]> git.lizzy.rs Git - rust.git/blob - src/comment.rs
handle lines prefixed with a # inside code blocks
[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(&item_block_buffer.replace("\n", " "), &item_fmt) {
583                 Some(s) => result.push_str(&join_block(
584                     &s,
585                     &format!("{}{}", &comment_line_separator, ib.line_start),
586                 )),
587                 None => result.push_str(&join_block(&item_block_buffer, &comment_line_separator)),
588             };
589             item_block_buffer.clear();
590         } else if let Some(ref attr) = code_block_attr {
591             if line.starts_with("```") {
592                 let code_block = match attr {
593                     CodeBlockAttribute::Ignore | CodeBlockAttribute::Text => {
594                         trim_custom_comment_prefix(&code_block_buffer)
595                     }
596                     _ if code_block_buffer.is_empty() => String::new(),
597                     _ => {
598                         let mut config = config.clone();
599                         config.set().format_doc_comments(false);
600                         match ::format_code_block(&code_block_buffer, &config) {
601                             Some(ref s) => trim_custom_comment_prefix(s),
602                             None => trim_custom_comment_prefix(&code_block_buffer),
603                         }
604                     }
605                 };
606                 if !code_block.is_empty() {
607                     result.push_str(&comment_line_separator);
608                     result.push_str(&join_block(&code_block, &comment_line_separator));
609                 }
610                 code_block_buffer.clear();
611                 result.push_str(&comment_line_separator);
612                 result.push_str(line);
613                 code_block_attr = None;
614             } else {
615                 code_block_buffer.push_str(&hide_sharp_behind_comment(line));
616                 code_block_buffer.push('\n');
617             }
618             continue;
619         }
620
621         code_block_attr = None;
622         item_block = None;
623         if line.starts_with("```") {
624             code_block_attr = Some(CodeBlockAttribute::new(&line[3..]))
625         } else if config.wrap_comments() && ItemizedBlock::is_itemized_line(&line) {
626             let ib = ItemizedBlock::new(&line);
627             item_block_buffer.push_str(&line[ib.indent..]);
628             item_block_buffer.push('\n');
629             item_block = Some(ib);
630             continue;
631         }
632
633         if result == opener {
634             let force_leading_whitespace = opener == "/* " && count_newlines(orig) == 0;
635             if !has_leading_whitespace && !force_leading_whitespace && result.ends_with(' ') {
636                 result.pop();
637             }
638             if line.is_empty() {
639                 continue;
640             }
641         } else if is_prev_line_multi_line && !line.is_empty() {
642             result.push(' ')
643         } else if is_last && line.is_empty() {
644             // trailing blank lines are unwanted
645             if !closer.is_empty() {
646                 result.push_str(&indent_str);
647             }
648             break;
649         } else {
650             result.push_str(&comment_line_separator);
651             if !has_leading_whitespace && result.ends_with(' ') {
652                 result.pop();
653             }
654         }
655
656         if config.wrap_comments() && line.len() > fmt.shape.width && !has_url(line) {
657             match rewrite_string(line, &fmt) {
658                 Some(ref s) => {
659                     is_prev_line_multi_line = s.contains('\n');
660                     result.push_str(s);
661                 }
662                 None if is_prev_line_multi_line => {
663                     // We failed to put the current `line` next to the previous `line`.
664                     // Remove the trailing space, then start rewrite on the next line.
665                     result.pop();
666                     result.push_str(&comment_line_separator);
667                     fmt.shape = Shape::legacy(max_chars, fmt_indent);
668                     match rewrite_string(line, &fmt) {
669                         Some(ref s) => {
670                             is_prev_line_multi_line = s.contains('\n');
671                             result.push_str(s);
672                         }
673                         None => {
674                             is_prev_line_multi_line = false;
675                             result.push_str(line);
676                         }
677                     }
678                 }
679                 None => {
680                     is_prev_line_multi_line = false;
681                     result.push_str(line);
682                 }
683             }
684
685             fmt.shape = if is_prev_line_multi_line {
686                 // 1 = " "
687                 let offset = 1 + last_line_width(&result) - line_start.len();
688                 Shape {
689                     width: max_chars.saturating_sub(offset),
690                     indent: fmt_indent,
691                     offset: fmt.shape.offset + offset,
692                 }
693             } else {
694                 Shape::legacy(max_chars, fmt_indent)
695             };
696         } else {
697             if line.is_empty() && result.ends_with(' ') && !is_last {
698                 // Remove space if this is an empty comment or a doc comment.
699                 result.pop();
700             }
701             result.push_str(line);
702             fmt.shape = Shape::legacy(max_chars, fmt_indent);
703             is_prev_line_multi_line = false;
704         }
705     }
706     if !code_block_buffer.is_empty() {
707         // There is a code block that is not properly enclosed by backticks.
708         // We will leave them untouched.
709         result.push_str(&comment_line_separator);
710         result.push_str(&join_block(
711             &trim_custom_comment_prefix(&code_block_buffer),
712             &comment_line_separator,
713         ));
714     }
715     if !item_block_buffer.is_empty() {
716         // the last few lines are part of an itemized block
717         let ib = item_block.unwrap();
718         fmt.shape = Shape::legacy(max_chars, fmt_indent);
719         let item_fmt = ib.create_string_format(&fmt);
720         result.push_str(&comment_line_separator);
721         result.push_str(&ib.opener);
722         match rewrite_string(&item_block_buffer.replace("\n", " "), &item_fmt) {
723             Some(s) => result.push_str(&join_block(
724                 &s,
725                 &format!("{}{}", &comment_line_separator, ib.line_start),
726             )),
727             None => result.push_str(&join_block(&item_block_buffer, &comment_line_separator)),
728         };
729     }
730
731     result.push_str(closer);
732     if result.ends_with(opener) && opener.ends_with(' ') {
733         // Trailing space.
734         result.pop();
735     }
736
737     Some(result)
738 }
739
740 const RUSTFMT_CUSTOM_COMMENT_PREFIX: &str = "//#### ";
741
742 fn hide_sharp_behind_comment(s: &str) -> Cow<str> {
743     if s.trim_left().starts_with("# ") {
744         Cow::from(format!("{}{}", RUSTFMT_CUSTOM_COMMENT_PREFIX, s))
745     } else {
746         Cow::from(s)
747     }
748 }
749
750 fn trim_custom_comment_prefix(s: &str) -> String {
751     s.lines()
752         .map(|line| {
753             let left_trimmed = line.trim_left();
754             if left_trimmed.starts_with(RUSTFMT_CUSTOM_COMMENT_PREFIX) {
755                 let orig = left_trimmed.trim_left_matches(RUSTFMT_CUSTOM_COMMENT_PREFIX);
756                 // due to comment wrapping, a line that was originaly behind `#` is split over
757                 // multiple lines, which needs then to be prefixed with a `#`
758                 if !orig.trim_left().starts_with("# ") {
759                     format!("# {}", orig)
760                 } else {
761                     orig.to_string()
762                 }
763             } else {
764                 line.to_string()
765             }
766         })
767         .collect::<Vec<_>>()
768         .join("\n")
769 }
770
771 /// Returns true if the given string MAY include URLs or alike.
772 fn has_url(s: &str) -> bool {
773     // This function may return false positive, but should get its job done in most cases.
774     s.contains("https://") || s.contains("http://") || s.contains("ftp://") || s.contains("file://")
775 }
776
777 /// Given the span, rewrite the missing comment inside it if available.
778 /// Note that the given span must only include comments (or leading/trailing whitespaces).
779 pub fn rewrite_missing_comment(
780     span: Span,
781     shape: Shape,
782     context: &RewriteContext,
783 ) -> Option<String> {
784     let missing_snippet = context.snippet(span);
785     let trimmed_snippet = missing_snippet.trim();
786     if !trimmed_snippet.is_empty() {
787         rewrite_comment(trimmed_snippet, false, shape, context.config)
788     } else {
789         Some(String::new())
790     }
791 }
792
793 /// Recover the missing comments in the specified span, if available.
794 /// The layout of the comments will be preserved as long as it does not break the code
795 /// and its total width does not exceed the max width.
796 pub fn recover_missing_comment_in_span(
797     span: Span,
798     shape: Shape,
799     context: &RewriteContext,
800     used_width: usize,
801 ) -> Option<String> {
802     let missing_comment = rewrite_missing_comment(span, shape, context)?;
803     if missing_comment.is_empty() {
804         Some(String::new())
805     } else {
806         let missing_snippet = context.snippet(span);
807         let pos = missing_snippet.find('/').unwrap_or(0);
808         // 1 = ` `
809         let total_width = missing_comment.len() + used_width + 1;
810         let force_new_line_before_comment =
811             missing_snippet[..pos].contains('\n') || total_width > context.config.max_width();
812         let sep = if force_new_line_before_comment {
813             shape.indent.to_string_with_newline(context.config)
814         } else {
815             Cow::from(" ")
816         };
817         Some(format!("{}{}", sep, missing_comment))
818     }
819 }
820
821 /// Trim trailing whitespaces unless they consist of two or more whitespaces.
822 fn trim_right_unless_two_whitespaces(s: &str, is_doc_comment: bool) -> &str {
823     if is_doc_comment && s.ends_with("  ") {
824         s
825     } else {
826         s.trim_right()
827     }
828 }
829
830 /// Trims whitespace and aligns to indent, but otherwise does not change comments.
831 fn light_rewrite_comment(
832     orig: &str,
833     offset: Indent,
834     config: &Config,
835     is_doc_comment: bool,
836 ) -> Option<String> {
837     let lines: Vec<&str> = orig
838         .lines()
839         .map(|l| {
840             // This is basically just l.trim(), but in the case that a line starts
841             // with `*` we want to leave one space before it, so it aligns with the
842             // `*` in `/*`.
843             let first_non_whitespace = l.find(|c| !char::is_whitespace(c));
844             let left_trimmed = if let Some(fnw) = first_non_whitespace {
845                 if l.as_bytes()[fnw] == b'*' && fnw > 0 {
846                     &l[fnw - 1..]
847                 } else {
848                     &l[fnw..]
849                 }
850             } else {
851                 ""
852             };
853             // Preserve markdown's double-space line break syntax in doc comment.
854             trim_right_unless_two_whitespaces(left_trimmed, is_doc_comment)
855         })
856         .collect();
857     Some(lines.join(&format!("\n{}", offset.to_string(config))))
858 }
859
860 /// Trims comment characters and possibly a single space from the left of a string.
861 /// Does not trim all whitespace. If a single space is trimmed from the left of the string,
862 /// this function returns true.
863 fn left_trim_comment_line<'a>(line: &'a str, style: &CommentStyle) -> (&'a str, bool) {
864     if line.starts_with("//! ")
865         || line.starts_with("/// ")
866         || line.starts_with("/*! ")
867         || line.starts_with("/** ")
868     {
869         (&line[4..], true)
870     } else if let CommentStyle::Custom(opener) = *style {
871         if line.starts_with(opener) {
872             (&line[opener.len()..], true)
873         } else {
874             (&line[opener.trim_right().len()..], false)
875         }
876     } else if line.starts_with("/* ")
877         || line.starts_with("// ")
878         || line.starts_with("//!")
879         || line.starts_with("///")
880         || line.starts_with("** ")
881         || line.starts_with("/*!")
882         || (line.starts_with("/**") && !line.starts_with("/**/"))
883     {
884         (&line[3..], line.chars().nth(2).unwrap() == ' ')
885     } else if line.starts_with("/*")
886         || line.starts_with("* ")
887         || line.starts_with("//")
888         || line.starts_with("**")
889     {
890         (&line[2..], line.chars().nth(1).unwrap() == ' ')
891     } else if line.starts_with('*') {
892         (&line[1..], false)
893     } else {
894         (line, line.starts_with(' '))
895     }
896 }
897
898 pub trait FindUncommented {
899     fn find_uncommented(&self, pat: &str) -> Option<usize>;
900 }
901
902 impl FindUncommented for str {
903     fn find_uncommented(&self, pat: &str) -> Option<usize> {
904         let mut needle_iter = pat.chars();
905         for (kind, (i, b)) in CharClasses::new(self.char_indices()) {
906             match needle_iter.next() {
907                 None => {
908                     return Some(i - pat.len());
909                 }
910                 Some(c) => match kind {
911                     FullCodeCharKind::Normal | FullCodeCharKind::InString if b == c => {}
912                     _ => {
913                         needle_iter = pat.chars();
914                     }
915                 },
916             }
917         }
918
919         // Handle case where the pattern is a suffix of the search string
920         match needle_iter.next() {
921             Some(_) => None,
922             None => Some(self.len() - pat.len()),
923         }
924     }
925 }
926
927 // Returns the first byte position after the first comment. The given string
928 // is expected to be prefixed by a comment, including delimiters.
929 // Good: "/* /* inner */ outer */ code();"
930 // Bad:  "code(); // hello\n world!"
931 pub fn find_comment_end(s: &str) -> Option<usize> {
932     let mut iter = CharClasses::new(s.char_indices());
933     for (kind, (i, _c)) in &mut iter {
934         if kind == FullCodeCharKind::Normal || kind == FullCodeCharKind::InString {
935             return Some(i);
936         }
937     }
938
939     // Handle case where the comment ends at the end of s.
940     if iter.status == CharClassesStatus::Normal {
941         Some(s.len())
942     } else {
943         None
944     }
945 }
946
947 /// Returns true if text contains any comment.
948 pub fn contains_comment(text: &str) -> bool {
949     CharClasses::new(text.chars()).any(|(kind, _)| kind.is_comment())
950 }
951
952 /// Remove trailing spaces from the specified snippet. We do not remove spaces
953 /// inside strings or comments.
954 pub fn remove_trailing_white_spaces(text: &str) -> String {
955     let mut buffer = String::with_capacity(text.len());
956     let mut space_buffer = String::with_capacity(128);
957     for (char_kind, c) in CharClasses::new(text.chars()) {
958         match c {
959             '\n' => {
960                 if char_kind == FullCodeCharKind::InString {
961                     buffer.push_str(&space_buffer);
962                 }
963                 space_buffer.clear();
964                 buffer.push('\n');
965             }
966             _ if c.is_whitespace() => {
967                 space_buffer.push(c);
968             }
969             _ => {
970                 if !space_buffer.is_empty() {
971                     buffer.push_str(&space_buffer);
972                     space_buffer.clear();
973                 }
974                 buffer.push(c);
975             }
976         }
977     }
978     buffer
979 }
980
981 pub struct CharClasses<T>
982 where
983     T: Iterator,
984     T::Item: RichChar,
985 {
986     base: MultiPeek<T>,
987     status: CharClassesStatus,
988 }
989
990 pub trait RichChar {
991     fn get_char(&self) -> char;
992 }
993
994 impl RichChar for char {
995     fn get_char(&self) -> char {
996         *self
997     }
998 }
999
1000 impl RichChar for (usize, char) {
1001     fn get_char(&self) -> char {
1002         self.1
1003     }
1004 }
1005
1006 #[derive(PartialEq, Eq, Debug, Clone, Copy)]
1007 enum CharClassesStatus {
1008     Normal,
1009     LitString,
1010     LitStringEscape,
1011     LitRawString(u32),
1012     RawStringPrefix(u32),
1013     RawStringSuffix(u32),
1014     LitChar,
1015     LitCharEscape,
1016     // The u32 is the nesting deepness of the comment
1017     BlockComment(u32),
1018     // Status when the '/' has been consumed, but not yet the '*', deepness is
1019     // the new deepness (after the comment opening).
1020     BlockCommentOpening(u32),
1021     // Status when the '*' has been consumed, but not yet the '/', deepness is
1022     // the new deepness (after the comment closing).
1023     BlockCommentClosing(u32),
1024     LineComment,
1025 }
1026
1027 /// Distinguish between functional part of code and comments
1028 #[derive(PartialEq, Eq, Debug, Clone, Copy)]
1029 pub enum CodeCharKind {
1030     Normal,
1031     Comment,
1032 }
1033
1034 /// Distinguish between functional part of code and comments,
1035 /// describing opening and closing of comments for ease when chunking
1036 /// code from tagged characters
1037 #[derive(PartialEq, Eq, Debug, Clone, Copy)]
1038 pub enum FullCodeCharKind {
1039     Normal,
1040     /// The first character of a comment, there is only one for a comment (always '/')
1041     StartComment,
1042     /// Any character inside a comment including the second character of comment
1043     /// marks ("//", "/*")
1044     InComment,
1045     /// Last character of a comment, '\n' for a line comment, '/' for a block comment.
1046     EndComment,
1047     /// Start of a mutlitine string
1048     StartString,
1049     /// End of a mutlitine string
1050     EndString,
1051     /// Inside a string.
1052     InString,
1053 }
1054
1055 impl FullCodeCharKind {
1056     pub fn is_comment(self) -> bool {
1057         match self {
1058             FullCodeCharKind::StartComment
1059             | FullCodeCharKind::InComment
1060             | FullCodeCharKind::EndComment => true,
1061             _ => false,
1062         }
1063     }
1064
1065     pub fn is_string(self) -> bool {
1066         self == FullCodeCharKind::InString || self == FullCodeCharKind::StartString
1067     }
1068
1069     fn to_codecharkind(self) -> CodeCharKind {
1070         if self.is_comment() {
1071             CodeCharKind::Comment
1072         } else {
1073             CodeCharKind::Normal
1074         }
1075     }
1076 }
1077
1078 impl<T> CharClasses<T>
1079 where
1080     T: Iterator,
1081     T::Item: RichChar,
1082 {
1083     pub fn new(base: T) -> CharClasses<T> {
1084         CharClasses {
1085             base: multipeek(base),
1086             status: CharClassesStatus::Normal,
1087         }
1088     }
1089 }
1090
1091 fn is_raw_string_suffix<T>(iter: &mut MultiPeek<T>, count: u32) -> bool
1092 where
1093     T: Iterator,
1094     T::Item: RichChar,
1095 {
1096     for _ in 0..count {
1097         match iter.peek() {
1098             Some(c) if c.get_char() == '#' => continue,
1099             _ => return false,
1100         }
1101     }
1102     true
1103 }
1104
1105 impl<T> Iterator for CharClasses<T>
1106 where
1107     T: Iterator,
1108     T::Item: RichChar,
1109 {
1110     type Item = (FullCodeCharKind, T::Item);
1111
1112     fn next(&mut self) -> Option<(FullCodeCharKind, T::Item)> {
1113         let item = self.base.next()?;
1114         let chr = item.get_char();
1115         let mut char_kind = FullCodeCharKind::Normal;
1116         self.status = match self.status {
1117             CharClassesStatus::LitRawString(sharps) => {
1118                 char_kind = FullCodeCharKind::InString;
1119                 match chr {
1120                     '"' => {
1121                         if sharps == 0 {
1122                             char_kind = FullCodeCharKind::Normal;
1123                             CharClassesStatus::Normal
1124                         } else if is_raw_string_suffix(&mut self.base, sharps) {
1125                             CharClassesStatus::RawStringSuffix(sharps)
1126                         } else {
1127                             CharClassesStatus::LitRawString(sharps)
1128                         }
1129                     }
1130                     _ => CharClassesStatus::LitRawString(sharps),
1131                 }
1132             }
1133             CharClassesStatus::RawStringPrefix(sharps) => {
1134                 char_kind = FullCodeCharKind::InString;
1135                 match chr {
1136                     '#' => CharClassesStatus::RawStringPrefix(sharps + 1),
1137                     '"' => CharClassesStatus::LitRawString(sharps),
1138                     _ => CharClassesStatus::Normal, // Unreachable.
1139                 }
1140             }
1141             CharClassesStatus::RawStringSuffix(sharps) => {
1142                 match chr {
1143                     '#' => {
1144                         if sharps == 1 {
1145                             CharClassesStatus::Normal
1146                         } else {
1147                             char_kind = FullCodeCharKind::InString;
1148                             CharClassesStatus::RawStringSuffix(sharps - 1)
1149                         }
1150                     }
1151                     _ => CharClassesStatus::Normal, // Unreachable
1152                 }
1153             }
1154             CharClassesStatus::LitString => {
1155                 char_kind = FullCodeCharKind::InString;
1156                 match chr {
1157                     '"' => CharClassesStatus::Normal,
1158                     '\\' => CharClassesStatus::LitStringEscape,
1159                     _ => CharClassesStatus::LitString,
1160                 }
1161             }
1162             CharClassesStatus::LitStringEscape => {
1163                 char_kind = FullCodeCharKind::InString;
1164                 CharClassesStatus::LitString
1165             }
1166             CharClassesStatus::LitChar => match chr {
1167                 '\\' => CharClassesStatus::LitCharEscape,
1168                 '\'' => CharClassesStatus::Normal,
1169                 _ => CharClassesStatus::LitChar,
1170             },
1171             CharClassesStatus::LitCharEscape => CharClassesStatus::LitChar,
1172             CharClassesStatus::Normal => match chr {
1173                 'r' => match self.base.peek().map(|c| c.get_char()) {
1174                     Some('#') | Some('"') => {
1175                         char_kind = FullCodeCharKind::InString;
1176                         CharClassesStatus::RawStringPrefix(0)
1177                     }
1178                     _ => CharClassesStatus::Normal,
1179                 },
1180                 '"' => {
1181                     char_kind = FullCodeCharKind::InString;
1182                     CharClassesStatus::LitString
1183                 }
1184                 '\'' => {
1185                     // HACK: Work around mut borrow.
1186                     match self.base.peek() {
1187                         Some(next) if next.get_char() == '\\' => {
1188                             self.status = CharClassesStatus::LitChar;
1189                             return Some((char_kind, item));
1190                         }
1191                         _ => (),
1192                     }
1193
1194                     match self.base.peek() {
1195                         Some(next) if next.get_char() == '\'' => CharClassesStatus::LitChar,
1196                         _ => CharClassesStatus::Normal,
1197                     }
1198                 }
1199                 '/' => match self.base.peek() {
1200                     Some(next) if next.get_char() == '*' => {
1201                         self.status = CharClassesStatus::BlockCommentOpening(1);
1202                         return Some((FullCodeCharKind::StartComment, item));
1203                     }
1204                     Some(next) if next.get_char() == '/' => {
1205                         self.status = CharClassesStatus::LineComment;
1206                         return Some((FullCodeCharKind::StartComment, item));
1207                     }
1208                     _ => CharClassesStatus::Normal,
1209                 },
1210                 _ => CharClassesStatus::Normal,
1211             },
1212             CharClassesStatus::BlockComment(deepness) => {
1213                 assert_ne!(deepness, 0);
1214                 self.status = match self.base.peek() {
1215                     Some(next) if next.get_char() == '/' && chr == '*' => {
1216                         CharClassesStatus::BlockCommentClosing(deepness - 1)
1217                     }
1218                     Some(next) if next.get_char() == '*' && chr == '/' => {
1219                         CharClassesStatus::BlockCommentOpening(deepness + 1)
1220                     }
1221                     _ => CharClassesStatus::BlockComment(deepness),
1222                 };
1223                 return Some((FullCodeCharKind::InComment, item));
1224             }
1225             CharClassesStatus::BlockCommentOpening(deepness) => {
1226                 assert_eq!(chr, '*');
1227                 self.status = CharClassesStatus::BlockComment(deepness);
1228                 return Some((FullCodeCharKind::InComment, item));
1229             }
1230             CharClassesStatus::BlockCommentClosing(deepness) => {
1231                 assert_eq!(chr, '/');
1232                 if deepness == 0 {
1233                     self.status = CharClassesStatus::Normal;
1234                     return Some((FullCodeCharKind::EndComment, item));
1235                 } else {
1236                     self.status = CharClassesStatus::BlockComment(deepness);
1237                     return Some((FullCodeCharKind::InComment, item));
1238                 }
1239             }
1240             CharClassesStatus::LineComment => match chr {
1241                 '\n' => {
1242                     self.status = CharClassesStatus::Normal;
1243                     return Some((FullCodeCharKind::EndComment, item));
1244                 }
1245                 _ => {
1246                     self.status = CharClassesStatus::LineComment;
1247                     return Some((FullCodeCharKind::InComment, item));
1248                 }
1249             },
1250         };
1251         Some((char_kind, item))
1252     }
1253 }
1254
1255 /// An iterator over the lines of a string, paired with the char kind at the
1256 /// end of the line.
1257 pub struct LineClasses<'a> {
1258     base: iter::Peekable<CharClasses<std::str::Chars<'a>>>,
1259     kind: FullCodeCharKind,
1260 }
1261
1262 impl<'a> LineClasses<'a> {
1263     pub fn new(s: &'a str) -> Self {
1264         LineClasses {
1265             base: CharClasses::new(s.chars()).peekable(),
1266             kind: FullCodeCharKind::Normal,
1267         }
1268     }
1269 }
1270
1271 impl<'a> Iterator for LineClasses<'a> {
1272     type Item = (FullCodeCharKind, String);
1273
1274     fn next(&mut self) -> Option<Self::Item> {
1275         self.base.peek()?;
1276
1277         let mut line = String::new();
1278
1279         let start_class = match self.base.peek() {
1280             Some((kind, _)) => *kind,
1281             None => FullCodeCharKind::Normal,
1282         };
1283
1284         while let Some((kind, c)) = self.base.next() {
1285             if c == '\n' {
1286                 self.kind = match (start_class, kind) {
1287                     (FullCodeCharKind::Normal, FullCodeCharKind::InString) => {
1288                         FullCodeCharKind::StartString
1289                     }
1290                     (FullCodeCharKind::InString, FullCodeCharKind::Normal) => {
1291                         FullCodeCharKind::EndString
1292                     }
1293                     _ => kind,
1294                 };
1295                 break;
1296             } else {
1297                 line.push(c);
1298             }
1299         }
1300
1301         Some((self.kind, line))
1302     }
1303 }
1304
1305 /// Iterator over functional and commented parts of a string. Any part of a string is either
1306 /// functional code, either *one* block comment, either *one* line comment. Whitespace between
1307 /// comments is functional code. Line comments contain their ending newlines.
1308 struct UngroupedCommentCodeSlices<'a> {
1309     slice: &'a str,
1310     iter: iter::Peekable<CharClasses<std::str::CharIndices<'a>>>,
1311 }
1312
1313 impl<'a> UngroupedCommentCodeSlices<'a> {
1314     fn new(code: &'a str) -> UngroupedCommentCodeSlices<'a> {
1315         UngroupedCommentCodeSlices {
1316             slice: code,
1317             iter: CharClasses::new(code.char_indices()).peekable(),
1318         }
1319     }
1320 }
1321
1322 impl<'a> Iterator for UngroupedCommentCodeSlices<'a> {
1323     type Item = (CodeCharKind, usize, &'a str);
1324
1325     fn next(&mut self) -> Option<Self::Item> {
1326         let (kind, (start_idx, _)) = self.iter.next()?;
1327         match kind {
1328             FullCodeCharKind::Normal | FullCodeCharKind::InString => {
1329                 // Consume all the Normal code
1330                 while let Some(&(char_kind, _)) = self.iter.peek() {
1331                     if char_kind.is_comment() {
1332                         break;
1333                     }
1334                     let _ = self.iter.next();
1335                 }
1336             }
1337             FullCodeCharKind::StartComment => {
1338                 // Consume the whole comment
1339                 while let Some((FullCodeCharKind::InComment, (_, _))) = self.iter.next() {}
1340             }
1341             _ => panic!(),
1342         }
1343         let slice = match self.iter.peek() {
1344             Some(&(_, (end_idx, _))) => &self.slice[start_idx..end_idx],
1345             None => &self.slice[start_idx..],
1346         };
1347         Some((
1348             if kind.is_comment() {
1349                 CodeCharKind::Comment
1350             } else {
1351                 CodeCharKind::Normal
1352             },
1353             start_idx,
1354             slice,
1355         ))
1356     }
1357 }
1358
1359 /// Iterator over an alternating sequence of functional and commented parts of
1360 /// a string. The first item is always a, possibly zero length, subslice of
1361 /// functional text. Line style comments contain their ending newlines.
1362 pub struct CommentCodeSlices<'a> {
1363     slice: &'a str,
1364     last_slice_kind: CodeCharKind,
1365     last_slice_end: usize,
1366 }
1367
1368 impl<'a> CommentCodeSlices<'a> {
1369     pub fn new(slice: &'a str) -> CommentCodeSlices<'a> {
1370         CommentCodeSlices {
1371             slice,
1372             last_slice_kind: CodeCharKind::Comment,
1373             last_slice_end: 0,
1374         }
1375     }
1376 }
1377
1378 impl<'a> Iterator for CommentCodeSlices<'a> {
1379     type Item = (CodeCharKind, usize, &'a str);
1380
1381     fn next(&mut self) -> Option<Self::Item> {
1382         if self.last_slice_end == self.slice.len() {
1383             return None;
1384         }
1385
1386         let mut sub_slice_end = self.last_slice_end;
1387         let mut first_whitespace = None;
1388         let subslice = &self.slice[self.last_slice_end..];
1389         let mut iter = CharClasses::new(subslice.char_indices());
1390
1391         for (kind, (i, c)) in &mut iter {
1392             let is_comment_connector = self.last_slice_kind == CodeCharKind::Normal
1393                 && &subslice[..2] == "//"
1394                 && [' ', '\t'].contains(&c);
1395
1396             if is_comment_connector && first_whitespace.is_none() {
1397                 first_whitespace = Some(i);
1398             }
1399
1400             if kind.to_codecharkind() == self.last_slice_kind && !is_comment_connector {
1401                 let last_index = match first_whitespace {
1402                     Some(j) => j,
1403                     None => i,
1404                 };
1405                 sub_slice_end = self.last_slice_end + last_index;
1406                 break;
1407             }
1408
1409             if !is_comment_connector {
1410                 first_whitespace = None;
1411             }
1412         }
1413
1414         if let (None, true) = (iter.next(), sub_slice_end == self.last_slice_end) {
1415             // This was the last subslice.
1416             sub_slice_end = match first_whitespace {
1417                 Some(i) => self.last_slice_end + i,
1418                 None => self.slice.len(),
1419             };
1420         }
1421
1422         let kind = match self.last_slice_kind {
1423             CodeCharKind::Comment => CodeCharKind::Normal,
1424             CodeCharKind::Normal => CodeCharKind::Comment,
1425         };
1426         let res = (
1427             kind,
1428             self.last_slice_end,
1429             &self.slice[self.last_slice_end..sub_slice_end],
1430         );
1431         self.last_slice_end = sub_slice_end;
1432         self.last_slice_kind = kind;
1433
1434         Some(res)
1435     }
1436 }
1437
1438 /// Checks is `new` didn't miss any comment from `span`, if it removed any, return previous text
1439 /// (if it fits in the width/offset, else return None), else return `new`
1440 pub fn recover_comment_removed(
1441     new: String,
1442     span: Span,
1443     context: &RewriteContext,
1444 ) -> Option<String> {
1445     let snippet = context.snippet(span);
1446     if snippet != new && changed_comment_content(snippet, &new) {
1447         // We missed some comments. Warn and keep the original text.
1448         if context.config.error_on_unformatted() {
1449             context.report.append(
1450                 context.source_map.span_to_filename(span).into(),
1451                 vec![FormattingError::from_span(
1452                     span,
1453                     &context.source_map,
1454                     ErrorKind::LostComment,
1455                 )],
1456             );
1457         }
1458         Some(snippet.to_owned())
1459     } else {
1460         Some(new)
1461     }
1462 }
1463
1464 pub fn filter_normal_code(code: &str) -> String {
1465     let mut buffer = String::with_capacity(code.len());
1466     LineClasses::new(code).for_each(|(kind, line)| match kind {
1467         FullCodeCharKind::Normal
1468         | FullCodeCharKind::StartString
1469         | FullCodeCharKind::InString
1470         | FullCodeCharKind::EndString => {
1471             buffer.push_str(&line);
1472             buffer.push('\n');
1473         }
1474         _ => (),
1475     });
1476     if !code.ends_with('\n') && buffer.ends_with('\n') {
1477         buffer.pop();
1478     }
1479     buffer
1480 }
1481
1482 /// Return true if the two strings of code have the same payload of comments.
1483 /// The payload of comments is everything in the string except:
1484 ///     - actual code (not comments)
1485 ///     - comment start/end marks
1486 ///     - whitespace
1487 ///     - '*' at the beginning of lines in block comments
1488 fn changed_comment_content(orig: &str, new: &str) -> bool {
1489     // Cannot write this as a fn since we cannot return types containing closures
1490     let code_comment_content = |code| {
1491         let slices = UngroupedCommentCodeSlices::new(code);
1492         slices
1493             .filter(|&(ref kind, _, _)| *kind == CodeCharKind::Comment)
1494             .flat_map(|(_, _, s)| CommentReducer::new(s))
1495     };
1496     let res = code_comment_content(orig).ne(code_comment_content(new));
1497     debug!(
1498         "comment::changed_comment_content: {}\norig: '{}'\nnew: '{}'\nraw_old: {}\nraw_new: {}",
1499         res,
1500         orig,
1501         new,
1502         code_comment_content(orig).collect::<String>(),
1503         code_comment_content(new).collect::<String>()
1504     );
1505     res
1506 }
1507
1508 /// Iterator over the 'payload' characters of a comment.
1509 /// It skips whitespace, comment start/end marks, and '*' at the beginning of lines.
1510 /// The comment must be one comment, ie not more than one start mark (no multiple line comments,
1511 /// for example).
1512 struct CommentReducer<'a> {
1513     is_block: bool,
1514     at_start_line: bool,
1515     iter: std::str::Chars<'a>,
1516 }
1517
1518 impl<'a> CommentReducer<'a> {
1519     fn new(comment: &'a str) -> CommentReducer<'a> {
1520         let is_block = comment.starts_with("/*");
1521         let comment = remove_comment_header(comment);
1522         CommentReducer {
1523             is_block,
1524             at_start_line: false, // There are no supplementary '*' on the first line
1525             iter: comment.chars(),
1526         }
1527     }
1528 }
1529
1530 impl<'a> Iterator for CommentReducer<'a> {
1531     type Item = char;
1532
1533     fn next(&mut self) -> Option<Self::Item> {
1534         loop {
1535             let mut c = self.iter.next()?;
1536             if self.is_block && self.at_start_line {
1537                 while c.is_whitespace() {
1538                     c = self.iter.next()?;
1539                 }
1540                 // Ignore leading '*'
1541                 if c == '*' {
1542                     c = self.iter.next()?;
1543                 }
1544             } else if c == '\n' {
1545                 self.at_start_line = true;
1546             }
1547             if !c.is_whitespace() {
1548                 return Some(c);
1549             }
1550         }
1551     }
1552 }
1553
1554 fn remove_comment_header(comment: &str) -> &str {
1555     if comment.starts_with("///") || comment.starts_with("//!") {
1556         &comment[3..]
1557     } else if comment.starts_with("//") {
1558         &comment[2..]
1559     } else if (comment.starts_with("/**") && !comment.starts_with("/**/"))
1560         || comment.starts_with("/*!")
1561     {
1562         &comment[3..comment.len() - 2]
1563     } else {
1564         assert!(
1565             comment.starts_with("/*"),
1566             format!("string '{}' is not a comment", comment)
1567         );
1568         &comment[2..comment.len() - 2]
1569     }
1570 }
1571
1572 #[cfg(test)]
1573 mod test {
1574     use super::*;
1575     use shape::{Indent, Shape};
1576
1577     #[test]
1578     fn char_classes() {
1579         let mut iter = CharClasses::new("//\n\n".chars());
1580
1581         assert_eq!((FullCodeCharKind::StartComment, '/'), iter.next().unwrap());
1582         assert_eq!((FullCodeCharKind::InComment, '/'), iter.next().unwrap());
1583         assert_eq!((FullCodeCharKind::EndComment, '\n'), iter.next().unwrap());
1584         assert_eq!((FullCodeCharKind::Normal, '\n'), iter.next().unwrap());
1585         assert_eq!(None, iter.next());
1586     }
1587
1588     #[test]
1589     fn comment_code_slices() {
1590         let input = "code(); /* test */ 1 + 1";
1591         let mut iter = CommentCodeSlices::new(input);
1592
1593         assert_eq!((CodeCharKind::Normal, 0, "code(); "), iter.next().unwrap());
1594         assert_eq!(
1595             (CodeCharKind::Comment, 8, "/* test */"),
1596             iter.next().unwrap()
1597         );
1598         assert_eq!((CodeCharKind::Normal, 18, " 1 + 1"), iter.next().unwrap());
1599         assert_eq!(None, iter.next());
1600     }
1601
1602     #[test]
1603     fn comment_code_slices_two() {
1604         let input = "// comment\n    test();";
1605         let mut iter = CommentCodeSlices::new(input);
1606
1607         assert_eq!((CodeCharKind::Normal, 0, ""), iter.next().unwrap());
1608         assert_eq!(
1609             (CodeCharKind::Comment, 0, "// comment\n"),
1610             iter.next().unwrap()
1611         );
1612         assert_eq!(
1613             (CodeCharKind::Normal, 11, "    test();"),
1614             iter.next().unwrap()
1615         );
1616         assert_eq!(None, iter.next());
1617     }
1618
1619     #[test]
1620     fn comment_code_slices_three() {
1621         let input = "1 // comment\n    // comment2\n\n";
1622         let mut iter = CommentCodeSlices::new(input);
1623
1624         assert_eq!((CodeCharKind::Normal, 0, "1 "), iter.next().unwrap());
1625         assert_eq!(
1626             (CodeCharKind::Comment, 2, "// comment\n    // comment2\n"),
1627             iter.next().unwrap()
1628         );
1629         assert_eq!((CodeCharKind::Normal, 29, "\n"), iter.next().unwrap());
1630         assert_eq!(None, iter.next());
1631     }
1632
1633     #[test]
1634     #[rustfmt::skip]
1635     fn format_doc_comments() {
1636         let mut wrap_normalize_config: ::config::Config = Default::default();
1637         wrap_normalize_config.set().wrap_comments(true);
1638         wrap_normalize_config.set().normalize_comments(true);
1639
1640         let mut wrap_config: ::config::Config = Default::default();
1641         wrap_config.set().wrap_comments(true);
1642
1643         let comment = rewrite_comment(" //test",
1644                                       true,
1645                                       Shape::legacy(100, Indent::new(0, 100)),
1646                                       &wrap_normalize_config).unwrap();
1647         assert_eq!("/* test */", comment);
1648
1649         let comment = rewrite_comment("// comment on a",
1650                                       false,
1651                                       Shape::legacy(10, Indent::empty()),
1652                                       &wrap_normalize_config).unwrap();
1653         assert_eq!("// comment\n// on a", comment);
1654
1655         let comment = rewrite_comment("//  A multi line comment\n             // between args.",
1656                                       false,
1657                                       Shape::legacy(60, Indent::new(0, 12)),
1658                                       &wrap_normalize_config).unwrap();
1659         assert_eq!("//  A multi line comment\n            // between args.", comment);
1660
1661         let input = "// comment";
1662         let expected =
1663             "/* comment */";
1664         let comment = rewrite_comment(input,
1665                                       true,
1666                                       Shape::legacy(9, Indent::new(0, 69)),
1667                                       &wrap_normalize_config).unwrap();
1668         assert_eq!(expected, comment);
1669
1670         let comment = rewrite_comment("/*   trimmed    */",
1671                                       true,
1672                                       Shape::legacy(100, Indent::new(0, 100)),
1673                                       &wrap_normalize_config).unwrap();
1674         assert_eq!("/* trimmed */", comment);
1675
1676         // check that different comment style are properly recognised
1677         let comment = rewrite_comment(r#"/// test1
1678                                          /// test2
1679                                          /*
1680                                           * test3
1681                                           */"#,
1682                                       false,
1683                                       Shape::legacy(100, Indent::new(0, 0)),
1684                                       &wrap_normalize_config).unwrap();
1685         assert_eq!("/// test1\n/// test2\n// test3", comment);
1686
1687         // check that the blank line marks the end of a commented paragraph
1688         let comment = rewrite_comment(r#"// test1
1689
1690                                          // test2"#,
1691                                       false,
1692                                       Shape::legacy(100, Indent::new(0, 0)),
1693                                       &wrap_normalize_config).unwrap();
1694         assert_eq!("// test1\n\n// test2", comment);
1695
1696         // check that the blank line marks the end of a custom-commented paragraph
1697         let comment = rewrite_comment(r#"//@ test1
1698
1699                                          //@ test2"#,
1700                                       false,
1701                                       Shape::legacy(100, Indent::new(0, 0)),
1702                                       &wrap_normalize_config).unwrap();
1703         assert_eq!("//@ test1\n\n//@ test2", comment);
1704
1705         // check that bare lines are just indented but left unchanged otherwise
1706         let comment = rewrite_comment(r#"// test1
1707                                          /*
1708                                            a bare line!
1709
1710                                                 another bare line!
1711                                           */"#,
1712                                       false,
1713                                       Shape::legacy(100, Indent::new(0, 0)),
1714                                       &wrap_config).unwrap();
1715         assert_eq!("// test1\n/*\n a bare line!\n\n      another bare line!\n*/", comment);
1716     }
1717
1718     // This is probably intended to be a non-test fn, but it is not used. I'm
1719     // keeping it around unless it helps us test stuff.
1720     fn uncommented(text: &str) -> String {
1721         CharClasses::new(text.chars())
1722             .filter_map(|(s, c)| match s {
1723                 FullCodeCharKind::Normal | FullCodeCharKind::InString => Some(c),
1724                 _ => None,
1725             })
1726             .collect()
1727     }
1728
1729     #[test]
1730     fn test_uncommented() {
1731         assert_eq!(&uncommented("abc/*...*/"), "abc");
1732         assert_eq!(
1733             &uncommented("// .... /* \n../* /* *** / */ */a/* // */c\n"),
1734             "..ac\n"
1735         );
1736         assert_eq!(&uncommented("abc \" /* */\" qsdf"), "abc \" /* */\" qsdf");
1737     }
1738
1739     #[test]
1740     fn test_contains_comment() {
1741         assert_eq!(contains_comment("abc"), false);
1742         assert_eq!(contains_comment("abc // qsdf"), true);
1743         assert_eq!(contains_comment("abc /* kqsdf"), true);
1744         assert_eq!(contains_comment("abc \" /* */\" qsdf"), false);
1745     }
1746
1747     #[test]
1748     fn test_find_uncommented() {
1749         fn check(haystack: &str, needle: &str, expected: Option<usize>) {
1750             assert_eq!(expected, haystack.find_uncommented(needle));
1751         }
1752
1753         check("/*/ */test", "test", Some(6));
1754         check("//test\ntest", "test", Some(7));
1755         check("/* comment only */", "whatever", None);
1756         check(
1757             "/* comment */ some text /* more commentary */ result",
1758             "result",
1759             Some(46),
1760         );
1761         check("sup // sup", "p", Some(2));
1762         check("sup", "x", None);
1763         check(r#"π? /**/ π is nice!"#, r#"π is nice"#, Some(9));
1764         check("/*sup yo? \n sup*/ sup", "p", Some(20));
1765         check("hel/*lohello*/lo", "hello", None);
1766         check("acb", "ab", None);
1767         check(",/*A*/ ", ",", Some(0));
1768         check("abc", "abc", Some(0));
1769         check("/* abc */", "abc", None);
1770         check("/**/abc/* */", "abc", Some(4));
1771         check("\"/* abc */\"", "abc", Some(4));
1772         check("\"/* abc", "abc", Some(4));
1773     }
1774
1775     #[test]
1776     fn test_remove_trailing_white_spaces() {
1777         let s = "    r#\"\n        test\n    \"#";
1778         assert_eq!(remove_trailing_white_spaces(&s), s);
1779     }
1780
1781     #[test]
1782     fn test_filter_normal_code() {
1783         let s = r#"
1784 fn main() {
1785     println!("hello, world");
1786 }
1787 "#;
1788         assert_eq!(s, filter_normal_code(s));
1789         let s_with_comment = r#"
1790 fn main() {
1791     // hello, world
1792     println!("hello, world");
1793 }
1794 "#;
1795         assert_eq!(s, filter_normal_code(s_with_comment));
1796     }
1797 }