(_) => true,
62             _ => false,
63         }
64     }
65
66     /// Returns true if the commenting style can span over multiple lines.
67     pub fn is_block_comment(&self) -> bool {
68         match *self {
69             CommentStyle::SingleBullet | CommentStyle::DoubleBullet | CommentStyle::Exclamation => {
70                 true
71             }
72             _ => false,
73         }
74     }
75
76     /// Returns true if the commenting style is for documentation.
77     pub fn is_doc_comment(&self) -> bool {
78         match *self {
79             CommentStyle::TripleSlash | CommentStyle::Doc => true,
80             _ => false,
81         }
82     }
83
84     pub fn opener(&self) -> &'a str {
85         match *self {
86             CommentStyle::DoubleSlash => "// ",
87             CommentStyle::TripleSlash => "/// ",
88             CommentStyle::Doc => "//! ",
89             CommentStyle::SingleBullet => "/* ",
90             CommentStyle::DoubleBullet => "/** ",
91             CommentStyle::Exclamation => "/*! ",
92             CommentStyle::Custom(opener) => opener,
93         }
94     }
95
96     pub fn closer(&self) -> &'a str {
97         match *self {
98             CommentStyle::DoubleSlash
99             | CommentStyle::TripleSlash
100             | CommentStyle::Custom(..)
101             | CommentStyle::Doc => "",
102             CommentStyle::DoubleBullet => " **/",
103             CommentStyle::SingleBullet | CommentStyle::Exclamation => " */",
104         }
105     }
106
107     pub fn line_start(&self) -> &'a str {
108         match *self {
109             CommentStyle::DoubleSlash => "// ",
110             CommentStyle::TripleSlash => "/// ",
111             CommentStyle::Doc => "//! ",
112             CommentStyle::SingleBullet | CommentStyle::Exclamation => " * ",
113             CommentStyle::DoubleBullet => " ** ",
114             CommentStyle::Custom(opener) => opener,
115         }
116     }
117
118     pub fn to_str_tuplet(&self) -> (&'a str, &'a str, &'a str) {
119         (self.opener(), self.closer(), self.line_start())
120     }
121 }
122
123 fn comment_style(orig: &str, normalize_comments: bool) -> CommentStyle {
124     if !normalize_comments {
125         if orig.starts_with("/**") && !orig.starts_with("/**/") {
126             CommentStyle::DoubleBullet
127         } else if orig.starts_with("/*!") {
128             CommentStyle::Exclamation
129         } else if orig.starts_with("/*") {
130             CommentStyle::SingleBullet
131         } else if orig.starts_with("///") && orig.chars().nth(3).map_or(true, |c| c != '/') {
132             CommentStyle::TripleSlash
133         } else if orig.starts_with("//!") {
134             CommentStyle::Doc
135         } else if is_custom_comment(orig) {
136             CommentStyle::Custom(custom_opener(orig))
137         } else {
138             CommentStyle::DoubleSlash
139         }
140     } else if (orig.starts_with("///") && orig.chars().nth(3).map_or(true, |c| c != '/'))
141         || (orig.starts_with("/**") && !orig.starts_with("/**/"))
142     {
143         CommentStyle::TripleSlash
144     } else if orig.starts_with("//!") || orig.starts_with("/*!") {
145         CommentStyle::Doc
146     } else if is_custom_comment(orig) {
147         CommentStyle::Custom(custom_opener(orig))
148     } else {
149         CommentStyle::DoubleSlash
150     }
151 }
152
153 /// Combine `prev_str` and `next_str` into a single `String`. `span` may contain
154 /// comments between two strings. If there are such comments, then that will be
155 /// recovered. If `allow_extend` is true and there is no comment between the two
156 /// strings, then they will be put on a single line as long as doing so does not
157 /// exceed max width.
158 pub fn combine_strs_with_missing_comments(
159     context: &RewriteContext,
160     prev_str: &str,
161     next_str: &str,
162     span: Span,
163     shape: Shape,
164     allow_extend: bool,
165 ) -> Option<String> {
166     let mut result =
167         String::with_capacity(prev_str.len() + next_str.len() + shape.indent.width() + 128);
168     result.push_str(prev_str);
169     let mut allow_one_line = !prev_str.contains('\n') && !next_str.contains('\n');
170     let first_sep = if prev_str.is_empty() || next_str.is_empty() {
171         ""
172     } else {
173         " "
174     };
175     let mut one_line_width =
176         last_line_width(prev_str) + first_line_width(next_str) + first_sep.len();
177
178     let config = context.config;
179     let indent = shape.indent;
180     let missing_comment = rewrite_missing_comment(span, shape, context)?;
181
182     if missing_comment.is_empty() {
183         if allow_extend && prev_str.len() + first_sep.len() + next_str.len() <= shape.width {
184             result.push_str(first_sep);
185         } else if !prev_str.is_empty() {
186             result.push_str(&indent.to_string_with_newline(config))
187         }
188         result.push_str(next_str);
189         return Some(result);
190     }
191
192     // We have a missing comment between the first expression and the second expression.
193
194     // Peek the the original source code and find out whether there is a newline between the first
195     // expression and the second expression or the missing comment. We will preserve the original
196     // layout whenever possible.
197     let original_snippet = context.snippet(span);
198     let prefer_same_line = if let Some(pos) = original_snippet.find('/') {
199         !original_snippet[..pos].contains('\n')
200     } else {
201         !original_snippet.contains('\n')
202     };
203
204     one_line_width -= first_sep.len();
205     let first_sep = if prev_str.is_empty() || missing_comment.is_empty() {
206         Cow::from("")
207     } else {
208         let one_line_width = last_line_width(prev_str) + first_line_width(&missing_comment) + 1;
209         if prefer_same_line && one_line_width <= shape.width {
210             Cow::from(" ")
211         } else {
212             indent.to_string_with_newline(config)
213         }
214     };
215     result.push_str(&first_sep);
216     result.push_str(&missing_comment);
217
218     let second_sep = if missing_comment.is_empty() || next_str.is_empty() {
219         Cow::from("")
220     } else if missing_comment.starts_with("//") {
221         indent.to_string_with_newline(config)
222     } else {
223         one_line_width += missing_comment.len() + first_sep.len() + 1;
224         allow_one_line &= !missing_comment.starts_with("//") && !missing_comment.contains('\n');
225         if prefer_same_line && allow_one_line && one_line_width <= shape.width {
226             Cow::from(" ")
227         } else {
228             indent.to_string_with_newline(config)
229         }
230     };
231     result.push_str(&second_sep);
232     result.push_str(next_str);
233
234     Some(result)
235 }
236
237 pub fn rewrite_doc_comment(orig: &str, shape: Shape, config: &Config) -> Option<String> {
238     identify_comment(orig, false, shape, config, true)
239 }
240
241 pub fn rewrite_comment(
242     orig: &str,
243     block_style: bool,
244     shape: Shape,
245     config: &Config,
246 ) -> Option<String> {
247     identify_comment(orig, block_style, shape, config, false)
248 }
249
250 fn identify_comment(
251     orig: &str,
252     block_style: bool,
253     shape: Shape,
254     config: &Config,
255     is_doc_comment: bool,
256 ) -> Option<String> {
257     let style = comment_style(orig, false);
258
259     // Computes the len of line taking into account a newline if the line is part of a paragraph.
260     fn compute_len(orig: &str, line: &str) -> usize {
261         if orig.len() > line.len() {
262             if orig.as_bytes()[line.len()] == b'\r' {
263                 line.len() + 2
264             } else {
265                 line.len() + 1
266             }
267         } else {
268             line.len()
269         }
270     }
271
272     // Get the first group of line comments having the same commenting style.
273     //
274     // Returns a tuple with:
275     // - a boolean indicating if there is a blank line
276     // - a number indicating the size of the first group of comments
277     fn consume_same_line_comments(
278         style: CommentStyle,
279         orig: &str,
280         line_start: &str,
281     ) -> (bool, usize) {
282         let mut first_group_ending = 0;
283         let mut hbl = false;
284
285         for line in orig.lines() {
286             let trimmed_line = line.trim_left();
287             if trimmed_line.is_empty() {
288                 hbl = true;
289                 break;
290             } else if trimmed_line.starts_with(line_start)
291                 || comment_style(trimmed_line, false) == style
292             {
293                 first_group_ending += compute_len(&orig[first_group_ending..], line);
294             } else {
295                 break;
296             }
297         }
298         (hbl, first_group_ending)
299     }
300
301     let (has_bare_lines, first_group_ending) = match style {
302         CommentStyle::DoubleSlash | CommentStyle::TripleSlash | CommentStyle::Doc => {
303             let line_start = style.line_start().trim_left();
304             consume_same_line_comments(style, orig, line_start)
305         }
306         CommentStyle::Custom(opener) => {
307             let trimmed_opener = opener.trim_right();
308             consume_same_line_comments(style, orig, trimmed_opener)
309         }
310         // for a block comment, search for the closing symbol
311         CommentStyle::DoubleBullet | CommentStyle::SingleBullet | CommentStyle::Exclamation => {
312             let closer = style.closer().trim_left();
313             let mut closing_symbol_offset = 0;
314             let mut hbl = false;
315             for line in orig.lines() {
316                 closing_symbol_offset += compute_len(&orig[closing_symbol_offset..], line);
317                 let trimmed_line = line.trim_left();
318                 if !trimmed_line.starts_with('*')
319                     && !trimmed_line.starts_with("//")
320                     && !trimmed_line.starts_with("/*")
321                 {
322                     hbl = true;
323                 }
324                 if trimmed_line.ends_with(closer) {
325                     break;
326                 }
327             }
328             (hbl, closing_symbol_offset)
329         }
330     };
331
332     let (first_group, rest) = orig.split_at(first_group_ending);
333     let rewritten_first_group =
334         if !config.normalize_comments() && has_bare_lines && style.is_block_comment() {
335             light_rewrite_block_comment_with_bare_lines(first_group, shape, config)?
336         } else if !config.normalize_comments()
337             && !config.wrap_comments()
338             && !config.format_doc_comments()
339         {
340             light_rewrite_comment(first_group, shape.indent, config, is_doc_comment)?
341         } else {
342             rewrite_comment_inner(
343                 first_group,
344                 block_style,
345                 style,
346                 shape,
347                 config,
348                 is_doc_comment || style.is_doc_comment(),
349             )?
350         };
351     if rest.is_empty() {
352         Some(rewritten_first_group)
353     } else {
354         identify_comment(rest.trim_left(), block_style, shape, config, is_doc_comment).map(
355             |rest_str| {
356                 format!(
357                     "{}\n{}{}{}",
358                     rewritten_first_group,
359                     // insert back the blank line
360                     if has_bare_lines && style.is_line_comment() {
361                         "\n"
362                     } else {
363                         ""
364                     },
365                     shape.indent.to_string(config),
366                     rest_str
367                 )
368             },
369         )
370     }
371 }
372
373 /// Trims a minimum of leading whitespaces so that the content layout is kept and aligns to indent.
374 fn light_rewrite_block_comment_with_bare_lines(
375     orig: &str,
376     shape: Shape,
377     config: &Config,
378 ) -> Option<String> {
379     let prefix_whitespace_min = orig
380         .lines()
381         // skip the line with the starting sigil since the leading whitespace is removed
382         // otherwise, the minimum would always be zero
383         .skip(1)
384         .filter(|line| !line.is_empty())
385         .map(|line| {
386             let mut width = 0;
387             for c in line.chars() {
388                 match c {
389                     ' ' => width += 1,
390                     '\t' => width += config.tab_spaces(),
391                     _ => break,
392                 }
393             }
394             width
395         })
396         .min()?;
397
398     let indent_str = shape.indent.to_string(config);
399     let mut lines = orig.lines();
400     let first_line = lines.next()?;
401     let rest = lines
402         .map(|line| {
403             if line.is_empty() {
404                 line
405             } else {
406                 &line[prefix_whitespace_min..]
407             }
408         })
409         .collect::<Vec<&str>>()
410         .join(&format!("\n{}", indent_str));
411     Some(format!("{}\n{}{}", first_line, indent_str, rest))
412 }
413
414 /// Attributes for code blocks in rustdoc.
415 /// See https://doc.rust-lang.org/rustdoc/print.html#attributes
416 enum CodeBlockAttribute {
417     Rust,
418     Ignore,
419     Text,
420     ShouldPanic,
421     NoRun,
422     CompileFail,
423 }
424
425 impl CodeBlockAttribute {
426     fn new(attribute: &str) -> CodeBlockAttribute {
427         match attribute {
428             "rust" | "" => CodeBlockAttribute::Rust,
429             "ignore" => CodeBlockAttribute::Ignore,
430             "text" => CodeBlockAttribute::Text,
431             "should_panic" => CodeBlockAttribute::ShouldPanic,
432             "no_run" => CodeBlockAttribute::NoRun,
433             "compile_fail" => CodeBlockAttribute::CompileFail,
434             _ => CodeBlockAttribute::Text,
435         }
436     }
437 }
438
439 /// Block that is formatted as an item.
440 ///
441 /// An item starts with either a star `*` or a dash `-`. Different level of indentation are
442 /// handled.
443 struct ItemizedBlock {
444     /// the number of whitespaces up to the item sigil
445     indent: usize,
446     /// the string that marks the start of an item
447     opener: String,
448     /// sequence of whitespaces to prefix new lines that are part of the item
449     line_start: String,
450 }
451
452 impl ItemizedBlock {
453     /// Returns true if the line is formatted as an item
454     fn is_itemized_line(line: &str) -> bool {
455         let trimmed = line.trim_left();
456         trimmed.starts_with("* ") || trimmed.starts_with("- ")
457     }
458
459     /// Creates a new ItemizedBlock described with the given line.
460     /// The `is_itemized_line` needs to be called first.
461     fn new(line: &str) -> ItemizedBlock {
462         let space_to_sigil = line.chars().take_while(|c| c.is_whitespace()).count();
463         let indent = space_to_sigil + 2;
464         ItemizedBlock {
465             indent,
466             opener: line[..indent].to_string(),
467             line_start: " ".repeat(indent),
468         }
469     }
470
471     /// Returns a `StringFormat` used for formatting the content of an item
472     fn create_string_format<'a>(&'a self, fmt: &'a StringFormat) -> StringFormat<'a> {
473         StringFormat {
474             opener: "",
475             closer: "",
476             line_start: "",
477             line_end: "",
478             shape: Shape::legacy(fmt.shape.width.saturating_sub(self.indent), Indent::empty()),
479             trim_end: true,
480             config: fmt.config,
481         }
482     }
483
484     /// Returns true if the line is part of the current itemized block
485     fn in_block(&self, line: &str) -> bool {
486         !ItemizedBlock::is_itemized_line(line)
487             && self.indent <= line.chars().take_while(|c| c.is_whitespace()).count()
488     }
489 }
490
491 fn rewrite_comment_inner(
492     orig: &str,
493     block_style: bool,
494     style: CommentStyle,
495     shape: Shape,
496     config: &Config,
497     is_doc_comment: bool,
498 ) -> Option<String> {
499     let (opener, closer, line_start) = if block_style {
500         CommentStyle::SingleBullet.to_str_tuplet()
501     } else {
502         comment_style(orig, config.normalize_comments()).to_str_tuplet()
503     };
504
505     let max_chars = shape
506         .width
507         .checked_sub(closer.len() + opener.len())
508         .unwrap_or(1);
509     let indent_str = shape.indent.to_string_with_newline(config);
510     let fmt_indent = shape.indent + (opener.len() - line_start.len());
511     let mut fmt = StringFormat {
512         opener: "",
513         closer: "",
514         line_start,
515         line_end: "",
516         shape: Shape::legacy(max_chars, fmt_indent),
517         trim_end: true,
518         config,
519     };
520
521     let line_breaks = count_newlines(orig.trim_right());
522     let lines = orig
523         .lines()
524         .enumerate()
525         .map(|(i, mut line)| {
526             line = trim_right_unless_two_whitespaces(line.trim_left(), is_doc_comment);
527             // Drop old closer.
528             if i == line_breaks && line.ends_with("*/") && !line.starts_with("//") {
529                 line = line[..(line.len() - 2)].trim_right();
530             }
531
532             line
533         })
534         .map(|s| left_trim_comment_line(s, &style))
535         .map(|(line, has_leading_whitespace)| {
536             if orig.starts_with("/*") && line_breaks == 0 {
537                 (
538                     line.trim_left(),
539                     has_leading_whitespace || config.normalize_comments(),
540                 )
541             } else {
542                 (line, has_leading_whitespace || config.normalize_comments())
543             }
544         });
545
546     let mut result = String::with_capacity(orig.len() * 2);
547     result.push_str(opener);
548     let mut code_block_buffer = String::with_capacity(128);
549     let mut is_prev_line_multi_line = false;
550     let mut code_block_attr = None;
551     let mut item_block_buffer = String::with_capacity(128);
552     let mut item_block: Option<ItemizedBlock> = None;
553     let comment_line_separator = format!("{}{}", indent_str, line_start);
554     let join_block = |s: &str, sep: &str| {
555         let mut result = String::with_capacity(s.len() + 128);
556         let mut iter = s.lines().peekable();
557         while let Some(line) = iter.next() {
558             result.push_str(line);
559             result.push_str(match iter.peek() {
560                 Some(next_line) if next_line.is_empty() => sep.trim_right(),
561                 Some(..) => &sep,
562                 None => "",
563             });
564         }
565         result
566     };
567
568     for (i, (line, has_leading_whitespace)) in lines.enumerate() {
569         let is_last = i == count_newlines(orig);
570
571         if let Some(ref ib) = item_block {
572             if ib.in_block(&line) {
573                 item_block_buffer.push_str(&line);
574                 item_block_buffer.push('\n');
575                 continue;
576             }
577             is_prev_line_multi_line = false;
578             fmt.shape = Shape::legacy(max_chars, fmt_indent);
579             let item_fmt = ib.create_string_format(&fmt);
580             result.push_str(&comment_line_separator);
581             result.push_str(&ib.opener);
582             match rewrite_string(
583                 &item_block_buffer.replace("\n", " "),
584                 &item_fmt,
585                 max_chars.saturating_sub(ib.indent),
586             ) {
587                 Some(s) => result.push_str(&join_block(
588                     &s,
589                     &format!("{}{}", &comment_line_separator, ib.line_start),
590                 )),
591                 None => result.push_str(&join_block(&item_block_buffer, &comment_line_separator)),
592             };
593             item_block_buffer.clear();
594         } else if let Some(ref attr) = code_block_attr {
595             if line.starts_with("```") {
596                 let code_block = match attr {
597                     CodeBlockAttribute::Ignore | CodeBlockAttribute::Text => {
598                         trim_custom_comment_prefix(&code_block_buffer)
599                     }
600                     _ if code_block_buffer.is_empty() => String::new(),
601                     _ => {
602                         let mut config = config.clone();
603                         config.set().format_doc_comments(false);
604                         match ::format_code_block(&code_block_buffer, &config) {
605                             Some(ref s) => trim_custom_comment_prefix(s),
606                             None => trim_custom_comment_prefix(&code_block_buffer),
607                         }
608                     }
609                 };
610                 if !code_block.is_empty() {
611                     result.push_str(&comment_line_separator);
612                     result.push_str(&join_block(&code_block, &comment_line_separator));
613                 }
614                 code_block_buffer.clear();
615                 result.push_str(&comment_line_separator);
616                 result.push_str(line);
617                 code_block_attr = None;
618             } else {
619                 code_block_buffer.push_str(&hide_sharp_behind_comment(line));
620                 code_block_buffer.push('\n');
621             }
622             continue;
623         }
624
625         code_block_attr = None;
626         item_block = None;
627         if line.starts_with("```") {
628             code_block_attr = Some(CodeBlockAttribute::new(&line[3..]))
629         } else if config.wrap_comments() && ItemizedBlock::is_itemized_line(&line) {
630             let ib = ItemizedBlock::new(&line);
631             item_block_buffer.push_str(&line[ib.indent..]);
632             item_block_buffer.push('\n');
633             item_block = Some(ib);
634             continue;
635         }
636
637         if result == opener {
638             let force_leading_whitespace = opener == "/* " && count_newlines(orig) == 0;
639             if !has_leading_whitespace && !force_leading_whitespace && result.ends_with(' ') {
640                 result.pop();
641             }
642             if line.is_empty() {
643                 continue;
644             }
645         } else if is_prev_line_multi_line && !line.is_empty() {
646             result.push(' ')
647         } else if is_last && line.is_empty() {
648             // trailing blank lines are unwanted
649             if !closer.is_empty() {
650                 result.push_str(&indent_str);
651             }
652             break;
653         } else {
654             result.push_str(&comment_line_separator);
655             if !has_leading_whitespace && result.ends_with(' ') {
656                 result.pop();
657             }
658         }
659
660         if config.wrap_comments() && line.len() > fmt.shape.width && !has_url(line) {
661             match rewrite_string(line, &fmt, max_chars) {
662                 Some(ref s) => {
663                     is_prev_line_multi_line = s.contains('\n');
664                     result.push_str(s);
665                 }
666                 None if is_prev_line_multi_line => {
667                     // We failed to put the current `line` next to the previous `line`.
668                     // Remove the trailing space, then start rewrite on the next line.
669                     result.pop();
670                     result.push_str(&comment_line_separator);
671                     fmt.shape = Shape::legacy(max_chars, fmt_indent);
672                     match rewrite_string(line, &fmt, max_chars) {
673                         Some(ref s) => {
674                             is_prev_line_multi_line = s.contains('\n');
675                             result.push_str(s);
676                         }
677                         None => {
678                             is_prev_line_multi_line = false;
679                             result.push_str(line);
680                         }
681                     }
682                 }
683                 None => {
684                     is_prev_line_multi_line = false;
685                     result.push_str(line);
686                 }
687             }
688
689             fmt.shape = if is_prev_line_multi_line {
690                 // 1 = " "
691                 let offset = 1 + last_line_width(&result) - line_start.len();
692                 Shape {
693                     width: max_chars.saturating_sub(offset),
694                     indent: fmt_indent,
695                     offset: fmt.shape.offset + offset,
696                 }
697             } else {
698                 Shape::legacy(max_chars, fmt_indent)
699             };
700         } else {
701             if line.is_empty() && result.ends_with(' ') && !is_last {
702                 // Remove space if this is an empty comment or a doc comment.
703                 result.pop();
704             }
705             result.push_str(line);
706             fmt.shape = Shape::legacy(max_chars, fmt_indent);
707             is_prev_line_multi_line = false;
708         }
709     }
710     if !code_block_buffer.is_empty() {
711         // There is a code block that is not properly enclosed by backticks.
712         // We will leave them untouched.
713         result.push_str(&comment_line_separator);
714         result.push_str(&join_block(
715             &trim_custom_comment_prefix(&code_block_buffer),
716             &comment_line_separator,
717         ));
718     }
719     if !item_block_buffer.is_empty() {
720         // the last few lines are part of an itemized block
721         let ib = item_block.unwrap();
722         fmt.shape = Shape::legacy(max_chars, fmt_indent);
723         let item_fmt = ib.create_string_format(&fmt);
724         result.push_str(&comment_line_separator);
725         result.push_str(&ib.opener);
726         match rewrite_string(
727             &item_block_buffer.replace("\n", " "),
728             &item_fmt,
729             max_chars.saturating_sub(ib.indent),
730         ) {
731             Some(s) => result.push_str(&join_block(
732                 &s,
733                 &format!("{}{}", &comment_line_separator, ib.line_start),
734             )),
735             None => result.push_str(&join_block(&item_block_buffer, &comment_line_separator)),
736         };
737     }
738
739     result.push_str(closer);
740     if result.ends_with(opener) && opener.ends_with(' ') {
741         // Trailing space.
742         result.pop();
743     }
744
745     Some(result)
746 }
747
748 const RUSTFMT_CUSTOM_COMMENT_PREFIX: &str = "//#### ";
749
750 fn hide_sharp_behind_comment(s: &str) -> Cow<str> {
751     if s.trim_left().starts_with("# ") {
752         Cow::from(format!("{}{}", RUSTFMT_CUSTOM_COMMENT_PREFIX, s))
753     } else {
754         Cow::from(s)
755     }
756 }
757
758 fn trim_custom_comment_prefix(s: &str) -> String {
759     s.lines()
760         .map(|line| {
761             let left_trimmed = line.trim_left();
762             if left_trimmed.starts_with(RUSTFMT_CUSTOM_COMMENT_PREFIX) {
763                 let orig = left_trimmed.trim_left_matches(RUSTFMT_CUSTOM_COMMENT_PREFIX);
764                 // due to comment wrapping, a line that was originaly behind `#` is split over
765                 // multiple lines, which needs then to be prefixed with a `#`
766                 if !orig.trim_left().starts_with("# ") {
767                     format!("# {}", orig)
768                 } else {
769                     orig.to_string()
770                 }
771             } else {
772                 line.to_string()
773             }
774         })
775         .collect::<Vec<_>>()
776         .join("\n")
777 }
778
779 /// Returns true if the given string MAY include URLs or alike.
780 fn has_url(s: &str) -> bool {
781     // This function may return false positive, but should get its job done in most cases.
782     s.contains("https://") || s.contains("http://") || s.contains("ftp://") || s.contains("file://")
783 }
784
785 /// Given the span, rewrite the missing comment inside it if available.
786 /// Note that the given span must only include comments (or leading/trailing whitespaces).
787 pub fn rewrite_missing_comment(
788     span: Span,
789     shape: Shape,
790     context: &RewriteContext,
791 ) -> Option<String> {
792     let missing_snippet = context.snippet(span);
793     let trimmed_snippet = missing_snippet.trim();
794     if !trimmed_snippet.is_empty() {
795         rewrite_comment(trimmed_snippet, false, shape, context.config)
796     } else {
797         Some(String::new())
798     }
799 }
800
801 /// Recover the missing comments in the specified span, if available.
802 /// The layout of the comments will be preserved as long as it does not break the code
803 /// and its total width does not exceed the max width.
804 pub fn recover_missing_comment_in_span(
805     span: Span,
806     shape: Shape,
807     context: &RewriteContext,
808     used_width: usize,
809 ) -> Option<String> {
810     let missing_comment = rewrite_missing_comment(span, shape, context)?;
811     if missing_comment.is_empty() {
812         Some(String::new())
813     } else {
814         let missing_snippet = context.snippet(span);
815         let pos = missing_snippet.find('/').unwrap_or(0);
816         // 1 = ` `
817         let total_width = missing_comment.len() + used_width + 1;
818         let force_new_line_before_comment =
819             missing_snippet[..pos].contains('\n') || total_width > context.config.max_width();
820         let sep = if force_new_line_before_comment {
821             shape.indent.to_string_with_newline(context.config)
822         } else {
823             Cow::from(" ")
824         };
825         Some(format!("{}{}", sep, missing_comment))
826     }
827 }
828
829 /// Trim trailing whitespaces unless they consist of two or more whitespaces.
830 fn trim_right_unless_two_whitespaces(s: &str, is_doc_comment: bool) -> &str {
831     if is_doc_comment && s.ends_with("  ") {
832         s
833     } else {
834         s.trim_right()
835     }
836 }
837
838 /// Trims whitespace and aligns to indent, but otherwise does not change comments.
839 fn light_rewrite_comment(
840     orig: &str,
841     offset: Indent,
842     config: &Config,
843     is_doc_comment: bool,
844 ) -> Option<String> {
845     let lines: Vec<&str> = orig
846         .lines()
847         .map(|l| {
848             // This is basically just l.trim(), but in the case that a line starts
849             // with `*` we want to leave one space before it, so it aligns with the
850             // `*` in `/*`.
851             let first_non_whitespace = l.find(|c| !char::is_whitespace(c));
852             let left_trimmed = if let Some(fnw) = first_non_whitespace {
853                 if l.as_bytes()[fnw] == b'*' && fnw > 0 {
854                     &l[fnw - 1..]
855                 } else {
856                     &l[fnw..]
857                 }
858             } else {
859                 ""
860             };
861             // Preserve markdown's double-space line break syntax in doc comment.
862             trim_right_unless_two_whitespaces(left_trimmed, is_doc_comment)
863         })
864         .collect();
865     Some(lines.join(&format!("\n{}", offset.to_string(config))))
866 }
867
868 /// Trims comment characters and possibly a single space from the left of a string.
869 /// Does not trim all whitespace. If a single space is trimmed from the left of the string,
870 /// this function returns true.
871 fn left_trim_comment_line<'a>(line: &'a str, style: &CommentStyle) -> (&'a str, bool) {
872     if line.starts_with("//! ")
873         || line.starts_with("/// ")
874         || line.starts_with("/*! ")
875         || line.starts_with("/** ")
876     {
877         (&line[4..], true)
878     } else if let CommentStyle::Custom(opener) = *style {
879         if line.starts_with(opener) {
880             (&line[opener.len()..], true)
881         } else {
882             (&line[opener.trim_right().len()..], false)
883         }
884     } else if line.starts_with("/* ")
885         || line.starts_with("// ")
886         || line.starts_with("//!")
887         || line.starts_with("///")
888         || line.starts_with("** ")
889         || line.starts_with("/*!")
890         || (line.starts_with("/**") && !line.starts_with("/**/"))
891     {
892         (&line[3..], line.chars().nth(2).unwrap() == ' ')
893     } else if line.starts_with("/*")
894         || line.starts_with("* ")
895         || line.starts_with("//")
896         || line.starts_with("**")
897     {
898         (&line[2..], line.chars().nth(1).unwrap() == ' ')
899     } else if line.starts_with('*') {
900         (&line[1..], false)
901     } else {
902         (line, line.starts_with(' '))
903     }
904 }
905
906 pub trait FindUncommented {
907     fn find_uncommented(&self, pat: &str) -> Option<usize>;
908 }
909
910 impl FindUncommented for str {
911     fn find_uncommented(&self, pat: &str) -> Option<usize> {
912         let mut needle_iter = pat.chars();
913         for (kind, (i, b)) in CharClasses::new(self.char_indices()) {
914             match needle_iter.next() {
915                 None => {
916                     return Some(i - pat.len());
917                 }
918                 Some(c) => match kind {
919                     FullCodeCharKind::Normal | FullCodeCharKind::InString if b == c => {}
920                     _ => {
921                         needle_iter = pat.chars();
922                     }
923                 },
924             }
925         }
926
927         // Handle case where the pattern is a suffix of the search string
928         match needle_iter.next() {
929             Some(_) => None,
930             None => Some(self.len() - pat.len()),
931         }
932     }
933 }
934
935 // Returns the first byte position after the first comment. The given string
936 // is expected to be prefixed by a comment, including delimiters.
937 // Good: "/* /* inner */ outer */ code();"
938 // Bad:  "code(); // hello\n world!"
939 pub fn find_comment_end(s: &str) -> Option<usize> {
940     let mut iter = CharClasses::new(s.char_indices());
941     for (kind, (i, _c)) in &mut iter {
942         if kind == FullCodeCharKind::Normal || kind == FullCodeCharKind::InString {
943             return Some(i);
944         }
945     }
946
947     // Handle case where the comment ends at the end of s.
948     if iter.status == CharClassesStatus::Normal {
949         Some(s.len())
950     } else {
951         None
952     }
953 }
954
955 /// Returns true if text contains any comment.
956 pub fn contains_comment(text: &str) -> bool {
957     CharClasses::new(text.chars()).any(|(kind, _)| kind.is_comment())
958 }
959
960 /// Remove trailing spaces from the specified snippet. We do not remove spaces
961 /// inside strings or comments.
962 pub fn remove_trailing_white_spaces(text: &str) -> String {
963     let mut buffer = String::with_capacity(text.len());
964     let mut space_buffer = String::with_capacity(128);
965     for (char_kind, c) in CharClasses::new(text.chars()) {
966         match c {
967             '\n' => {
968                 if char_kind == FullCodeCharKind::InString {
969                     buffer.push_str(&space_buffer);
970                 }
971                 space_buffer.clear();
972                 buffer.push('\n');
973             }
974             _ if c.is_whitespace() => {
975                 space_buffer.push(c);
976             }
977             _ => {
978                 if !space_buffer.is_empty() {
979                     buffer.push_str(&space_buffer);
980                     space_buffer.clear();
981                 }
982                 buffer.push(c);
983             }
984         }
985     }
986     buffer
987 }
988
989 pub struct CharClasses<T>
990 where
991     T: Iterator,
992     T::Item: RichChar,
993 {
994     base: MultiPeek<T>,
995     status: CharClassesStatus,
996 }
997
998 pub trait RichChar {
999     fn get_char(&self) -> char;
1000 }
1001
1002 impl RichChar for char {
1003     fn get_char(&self) -> char {
1004         *self
1005     }
1006 }
1007
1008 impl RichChar for (usize, char) {
1009     fn get_char(&self) -> char {
1010         self.1
1011     }
1012 }
1013
1014 #[derive(PartialEq, Eq, Debug, Clone, Copy)]
1015 enum CharClassesStatus {
1016     Normal,
1017     LitString,
1018     LitStringEscape,
1019     LitRawString(u32),
1020     RawStringPrefix(u32),
1021     RawStringSuffix(u32),
1022     LitChar,
1023     LitCharEscape,
1024     // The u32 is the nesting deepness of the comment
1025     BlockComment(u32),
1026     // Status when the '/' has been consumed, but not yet the '*', deepness is
1027     // the new deepness (after the comment opening).
1028     BlockCommentOpening(u32),
1029     // Status when the '*' has been consumed, but not yet the '/', deepness is
1030     // the new deepness (after the comment closing).
1031     BlockCommentClosing(u32),
1032     LineComment,
1033 }
1034
1035 /// Distinguish between functional part of code and comments
1036 #[derive(PartialEq, Eq, Debug, Clone, Copy)]
1037 pub enum CodeCharKind {
1038     Normal,
1039     Comment,
1040 }
1041
1042 /// Distinguish between functional part of code and comments,
1043 /// describing opening and closing of comments for ease when chunking
1044 /// code from tagged characters
1045 #[derive(PartialEq, Eq, Debug, Clone, Copy)]
1046 pub enum FullCodeCharKind {
1047     Normal,
1048     /// The first character of a comment, there is only one for a comment (always '/')
1049     StartComment,
1050     /// Any character inside a comment including the second character of comment
1051     /// marks ("//", "/*")
1052     InComment,
1053     /// Last character of a comment, '\n' for a line comment, '/' for a block comment.
1054     EndComment,
1055     /// Start of a mutlitine string
1056     StartString,
1057     /// End of a mutlitine string
1058     EndString,
1059     /// Inside a string.
1060     InString,
1061 }
1062
1063 impl FullCodeCharKind {
1064     pub fn is_comment(self) -> bool {
1065         match self {
1066             FullCodeCharKind::StartComment
1067             | FullCodeCharKind::InComment
1068             | FullCodeCharKind::EndComment => true,
1069             _ => false,
1070         }
1071     }
1072
1073     pub fn is_string(self) -> bool {
1074         self == FullCodeCharKind::InString || self == FullCodeCharKind::StartString
1075     }
1076
1077     fn to_codecharkind(self) -> CodeCharKind {
1078         if self.is_comment() {
1079             CodeCharKind::Comment
1080         } else {
1081             CodeCharKind::Normal
1082         }
1083     }
1084 }
1085
1086 impl<T> CharClasses<T>
1087 where
1088     T: Iterator,
1089     T::Item: RichChar,
1090 {
1091     pub fn new(base: T) -> CharClasses<T> {
1092         CharClasses {
1093             base: multipeek(base),
1094             status: CharClassesStatus::Normal,
1095         }
1096     }
1097 }
1098
1099 fn is_raw_string_suffix<T>(iter: &mut MultiPeek<T>, count: u32) -> bool
1100 where
1101     T: Iterator,
1102     T::Item: RichChar,
1103 {
1104     for _ in 0..count {
1105         match iter.peek() {
1106             Some(c) if c.get_char() == '#' => continue,
1107             _ => return false,
1108         }
1109     }
1110     true
1111 }
1112
1113 impl<T> Iterator for CharClasses<T>
1114 where
1115     T: Iterator,
1116     T::Item: RichChar,
1117 {
1118     type Item = (FullCodeCharKind, T::Item);
1119
1120     fn next(&mut self) -> Option<(FullCodeCharKind, T::Item)> {
1121         let item = self.base.next()?;
1122         let chr = item.get_char();
1123         let mut char_kind = FullCodeCharKind::Normal;
1124         self.status = match self.status {
1125             CharClassesStatus::LitRawString(sharps) => {
1126                 char_kind = FullCodeCharKind::InString;
1127                 match chr {
1128                     '"' => {
1129                         if sharps == 0 {
1130                             char_kind = FullCodeCharKind::Normal;
1131                             CharClassesStatus::Normal
1132                         } else if is_raw_string_suffix(&mut self.base, sharps) {
1133                             CharClassesStatus::RawStringSuffix(sharps)
1134                         } else {
1135                             CharClassesStatus::LitRawString(sharps)
1136                         }
1137                     }
1138                     _ => CharClassesStatus::LitRawString(sharps),
1139                 }
1140             }
1141             CharClassesStatus::RawStringPrefix(sharps) => {
1142                 char_kind = FullCodeCharKind::InString;
1143                 match chr {
1144                     '#' => CharClassesStatus::RawStringPrefix(sharps + 1),
1145                     '"' => CharClassesStatus::LitRawString(sharps),
1146                     _ => CharClassesStatus::Normal, // Unreachable.
1147                 }
1148             }
1149             CharClassesStatus::RawStringSuffix(sharps) => {
1150                 match chr {
1151                     '#' => {
1152                         if sharps == 1 {
1153                             CharClassesStatus::Normal
1154                         } else {
1155                             char_kind = FullCodeCharKind::InString;
1156                             CharClassesStatus::RawStringSuffix(sharps - 1)
1157                         }
1158                     }
1159                     _ => CharClassesStatus::Normal, // Unreachable
1160                 }
1161             }
1162             CharClassesStatus::LitString => {
1163                 char_kind = FullCodeCharKind::InString;
1164                 match chr {
1165                     '"' => CharClassesStatus::Normal,
1166                     '\\' => CharClassesStatus::LitStringEscape,
1167                     _ => CharClassesStatus::LitString,
1168                 }
1169             }
1170             CharClassesStatus::LitStringEscape => {
1171                 char_kind = FullCodeCharKind::InString;
1172                 CharClassesStatus::LitString
1173             }
1174             CharClassesStatus::LitChar => match chr {
1175                 '\\' => CharClassesStatus::LitCharEscape,
1176                 '\'' => CharClassesStatus::Normal,
1177                 _ => CharClassesStatus::LitChar,
1178             },
1179             CharClassesStatus::LitCharEscape => CharClassesStatus::LitChar,
1180             CharClassesStatus::Normal => match chr {
1181                 'r' => match self.base.peek().map(|c| c.get_char()) {
1182                     Some('#') | Some('"') => {
1183                         char_kind = FullCodeCharKind::InString;
1184                         CharClassesStatus::RawStringPrefix(0)
1185                     }
1186                     _ => CharClassesStatus::Normal,
1187                 },
1188                 '"' => {
1189                     char_kind = FullCodeCharKind::InString;
1190                     CharClassesStatus::LitString
1191                 }
1192                 '\'' => {
1193                     // HACK: Work around mut borrow.
1194                     match self.base.peek() {
1195                         Some(next) if next.get_char() == '\\' => {
1196                             self.status = CharClassesStatus::LitChar;
1197                             return Some((char_kind, item));
1198                         }
1199                         _ => (),
1200                     }
1201
1202                     match self.base.peek() {
1203                         Some(next) if next.get_char() == '\'' => CharClassesStatus::LitChar,
1204                         _ => CharClassesStatus::Normal,
1205                     }
1206                 }
1207                 '/' => match self.base.peek() {
1208                     Some(next) if next.get_char() == '*' => {
1209                         self.status = CharClassesStatus::BlockCommentOpening(1);
1210                         return Some((FullCodeCharKind::StartComment, item));
1211                     }
1212                     Some(next) if next.get_char() == '/' => {
1213                         self.status = CharClassesStatus::LineComment;
1214                         return Some((FullCodeCharKind::StartComment, item));
1215                     }
1216                     _ => CharClassesStatus::Normal,
1217                 },
1218                 _ => CharClassesStatus::Normal,
1219             },
1220             CharClassesStatus::BlockComment(deepness) => {
1221                 assert_ne!(deepness, 0);
1222                 self.status = match self.base.peek() {
1223                     Some(next) if next.get_char() == '/' && chr == '*' => {
1224                         CharClassesStatus::BlockCommentClosing(deepness - 1)
1225                     }
1226                     Some(next) if next.get_char() == '*' && chr == '/' => {
1227                         CharClassesStatus::BlockCommentOpening(deepness + 1)
1228                     }
1229                     _ => CharClassesStatus::BlockComment(deepness),
1230                 };
1231                 return Some((FullCodeCharKind::InComment, item));
1232             }
1233             CharClassesStatus::BlockCommentOpening(deepness) => {
1234                 assert_eq!(chr, '*');
1235                 self.status = CharClassesStatus::BlockComment(deepness);
1236                 return Some((FullCodeCharKind::InComment, item));
1237             }
1238             CharClassesStatus::BlockCommentClosing(deepness) => {
1239                 assert_eq!(chr, '/');
1240                 if deepness == 0 {
1241                     self.status = CharClassesStatus::Normal;
1242                     return Some((FullCodeCharKind::EndComment, item));
1243                 } else {
1244                     self.status = CharClassesStatus::BlockComment(deepness);
1245                     return Some((FullCodeCharKind::InComment, item));
1246                 }
1247             }
1248             CharClassesStatus::LineComment => match chr {
1249                 '\n' => {
1250                     self.status = CharClassesStatus::Normal;
1251                     return Some((FullCodeCharKind::EndComment, item));
1252                 }
1253                 _ => {
1254                     self.status = CharClassesStatus::LineComment;
1255                     return Some((FullCodeCharKind::InComment, item));
1256                 }
1257             },
1258         };
1259         Some((char_kind, item))
1260     }
1261 }
1262
1263 /// An iterator over the lines of a string, paired with the char kind at the
1264 /// end of the line.
1265 pub struct LineClasses<'a> {
1266     base: iter::Peekable<CharClasses<std::str::Chars<'a>>>,
1267     kind: FullCodeCharKind,
1268 }
1269
1270 impl<'a> LineClasses<'a> {
1271     pub fn new(s: &'a str) -> Self {
1272         LineClasses {
1273             base: CharClasses::new(s.chars()).peekable(),
1274             kind: FullCodeCharKind::Normal,
1275         }
1276     }
1277 }
1278
1279 impl<'a> Iterator for LineClasses<'a> {
1280     type Item = (FullCodeCharKind, String);
1281
1282     fn next(&mut self) -> Option<Self::Item> {
1283         self.base.peek()?;
1284
1285         let mut line = String::new();
1286
1287         let start_class = match self.base.peek() {
1288             Some((kind, _)) => *kind,
1289             None => FullCodeCharKind::Normal,
1290         };
1291
1292         while let Some((kind, c)) = self.base.next() {
1293             if c == '\n' {
1294                 self.kind = match (start_class, kind) {
1295                     (FullCodeCharKind::Normal, FullCodeCharKind::InString) => {
1296                         FullCodeCharKind::StartString
1297                     }
1298                     (FullCodeCharKind::InString, FullCodeCharKind::Normal) => {
1299                         FullCodeCharKind::EndString
1300                     }
1301                     _ => kind,
1302                 };
1303                 break;
1304             } else {
1305                 line.push(c);
1306             }
1307         }
1308
1309         Some((self.kind, line))
1310     }
1311 }
1312
1313 /// Iterator over functional and commented parts of a string. Any part of a string is either
1314 /// functional code, either *one* block comment, either *one* line comment. Whitespace between
1315 /// comments is functional code. Line comments contain their ending newlines.
1316 struct UngroupedCommentCodeSlices<'a> {
1317     slice: &'a str,
1318     iter: iter::Peekable<CharClasses<std::str::CharIndices<'a>>>,
1319 }
1320
1321 impl<'a> UngroupedCommentCodeSlices<'a> {
1322     fn new(code: &'a str) -> UngroupedCommentCodeSlices<'a> {
1323         UngroupedCommentCodeSlices {
1324             slice: code,
1325             iter: CharClasses::new(code.char_indices()).peekable(),
1326         }
1327     }
1328 }
1329
1330 impl<'a> Iterator for UngroupedCommentCodeSlices<'a> {
1331     type Item = (CodeCharKind, usize, &'a str);
1332
1333     fn next(&mut self) -> Option<Self::Item> {
1334         let (kind, (start_idx, _)) = self.iter.next()?;
1335         match kind {
1336             FullCodeCharKind::Normal | FullCodeCharKind::InString => {
1337                 // Consume all the Normal code
1338                 while let Some(&(char_kind, _)) = self.iter.peek() {
1339                     if char_kind.is_comment() {
1340                         break;
1341                     }
1342                     let _ = self.iter.next();
1343                 }
1344             }
1345             FullCodeCharKind::StartComment => {
1346                 // Consume the whole comment
1347                 while let Some((FullCodeCharKind::InComment, (_, _))) = self.iter.next() {}
1348             }
1349             _ => panic!(),
1350         }
1351         let slice = match self.iter.peek() {
1352             Some(&(_, (end_idx, _))) => &self.slice[start_idx..end_idx],
1353             None => &self.slice[start_idx..],
1354         };
1355         Some((
1356             if kind.is_comment() {
1357                 CodeCharKind::Comment
1358             } else {
1359                 CodeCharKind::Normal
1360             },
1361             start_idx,
1362             slice,
1363         ))
1364     }
1365 }
1366
1367 /// Iterator over an alternating sequence of functional and commented parts of
1368 /// a string. The first item is always a, possibly zero length, subslice of
1369 /// functional text. Line style comments contain their ending newlines.
1370 pub struct CommentCodeSlices<'a> {
1371     slice: &'a str,
1372     last_slice_kind: CodeCharKind,
1373     last_slice_end: usize,
1374 }
1375
1376 impl<'a> CommentCodeSlices<'a> {
1377     pub fn new(slice: &'a str) -> CommentCodeSlices<'a> {
1378         CommentCodeSlices {
1379             slice,
1380             last_slice_kind: CodeCharKind::Comment,
1381             last_slice_end: 0,
1382         }
1383     }
1384 }
1385
1386 impl<'a> Iterator for CommentCodeSlices<'a> {
1387     type Item = (CodeCharKind, usize, &'a str);
1388
1389     fn next(&mut self) -> Option<Self::Item> {
1390         if self.last_slice_end == self.slice.len() {
1391             return None;
1392         }
1393
1394         let mut sub_slice_end = self.last_slice_end;
1395         let mut first_whitespace = None;
1396         let subslice = &self.slice[self.last_slice_end..];
1397         let mut iter = CharClasses::new(subslice.char_indices());
1398
1399         for (kind, (i, c)) in &mut iter {
1400             let is_comment_connector = self.last_slice_kind == CodeCharKind::Normal
1401                 && &subslice[..2] == "//"
1402                 && [' ', '\t'].contains(&c);
1403
1404             if is_comment_connector && first_whitespace.is_none() {
1405                 first_whitespace = Some(i);
1406             }
1407
1408             if kind.to_codecharkind() == self.last_slice_kind && !is_comment_connector {
1409                 let last_index = match first_whitespace {
1410                     Some(j) => j,
1411                     None => i,
1412                 };
1413                 sub_slice_end = self.last_slice_end + last_index;
1414                 break;
1415             }
1416
1417             if !is_comment_connector {
1418                 first_whitespace = None;
1419             }
1420         }
1421
1422         if let (None, true) = (iter.next(), sub_slice_end == self.last_slice_end) {
1423             // This was the last subslice.
1424             sub_slice_end = match first_whitespace {
1425                 Some(i) => self.last_slice_end + i,
1426                 None => self.slice.len(),
1427             };
1428         }
1429
1430         let kind = match self.last_slice_kind {
1431             CodeCharKind::Comment => CodeCharKind::Normal,
1432             CodeCharKind::Normal => CodeCharKind::Comment,
1433         };
1434         let res = (
1435             kind,
1436             self.last_slice_end,
1437             &self.slice[self.last_slice_end..sub_slice_end],
1438         );
1439         self.last_slice_end = sub_slice_end;
1440         self.last_slice_kind = kind;
1441
1442         Some(res)
1443     }
1444 }
1445
1446 /// Checks is `new` didn't miss any comment from `span`, if it removed any, return previous text
1447 /// (if it fits in the width/offset, else return None), else return `new`
1448 pub fn recover_comment_removed(
1449     new: String,
1450     span: Span,
1451     context: &RewriteContext,
1452 ) -> Option<String> {
1453     let snippet = context.snippet(span);
1454     if snippet != new && changed_comment_content(snippet, &new) {
1455         // We missed some comments. Warn and keep the original text.
1456         if context.config.error_on_unformatted() {
1457             context.report.append(
1458                 context.source_map.span_to_filename(span).into(),
1459                 vec![FormattingError::from_span(
1460                     span,
1461                     &context.source_map,
1462                     ErrorKind::LostComment,
1463                 )],
1464             );
1465         }
1466         Some(snippet.to_owned())
1467     } else {
1468         Some(new)
1469     }
1470 }
1471
1472 pub fn filter_normal_code(code: &str) -> String {
1473     let mut buffer = String::with_capacity(code.len());
1474     LineClasses::new(code).for_each(|(kind, line)| match kind {
1475         FullCodeCharKind::Normal
1476         | FullCodeCharKind::StartString
1477         | FullCodeCharKind::InString
1478         | FullCodeCharKind::EndString => {
1479             buffer.push_str(&line);
1480             buffer.push('\n');
1481         }
1482         _ => (),
1483     });
1484     if !code.ends_with('\n') && buffer.ends_with('\n') {
1485         buffer.pop();
1486     }
1487     buffer
1488 }
1489
1490 /// Return true if the two strings of code have the same payload of comments.
1491 /// The payload of comments is everything in the string except:
1492 ///     - actual code (not comments)
1493 ///     - comment start/end marks
1494 ///     - whitespace
1495 ///     - '*' at the beginning of lines in block comments
1496 fn changed_comment_content(orig: &str, new: &str) -> bool {
1497     // Cannot write this as a fn since we cannot return types containing closures
1498     let code_comment_content = |code| {
1499         let slices = UngroupedCommentCodeSlices::new(code);
1500         slices
1501             .filter(|&(ref kind, _, _)| *kind == CodeCharKind::Comment)
1502             .flat_map(|(_, _, s)| CommentReducer::new(s))
1503     };
1504     let res = code_comment_content(orig).ne(code_comment_content(new));
1505     debug!(
1506         "comment::changed_comment_content: {}\norig: '{}'\nnew: '{}'\nraw_old: {}\nraw_new: {}",
1507         res,
1508         orig,
1509         new,
1510         code_comment_content(orig).collect::<String>(),
1511         code_comment_content(new).collect::<String>()
1512     );
1513     res
1514 }
1515
1516 /// Iterator over the 'payload' characters of a comment.
1517 /// It skips whitespace, comment start/end marks, and '*' at the beginning of lines.
1518 /// The comment must be one comment, ie not more than one start mark (no multiple line comments,
1519 /// for example).
1520 struct CommentReducer<'a> {
1521     is_block: bool,
1522     at_start_line: bool,
1523     iter: std::str::Chars<'a>,
1524 }
1525
1526 impl<'a> CommentReducer<'a> {
1527     fn new(comment: &'a str) -> CommentReducer<'a> {
1528         let is_block = comment.starts_with("/*");
1529         let comment = remove_comment_header(comment);
1530         CommentReducer {
1531             is_block,
1532             at_start_line: false, // There are no supplementary '*' on the first line
1533             iter: comment.chars(),
1534         }
1535     }
1536 }
1537
1538 impl<'a> Iterator for CommentReducer<'a> {
1539     type Item = char;
1540
1541     fn next(&mut self) -> Option<Self::Item> {
1542         loop {
1543             let mut c = self.iter.next()?;
1544             if self.is_block && self.at_start_line {
1545                 while c.is_whitespace() {
1546                     c = self.iter.next()?;
1547                 }
1548                 // Ignore leading '*'
1549                 if c == '*' {
1550                     c = self.iter.next()?;
1551                 }
1552             } else if c == '\n' {
1553                 self.at_start_line = true;
1554             }
1555             if !c.is_whitespace() {
1556                 return Some(c);
1557             }
1558         }
1559     }
1560 }
1561
1562 fn remove_comment_header(comment: &str) -> &str {
1563     if comment.starts_with("///") || comment.starts_with("//!") {
1564         &comment[3..]
1565     } else if comment.starts_with("//") {
1566         &comment[2..]
1567     } else if (comment.starts_with("/**") && !comment.starts_with("/**/"))
1568         || comment.starts_with("/*!")
1569     {
1570         &comment[3..comment.len() - 2]
1571     } else {
1572         assert!(
1573             comment.starts_with("/*"),
1574             format!("string '{}' is not a comment", comment)
1575         );
1576         &comment[2..comment.len() - 2]
1577     }
1578 }
1579
1580 #[cfg(test)]
1581 mod test {
1582     use super::*;
1583     use shape::{Indent, Shape};
1584
1585     #[test]
1586     fn char_classes() {
1587         let mut iter = CharClasses::new("//\n\n".chars());
1588
1589         assert_eq!((FullCodeCharKind::StartComment, '/'), iter.next().unwrap());
1590         assert_eq!((FullCodeCharKind::InComment, '/'), iter.next().unwrap());
1591         assert_eq!((FullCodeCharKind::EndComment, '\n'), iter.next().unwrap());
1592         assert_eq!((FullCodeCharKind::Normal, '\n'), iter.next().unwrap());
1593         assert_eq!(None, iter.next());
1594     }
1595
1596     #[test]
1597     fn comment_code_slices() {
1598         let input = "code(); /* test */ 1 + 1";
1599         let mut iter = CommentCodeSlices::new(input);
1600
1601         assert_eq!((CodeCharKind::Normal, 0, "code(); "), iter.next().unwrap());
1602         assert_eq!(
1603             (CodeCharKind::Comment, 8, "/* test */"),
1604             iter.next().unwrap()
1605         );
1606         assert_eq!((CodeCharKind::Normal, 18, " 1 + 1"), iter.next().unwrap());
1607         assert_eq!(None, iter.next());
1608     }
1609
1610     #[test]
1611     fn comment_code_slices_two() {
1612         let input = "// comment\n    test();";
1613         let mut iter = CommentCodeSlices::new(input);
1614
1615         assert_eq!((CodeCharKind::Normal, 0, ""), iter.next().unwrap());
1616         assert_eq!(
1617             (CodeCharKind::Comment, 0, "// comment\n"),
1618             iter.next().unwrap()
1619         );
1620         assert_eq!(
1621             (CodeCharKind::Normal, 11, "    test();"),
1622             iter.next().unwrap()
1623         );
1624         assert_eq!(None, iter.next());
1625     }
1626
1627     #[test]
1628     fn comment_code_slices_three() {
1629         let input = "1 // comment\n    // comment2\n\n";
1630         let mut iter = CommentCodeSlices::new(input);
1631
1632         assert_eq!((CodeCharKind::Normal, 0, "1 "), iter.next().unwrap());
1633         assert_eq!(
1634             (CodeCharKind::Comment, 2, "// comment\n    // comment2\n"),
1635             iter.next().unwrap()
1636         );
1637         assert_eq!((CodeCharKind::Normal, 29, "\n"), iter.next().unwrap());
1638         assert_eq!(None, iter.next());
1639     }
1640
1641     #[test]
1642     #[rustfmt::skip]
1643     fn format_doc_comments() {
1644         let mut wrap_normalize_config: ::config::Config = Default::default();
1645         wrap_normalize_config.set().wrap_comments(true);
1646         wrap_normalize_config.set().normalize_comments(true);
1647
1648         let mut wrap_config: ::config::Config = Default::default();
1649         wrap_config.set().wrap_comments(true);
1650
1651         let comment = rewrite_comment(" //test",
1652                                       true,
1653                                       Shape::legacy(100, Indent::new(0, 100)),
1654                                       &wrap_normalize_config).unwrap();
1655         assert_eq!("/* test */", comment);
1656
1657         let comment = rewrite_comment("// comment on a",
1658                                       false,
1659                                       Shape::legacy(10, Indent::empty()),
1660                                       &wrap_normalize_config).unwrap();
1661         assert_eq!("// comment\n// on a", comment);
1662
1663         let comment = rewrite_comment("//  A multi line comment\n             // between args.",
1664                                       false,
1665                                       Shape::legacy(60, Indent::new(0, 12)),
1666                                       &wrap_normalize_config).unwrap();
1667         assert_eq!("//  A multi line comment\n            // between args.", comment);
1668
1669         let input = "// comment";
1670         let expected =
1671             "/* comment */";
1672         let comment = rewrite_comment(input,
1673                                       true,
1674                                       Shape::legacy(9, Indent::new(0, 69)),
1675                                       &wrap_normalize_config).unwrap();
1676         assert_eq!(expected, comment);
1677
1678         let comment = rewrite_comment("/*   trimmed    */",
1679                                       true,
1680                                       Shape::legacy(100, Indent::new(0, 100)),
1681                                       &wrap_normalize_config).unwrap();
1682         assert_eq!("/* trimmed */", comment);
1683
1684         // check that different comment style are properly recognised
1685         let comment = rewrite_comment(r#"/// test1
1686                                          /// test2
1687                                          /*
1688                                           * test3
1689                                           */"#,
1690                                       false,
1691                                       Shape::legacy(100, Indent::new(0, 0)),
1692                                       &wrap_normalize_config).unwrap();
1693         assert_eq!("/// test1\n/// test2\n// test3", comment);
1694
1695         // check that the blank line marks the end of a commented paragraph
1696         let comment = rewrite_comment(r#"// test1
1697
1698                                          // test2"#,
1699                                       false,
1700                                       Shape::legacy(100, Indent::new(0, 0)),
1701                                       &wrap_normalize_config).unwrap();
1702         assert_eq!("// test1\n\n// test2", comment);
1703
1704         // check that the blank line marks the end of a custom-commented paragraph
1705         let comment = rewrite_comment(r#"//@ test1
1706
1707                                          //@ test2"#,
1708                                       false,
1709                                       Shape::legacy(100, Indent::new(0, 0)),
1710                                       &wrap_normalize_config).unwrap();
1711         assert_eq!("//@ test1\n\n//@ test2", comment);
1712
1713         // check that bare lines are just indented but left unchanged otherwise
1714         let comment = rewrite_comment(r#"// test1
1715                                          /*
1716                                            a bare line!
1717
1718                                                 another bare line!
1719                                           */"#,
1720                                       false,
1721                                       Shape::legacy(100, Indent::new(0, 0)),
1722                                       &wrap_config).unwrap();
1723         assert_eq!("// test1\n/*\n a bare line!\n\n      another bare line!\n*/", comment);
1724     }
1725
1726     // This is probably intended to be a non-test fn, but it is not used. I'm
1727     // keeping it around unless it helps us test stuff.
1728     fn uncommented(text: &str) -> String {
1729         CharClasses::new(text.chars())
1730             .filter_map(|(s, c)| match s {
1731                 FullCodeCharKind::Normal | FullCodeCharKind::InString => Some(c),
1732                 _ => None,
1733             })
1734             .collect()
1735     }
1736
1737     #[test]
1738     fn test_uncommented() {
1739         assert_eq!(&uncommented("abc/*...*/"), "abc");
1740         assert_eq!(
1741             &uncommented("// .... /* \n../* /* *** / */ */a/* // */c\n"),
1742             "..ac\n"
1743         );
1744         assert_eq!(&uncommented("abc \" /* */\" qsdf"), "abc \" /* */\" qsdf");
1745     }
1746
1747     #[test]
1748     fn test_contains_comment() {
1749         assert_eq!(contains_comment("abc"), false);
1750         assert_eq!(contains_comment("abc // qsdf"), true);
1751         assert_eq!(contains_comment("abc /* kqsdf"), true);
1752         assert_eq!(contains_comment("abc \" /* */\" qsdf"), false);
1753     }
1754
1755     #[test]
1756     fn test_find_uncommented() {
1757         fn check(haystack: &str, needle: &str, expected: Option<usize>) {
1758             assert_eq!(expected, haystack.find_uncommented(needle));
1759         }
1760
1761         check("/*/ */test", "test", Some(6));
1762         check("//test\ntest", "test", Some(7));
1763         check("/* comment only */", "whatever", None);
1764         check(
1765             "/* comment */ some text /* more commentary */ result",
1766             "result",
1767             Some(46),
1768         );
1769         check("sup // sup", "p", Some(2));
1770         check("sup", "x", None);
1771         check(r#"π? /**/ π is nice!"#, r#"π is nice"#, Some(9));
1772         check("/*sup yo? \n sup*/ sup", "p", Some(20));
1773         check("hel/*lohello*/lo", "hello", None);
1774         check("acb", "ab", None);
1775         check(",/*A*/ ", ",", Some(0));
1776         check("abc", "abc", Some(0));
1777         check("/* abc */", "abc", None);
1778         check("/**/abc/* */", "abc", Some(4));
1779         check("\"/* abc */\"", "abc", Some(4));
1780         check("\"/* abc", "abc", Some(4));
1781     }
1782
1783     #[test]
1784     fn test_remove_trailing_white_spaces() {
1785         let s = "    r#\"\n        test\n    \"#";
1786         assert_eq!(remove_trailing_white_spaces(&s), s);
1787     }
1788
1789     #[test]
1790     fn test_filter_normal_code() {
1791         let s = r#"
1792 fn main() {
1793     println!("hello, world");
1794 }
1795 "#;
1796         assert_eq!(s, filter_normal_code(s));
1797         let s_with_comment = r#"
1798 fn main() {
1799     // hello, world
1800     println!("hello, world");
1801 }
1802 "#;
1803         assert_eq!(s, filter_normal_code(s_with_comment));
1804     }
1805 }