]> git.lizzy.rs Git - rust.git/blobdiff - src/comment.rs
handle lines prefixed with a # inside code blocks
[rust.git] / src / comment.rs
index 431be847678abfb69177f19d4c00d35e3c550ae4..c946cc6e1c013e0a534b2408186a8cf84e7ca8ce 100644 (file)
 use std::{self, borrow::Cow, iter};
 
 use itertools::{multipeek, MultiPeek};
-use syntax::codemap::Span;
+use syntax::source_map::Span;
 
 use config::Config;
 use rewrite::RewriteContext;
 use shape::{Indent, Shape};
 use string::{rewrite_string, StringFormat};
 use utils::{count_newlines, first_line_width, last_line_width};
+use {ErrorKind, FormattingError};
 
 fn is_custom_comment(comment: &str) -> bool {
     if !comment.starts_with("//") {
@@ -46,11 +47,33 @@ fn custom_opener(s: &str) -> &str {
     s.lines().next().map_or("", |first_line| {
         first_line
             .find(' ')
-            .map_or(first_line, |space_index| &first_line[0..space_index + 1])
+            .map_or(first_line, |space_index| &first_line[0..=space_index])
     })
 }
 
 impl<'a> CommentStyle<'a> {
+    /// Returns true if the commenting style covers a line only.
+    pub fn is_line_comment(&self) -> bool {
+        match *self {
+            CommentStyle::DoubleSlash
+            | CommentStyle::TripleSlash
+            | CommentStyle::Doc
+            | CommentStyle::Custom(_) => true,
+            _ => false,
+        }
+    }
+
+    /// Returns true if the commenting style can span over multiple lines.
+    pub fn is_block_comment(&self) -> bool {
+        match *self {
+            CommentStyle::SingleBullet | CommentStyle::DoubleBullet | CommentStyle::Exclamation => {
+                true
+            }
+            _ => false,
+        }
+    }
+
+    /// Returns true if the commenting style is for documentation.
     pub fn is_doc_comment(&self) -> bool {
         match *self {
             CommentStyle::TripleSlash | CommentStyle::Doc => true,
@@ -95,21 +118,6 @@ pub fn line_start(&self) -> &'a str {
     pub fn to_str_tuplet(&self) -> (&'a str, &'a str, &'a str) {
         (self.opener(), self.closer(), self.line_start())
     }
-
-    pub fn line_with_same_comment_style(&self, line: &str, normalize_comments: bool) -> bool {
-        match *self {
-            CommentStyle::DoubleSlash | CommentStyle::TripleSlash | CommentStyle::Doc => {
-                line.trim_left().starts_with(self.line_start().trim_left())
-                    || comment_style(line, normalize_comments) == *self
-            }
-            CommentStyle::DoubleBullet | CommentStyle::SingleBullet | CommentStyle::Exclamation => {
-                line.trim_left().starts_with(self.closer().trim_left())
-                    || line.trim_left().starts_with(self.line_start().trim_left())
-                    || comment_style(line, normalize_comments) == *self
-            }
-            CommentStyle::Custom(opener) => line.trim_left().starts_with(opener.trim_right()),
-        }
-    }
 }
 
 fn comment_style(orig: &str, normalize_comments: bool) -> CommentStyle {
@@ -227,7 +235,7 @@ pub fn combine_strs_with_missing_comments(
 }
 
 pub fn rewrite_doc_comment(orig: &str, shape: Shape, config: &Config) -> Option<String> {
-    _rewrite_comment(orig, false, shape, config, true)
+    identify_comment(orig, false, shape, config, true)
 }
 
 pub fn rewrite_comment(
@@ -236,69 +244,247 @@ pub fn rewrite_comment(
     shape: Shape,
     config: &Config,
 ) -> Option<String> {
-    _rewrite_comment(orig, block_style, shape, config, false)
+    identify_comment(orig, block_style, shape, config, false)
 }
 
-fn _rewrite_comment(
+fn identify_comment(
     orig: &str,
     block_style: bool,
     shape: Shape,
     config: &Config,
     is_doc_comment: bool,
 ) -> Option<String> {
-    // If there are lines without a starting sigil, we won't format them correctly
-    // so in that case we won't even re-align (if !config.normalize_comments()) and
-    // we should stop now.
-    let num_bare_lines = orig.lines()
-        .map(|line| line.trim())
-        .filter(|l| !(l.starts_with('*') || l.starts_with("//") || l.starts_with("/*")))
-        .count();
-    if num_bare_lines > 0 && !config.normalize_comments() {
-        return Some(orig.to_owned());
+    let style = comment_style(orig, false);
+
+    // Computes the len of line taking into account a newline if the line is part of a paragraph.
+    fn compute_len(orig: &str, line: &str) -> usize {
+        if orig.len() > line.len() {
+            if orig.as_bytes()[line.len()] == b'\r' {
+                line.len() + 2
+            } else {
+                line.len() + 1
+            }
+        } else {
+            line.len()
+        }
     }
-    if !config.normalize_comments() && !config.wrap_comments() {
-        return light_rewrite_comment(orig, shape.indent, config, is_doc_comment);
+
+    // Get the first group of line comments having the same commenting style.
+    //
+    // Returns a tuple with:
+    // - a boolean indicating if there is a blank line
+    // - a number indicating the size of the first group of comments
+    fn consume_same_line_comments(
+        style: CommentStyle,
+        orig: &str,
+        line_start: &str,
+    ) -> (bool, usize) {
+        let mut first_group_ending = 0;
+        let mut hbl = false;
+
+        for line in orig.lines() {
+            let trimmed_line = line.trim_left();
+            if trimmed_line.is_empty() {
+                hbl = true;
+                break;
+            } else if trimmed_line.starts_with(line_start)
+                || comment_style(trimmed_line, false) == style
+            {
+                first_group_ending += compute_len(&orig[first_group_ending..], line);
+            } else {
+                break;
+            }
+        }
+        (hbl, first_group_ending)
     }
 
-    identify_comment(orig, block_style, shape, config, is_doc_comment)
+    let (has_bare_lines, first_group_ending) = match style {
+        CommentStyle::DoubleSlash | CommentStyle::TripleSlash | CommentStyle::Doc => {
+            let line_start = style.line_start().trim_left();
+            consume_same_line_comments(style, orig, line_start)
+        }
+        CommentStyle::Custom(opener) => {
+            let trimmed_opener = opener.trim_right();
+            consume_same_line_comments(style, orig, trimmed_opener)
+        }
+        // for a block comment, search for the closing symbol
+        CommentStyle::DoubleBullet | CommentStyle::SingleBullet | CommentStyle::Exclamation => {
+            let closer = style.closer().trim_left();
+            let mut closing_symbol_offset = 0;
+            let mut hbl = false;
+            for line in orig.lines() {
+                closing_symbol_offset += compute_len(&orig[closing_symbol_offset..], line);
+                let trimmed_line = line.trim_left();
+                if !trimmed_line.starts_with('*')
+                    && !trimmed_line.starts_with("//")
+                    && !trimmed_line.starts_with("/*")
+                {
+                    hbl = true;
+                }
+                if trimmed_line.ends_with(closer) {
+                    break;
+                }
+            }
+            (hbl, closing_symbol_offset)
+        }
+    };
+
+    let (first_group, rest) = orig.split_at(first_group_ending);
+    let rewritten_first_group =
+        if !config.normalize_comments() && has_bare_lines && style.is_block_comment() {
+            light_rewrite_block_comment_with_bare_lines(first_group, shape, config)?
+        } else if !config.normalize_comments()
+            && !config.wrap_comments()
+            && !config.format_doc_comments()
+        {
+            light_rewrite_comment(first_group, shape.indent, config, is_doc_comment)?
+        } else {
+            rewrite_comment_inner(
+                first_group,
+                block_style,
+                style,
+                shape,
+                config,
+                is_doc_comment || style.is_doc_comment(),
+            )?
+        };
+    if rest.is_empty() {
+        Some(rewritten_first_group)
+    } else {
+        identify_comment(rest.trim_left(), block_style, shape, config, is_doc_comment).map(
+            |rest_str| {
+                format!(
+                    "{}\n{}{}{}",
+                    rewritten_first_group,
+                    // insert back the blank line
+                    if has_bare_lines && style.is_line_comment() {
+                        "\n"
+                    } else {
+                        ""
+                    },
+                    shape.indent.to_string(config),
+                    rest_str
+                )
+            },
+        )
+    }
 }
 
-fn identify_comment(
+/// Trims a minimum of leading whitespaces so that the content layout is kept and aligns to indent.
+fn light_rewrite_block_comment_with_bare_lines(
     orig: &str,
-    block_style: bool,
     shape: Shape,
     config: &Config,
-    is_doc_comment: bool,
 ) -> Option<String> {
-    let style = comment_style(orig, false);
-    let first_group = orig.lines()
-        .take_while(|l| style.line_with_same_comment_style(l, false))
-        .collect::<Vec<_>>()
-        .join("\n");
-    let rest = orig.lines()
-        .skip(first_group.lines().count())
-        .collect::<Vec<_>>()
-        .join("\n");
-
-    let first_group_str = rewrite_comment_inner(
-        &first_group,
-        block_style,
-        style,
-        shape,
-        config,
-        is_doc_comment || style.is_doc_comment(),
-    )?;
-    if rest.is_empty() {
-        Some(first_group_str)
-    } else {
-        identify_comment(&rest, block_style, shape, config, is_doc_comment).map(|rest_str| {
-            format!(
-                "{}\n{}{}",
-                first_group_str,
-                shape.indent.to_string(config),
-                rest_str
-            )
+    let prefix_whitespace_min = orig
+        .lines()
+        // skip the line with the starting sigil since the leading whitespace is removed
+        // otherwise, the minimum would always be zero
+        .skip(1)
+        .filter(|line| !line.is_empty())
+        .map(|line| {
+            let mut width = 0;
+            for c in line.chars() {
+                match c {
+                    ' ' => width += 1,
+                    '\t' => width += config.tab_spaces(),
+                    _ => break,
+                }
+            }
+            width
         })
+        .min()?;
+
+    let indent_str = shape.indent.to_string(config);
+    let mut lines = orig.lines();
+    let first_line = lines.next()?;
+    let rest = lines
+        .map(|line| {
+            if line.is_empty() {
+                line
+            } else {
+                &line[prefix_whitespace_min..]
+            }
+        })
+        .collect::<Vec<&str>>()
+        .join(&format!("\n{}", indent_str));
+    Some(format!("{}\n{}{}", first_line, indent_str, rest))
+}
+
+/// Attributes for code blocks in rustdoc.
+/// See https://doc.rust-lang.org/rustdoc/print.html#attributes
+enum CodeBlockAttribute {
+    Rust,
+    Ignore,
+    Text,
+    ShouldPanic,
+    NoRun,
+    CompileFail,
+}
+
+impl CodeBlockAttribute {
+    fn new(attribute: &str) -> CodeBlockAttribute {
+        match attribute {
+            "rust" | "" => CodeBlockAttribute::Rust,
+            "ignore" => CodeBlockAttribute::Ignore,
+            "text" => CodeBlockAttribute::Text,
+            "should_panic" => CodeBlockAttribute::ShouldPanic,
+            "no_run" => CodeBlockAttribute::NoRun,
+            "compile_fail" => CodeBlockAttribute::CompileFail,
+            _ => CodeBlockAttribute::Text,
+        }
+    }
+}
+
+/// Block that is formatted as an item.
+///
+/// An item starts with either a star `*` or a dash `-`. Different level of indentation are
+/// handled.
+struct ItemizedBlock {
+    /// the number of whitespaces up to the item sigil
+    indent: usize,
+    /// the string that marks the start of an item
+    opener: String,
+    /// sequence of whitespaces to prefix new lines that are part of the item
+    line_start: String,
+}
+
+impl ItemizedBlock {
+    /// Returns true if the line is formatted as an item
+    fn is_itemized_line(line: &str) -> bool {
+        let trimmed = line.trim_left();
+        trimmed.starts_with("* ") || trimmed.starts_with("- ")
+    }
+
+    /// Creates a new ItemizedBlock described with the given line.
+    /// The `is_itemized_line` needs to be called first.
+    fn new(line: &str) -> ItemizedBlock {
+        let space_to_sigil = line.chars().take_while(|c| c.is_whitespace()).count();
+        let indent = space_to_sigil + 2;
+        ItemizedBlock {
+            indent,
+            opener: line[..indent].to_string(),
+            line_start: " ".repeat(indent),
+        }
+    }
+
+    /// Returns a `StringFormat` used for formatting the content of an item
+    fn create_string_format<'a>(&'a self, fmt: &'a StringFormat) -> StringFormat<'a> {
+        StringFormat {
+            opener: "",
+            closer: "",
+            line_start: "",
+            line_end: "",
+            shape: Shape::legacy(fmt.shape.width.saturating_sub(self.indent), Indent::empty()),
+            trim_end: true,
+            config: fmt.config,
+        }
+    }
+
+    /// Returns true if the line is part of the current itemized block
+    fn in_block(&self, line: &str) -> bool {
+        !ItemizedBlock::is_itemized_line(line)
+            && self.indent <= line.chars().take_while(|c| c.is_whitespace()).count()
     }
 }
 
@@ -333,7 +519,8 @@ fn rewrite_comment_inner(
     };
 
     let line_breaks = count_newlines(orig.trim_right());
-    let lines = orig.lines()
+    let lines = orig
+        .lines()
         .enumerate()
         .map(|(i, mut line)| {
             line = trim_right_unless_two_whitespaces(line.trim_left(), is_doc_comment);
@@ -360,16 +547,18 @@ fn rewrite_comment_inner(
     result.push_str(opener);
     let mut code_block_buffer = String::with_capacity(128);
     let mut is_prev_line_multi_line = false;
-    let mut inside_code_block = false;
+    let mut code_block_attr = None;
+    let mut item_block_buffer = String::with_capacity(128);
+    let mut item_block: Option<ItemizedBlock> = None;
     let comment_line_separator = format!("{}{}", indent_str, line_start);
-    let join_code_block_with_comment_line_separator = |s: &str| {
+    let join_block = |s: &str, sep: &str| {
         let mut result = String::with_capacity(s.len() + 128);
         let mut iter = s.lines().peekable();
         while let Some(line) = iter.next() {
             result.push_str(line);
             result.push_str(match iter.peek() {
-                Some(next_line) if next_line.is_empty() => comment_line_separator.trim_right(),
-                Some(..) => &comment_line_separator,
+                Some(next_line) if next_line.is_empty() => sep.trim_right(),
+                Some(..) => &sep,
                 None => "",
             });
         }
@@ -379,56 +568,93 @@ fn rewrite_comment_inner(
     for (i, (line, has_leading_whitespace)) in lines.enumerate() {
         let is_last = i == count_newlines(orig);
 
-        if inside_code_block {
+        if let Some(ref ib) = item_block {
+            if ib.in_block(&line) {
+                item_block_buffer.push_str(&line);
+                item_block_buffer.push('\n');
+                continue;
+            }
+            is_prev_line_multi_line = false;
+            fmt.shape = Shape::legacy(max_chars, fmt_indent);
+            let item_fmt = ib.create_string_format(&fmt);
+            result.push_str(&comment_line_separator);
+            result.push_str(&ib.opener);
+            match rewrite_string(&item_block_buffer.replace("\n", " "), &item_fmt) {
+                Some(s) => result.push_str(&join_block(
+                    &s,
+                    &format!("{}{}", &comment_line_separator, ib.line_start),
+                )),
+                None => result.push_str(&join_block(&item_block_buffer, &comment_line_separator)),
+            };
+            item_block_buffer.clear();
+        } else if let Some(ref attr) = code_block_attr {
             if line.starts_with("```") {
-                inside_code_block = false;
-                result.push_str(&comment_line_separator);
-                let code_block = ::format_code_block(&code_block_buffer, config)
-                    .unwrap_or_else(|| code_block_buffer.to_owned());
-                result.push_str(&join_code_block_with_comment_line_separator(&code_block));
+                let code_block = match attr {
+                    CodeBlockAttribute::Ignore | CodeBlockAttribute::Text => {
+                        trim_custom_comment_prefix(&code_block_buffer)
+                    }
+                    _ if code_block_buffer.is_empty() => String::new(),
+                    _ => {
+                        let mut config = config.clone();
+                        config.set().format_doc_comments(false);
+                        match ::format_code_block(&code_block_buffer, &config) {
+                            Some(ref s) => trim_custom_comment_prefix(s),
+                            None => trim_custom_comment_prefix(&code_block_buffer),
+                        }
+                    }
+                };
+                if !code_block.is_empty() {
+                    result.push_str(&comment_line_separator);
+                    result.push_str(&join_block(&code_block, &comment_line_separator));
+                }
                 code_block_buffer.clear();
                 result.push_str(&comment_line_separator);
                 result.push_str(line);
+                code_block_attr = None;
             } else {
-                code_block_buffer.push_str(line);
+                code_block_buffer.push_str(&hide_sharp_behind_comment(line));
                 code_block_buffer.push('\n');
-
-                if is_last {
-                    // There is an code block that is not properly enclosed by backticks.
-                    // We will leave them untouched.
-                    result.push_str(&comment_line_separator);
-                    result.push_str(&join_code_block_with_comment_line_separator(
-                        &code_block_buffer,
-                    ));
-                }
             }
+            continue;
+        }
 
+        code_block_attr = None;
+        item_block = None;
+        if line.starts_with("```") {
+            code_block_attr = Some(CodeBlockAttribute::new(&line[3..]))
+        } else if config.wrap_comments() && ItemizedBlock::is_itemized_line(&line) {
+            let ib = ItemizedBlock::new(&line);
+            item_block_buffer.push_str(&line[ib.indent..]);
+            item_block_buffer.push('\n');
+            item_block = Some(ib);
             continue;
-        } else {
-            inside_code_block = line.starts_with("```rust");
+        }
 
-            if result == opener {
-                let force_leading_whitespace = opener == "/* " && count_newlines(orig) == 0;
-                if !has_leading_whitespace && !force_leading_whitespace && result.ends_with(' ') {
-                    result.pop();
-                }
-                if line.is_empty() {
-                    continue;
-                }
-            } else if is_prev_line_multi_line && !line.is_empty() {
-                result.push(' ')
-            } else if is_last && !closer.is_empty() && line.is_empty() {
+        if result == opener {
+            let force_leading_whitespace = opener == "/* " && count_newlines(orig) == 0;
+            if !has_leading_whitespace && !force_leading_whitespace && result.ends_with(' ') {
+                result.pop();
+            }
+            if line.is_empty() {
+                continue;
+            }
+        } else if is_prev_line_multi_line && !line.is_empty() {
+            result.push(' ')
+        } else if is_last && line.is_empty() {
+            // trailing blank lines are unwanted
+            if !closer.is_empty() {
                 result.push_str(&indent_str);
-            } else {
-                result.push_str(&comment_line_separator);
-                if !has_leading_whitespace && result.ends_with(' ') {
-                    result.pop();
-                }
+            }
+            break;
+        } else {
+            result.push_str(&comment_line_separator);
+            if !has_leading_whitespace && result.ends_with(' ') {
+                result.pop();
             }
         }
 
         if config.wrap_comments() && line.len() > fmt.shape.width && !has_url(line) {
-            match rewrite_string(line, &fmt, Some(max_chars)) {
+            match rewrite_string(line, &fmt) {
                 Some(ref s) => {
                     is_prev_line_multi_line = s.contains('\n');
                     result.push_str(s);
@@ -439,7 +665,7 @@ fn rewrite_comment_inner(
                     result.pop();
                     result.push_str(&comment_line_separator);
                     fmt.shape = Shape::legacy(max_chars, fmt_indent);
-                    match rewrite_string(line, &fmt, Some(max_chars)) {
+                    match rewrite_string(line, &fmt) {
                         Some(ref s) => {
                             is_prev_line_multi_line = s.contains('\n');
                             result.push_str(s);
@@ -460,7 +686,7 @@ fn rewrite_comment_inner(
                 // 1 = " "
                 let offset = 1 + last_line_width(&result) - line_start.len();
                 Shape {
-                    width: max_chars.checked_sub(offset).unwrap_or(0),
+                    width: max_chars.saturating_sub(offset),
                     indent: fmt_indent,
                     offset: fmt.shape.offset + offset,
                 }
@@ -477,6 +703,30 @@ fn rewrite_comment_inner(
             is_prev_line_multi_line = false;
         }
     }
+    if !code_block_buffer.is_empty() {
+        // There is a code block that is not properly enclosed by backticks.
+        // We will leave them untouched.
+        result.push_str(&comment_line_separator);
+        result.push_str(&join_block(
+            &trim_custom_comment_prefix(&code_block_buffer),
+            &comment_line_separator,
+        ));
+    }
+    if !item_block_buffer.is_empty() {
+        // the last few lines are part of an itemized block
+        let ib = item_block.unwrap();
+        fmt.shape = Shape::legacy(max_chars, fmt_indent);
+        let item_fmt = ib.create_string_format(&fmt);
+        result.push_str(&comment_line_separator);
+        result.push_str(&ib.opener);
+        match rewrite_string(&item_block_buffer.replace("\n", " "), &item_fmt) {
+            Some(s) => result.push_str(&join_block(
+                &s,
+                &format!("{}{}", &comment_line_separator, ib.line_start),
+            )),
+            None => result.push_str(&join_block(&item_block_buffer, &comment_line_separator)),
+        };
+    }
 
     result.push_str(closer);
     if result.ends_with(opener) && opener.ends_with(' ') {
@@ -487,6 +737,37 @@ fn rewrite_comment_inner(
     Some(result)
 }
 
+const RUSTFMT_CUSTOM_COMMENT_PREFIX: &str = "//#### ";
+
+fn hide_sharp_behind_comment(s: &str) -> Cow<str> {
+    if s.trim_left().starts_with("# ") {
+        Cow::from(format!("{}{}", RUSTFMT_CUSTOM_COMMENT_PREFIX, s))
+    } else {
+        Cow::from(s)
+    }
+}
+
+fn trim_custom_comment_prefix(s: &str) -> String {
+    s.lines()
+        .map(|line| {
+            let left_trimmed = line.trim_left();
+            if left_trimmed.starts_with(RUSTFMT_CUSTOM_COMMENT_PREFIX) {
+                let orig = left_trimmed.trim_left_matches(RUSTFMT_CUSTOM_COMMENT_PREFIX);
+                // due to comment wrapping, a line that was originaly behind `#` is split over
+                // multiple lines, which needs then to be prefixed with a `#`
+                if !orig.trim_left().starts_with("# ") {
+                    format!("# {}", orig)
+                } else {
+                    orig.to_string()
+                }
+            } else {
+                line.to_string()
+            }
+        })
+        .collect::<Vec<_>>()
+        .join("\n")
+}
+
 /// Returns true if the given string MAY include URLs or alike.
 fn has_url(s: &str) -> bool {
     // This function may return false positive, but should get its job done in most cases.
@@ -553,7 +834,8 @@ fn light_rewrite_comment(
     config: &Config,
     is_doc_comment: bool,
 ) -> Option<String> {
-    let lines: Vec<&str> = orig.lines()
+    let lines: Vec<&str> = orig
+        .lines()
         .map(|l| {
             // This is basically just l.trim(), but in the case that a line starts
             // with `*` we want to leave one space before it, so it aligns with the
@@ -579,7 +861,9 @@ fn light_rewrite_comment(
 /// Does not trim all whitespace. If a single space is trimmed from the left of the string,
 /// this function returns true.
 fn left_trim_comment_line<'a>(line: &'a str, style: &CommentStyle) -> (&'a str, bool) {
-    if line.starts_with("//! ") || line.starts_with("/// ") || line.starts_with("/*! ")
+    if line.starts_with("//! ")
+        || line.starts_with("/// ")
+        || line.starts_with("/*! ")
         || line.starts_with("/** ")
     {
         (&line[4..], true)
@@ -589,13 +873,18 @@ fn left_trim_comment_line<'a>(line: &'a str, style: &CommentStyle) -> (&'a str,
         } else {
             (&line[opener.trim_right().len()..], false)
         }
-    } else if line.starts_with("/* ") || line.starts_with("// ") || line.starts_with("//!")
-        || line.starts_with("///") || line.starts_with("** ")
+    } else if line.starts_with("/* ")
+        || line.starts_with("// ")
+        || line.starts_with("//!")
+        || line.starts_with("///")
+        || line.starts_with("** ")
         || line.starts_with("/*!")
         || (line.starts_with("/**") && !line.starts_with("/**/"))
     {
         (&line[3..], line.chars().nth(2).unwrap() == ' ')
-    } else if line.starts_with("/*") || line.starts_with("* ") || line.starts_with("//")
+    } else if line.starts_with("/*")
+        || line.starts_with("* ")
+        || line.starts_with("//")
         || line.starts_with("**")
     {
         (&line[2..], line.chars().nth(1).unwrap() == ' ')
@@ -719,6 +1008,9 @@ enum CharClassesStatus {
     Normal,
     LitString,
     LitStringEscape,
+    LitRawString(u32),
+    RawStringPrefix(u32),
+    RawStringSuffix(u32),
     LitChar,
     LitCharEscape,
     // The u32 is the nesting deepness of the comment
@@ -752,13 +1044,17 @@ pub enum FullCodeCharKind {
     InComment,
     /// Last character of a comment, '\n' for a line comment, '/' for a block comment.
     EndComment,
+    /// Start of a mutlitine string
+    StartString,
+    /// End of a mutlitine string
+    EndString,
     /// Inside a string.
     InString,
 }
 
 impl FullCodeCharKind {
-    pub fn is_comment(&self) -> bool {
-        match *self {
+    pub fn is_comment(self) -> bool {
+        match self {
             FullCodeCharKind::StartComment
             | FullCodeCharKind::InComment
             | FullCodeCharKind::EndComment => true,
@@ -766,11 +1062,11 @@ pub fn is_comment(&self) -> bool {
         }
     }
 
-    pub fn is_string(&self) -> bool {
-        *self == FullCodeCharKind::InString
+    pub fn is_string(self) -> bool {
+        self == FullCodeCharKind::InString || self == FullCodeCharKind::StartString
     }
 
-    fn to_codecharkind(&self) -> CodeCharKind {
+    fn to_codecharkind(self) -> CodeCharKind {
         if self.is_comment() {
             CodeCharKind::Comment
         } else {
@@ -792,6 +1088,20 @@ pub fn new(base: T) -> CharClasses<T> {
     }
 }
 
+fn is_raw_string_suffix<T>(iter: &mut MultiPeek<T>, count: u32) -> bool
+where
+    T: Iterator,
+    T::Item: RichChar,
+{
+    for _ in 0..count {
+        match iter.peek() {
+            Some(c) if c.get_char() == '#' => continue,
+            _ => return false,
+        }
+    }
+    true
+}
+
 impl<T> Iterator for CharClasses<T>
 where
     T: Iterator,
@@ -804,17 +1114,51 @@ fn next(&mut self) -> Option<(FullCodeCharKind, T::Item)> {
         let chr = item.get_char();
         let mut char_kind = FullCodeCharKind::Normal;
         self.status = match self.status {
-            CharClassesStatus::LitString => match chr {
-                '"' => CharClassesStatus::Normal,
-                '\\' => {
-                    char_kind = FullCodeCharKind::InString;
-                    CharClassesStatus::LitStringEscape
+            CharClassesStatus::LitRawString(sharps) => {
+                char_kind = FullCodeCharKind::InString;
+                match chr {
+                    '"' => {
+                        if sharps == 0 {
+                            char_kind = FullCodeCharKind::Normal;
+                            CharClassesStatus::Normal
+                        } else if is_raw_string_suffix(&mut self.base, sharps) {
+                            CharClassesStatus::RawStringSuffix(sharps)
+                        } else {
+                            CharClassesStatus::LitRawString(sharps)
+                        }
+                    }
+                    _ => CharClassesStatus::LitRawString(sharps),
                 }
-                _ => {
-                    char_kind = FullCodeCharKind::InString;
-                    CharClassesStatus::LitString
+            }
+            CharClassesStatus::RawStringPrefix(sharps) => {
+                char_kind = FullCodeCharKind::InString;
+                match chr {
+                    '#' => CharClassesStatus::RawStringPrefix(sharps + 1),
+                    '"' => CharClassesStatus::LitRawString(sharps),
+                    _ => CharClassesStatus::Normal, // Unreachable.
                 }
-            },
+            }
+            CharClassesStatus::RawStringSuffix(sharps) => {
+                match chr {
+                    '#' => {
+                        if sharps == 1 {
+                            CharClassesStatus::Normal
+                        } else {
+                            char_kind = FullCodeCharKind::InString;
+                            CharClassesStatus::RawStringSuffix(sharps - 1)
+                        }
+                    }
+                    _ => CharClassesStatus::Normal, // Unreachable
+                }
+            }
+            CharClassesStatus::LitString => {
+                char_kind = FullCodeCharKind::InString;
+                match chr {
+                    '"' => CharClassesStatus::Normal,
+                    '\\' => CharClassesStatus::LitStringEscape,
+                    _ => CharClassesStatus::LitString,
+                }
+            }
             CharClassesStatus::LitStringEscape => {
                 char_kind = FullCodeCharKind::InString;
                 CharClassesStatus::LitString
@@ -826,6 +1170,13 @@ fn next(&mut self) -> Option<(FullCodeCharKind, T::Item)> {
             },
             CharClassesStatus::LitCharEscape => CharClassesStatus::LitChar,
             CharClassesStatus::Normal => match chr {
+                'r' => match self.base.peek().map(|c| c.get_char()) {
+                    Some('#') | Some('"') => {
+                        char_kind = FullCodeCharKind::InString;
+                        CharClassesStatus::RawStringPrefix(0)
+                    }
+                    _ => CharClassesStatus::Normal,
+                },
                 '"' => {
                     char_kind = FullCodeCharKind::InString;
                     CharClassesStatus::LitString
@@ -901,6 +1252,56 @@ fn next(&mut self) -> Option<(FullCodeCharKind, T::Item)> {
     }
 }
 
+/// An iterator over the lines of a string, paired with the char kind at the
+/// end of the line.
+pub struct LineClasses<'a> {
+    base: iter::Peekable<CharClasses<std::str::Chars<'a>>>,
+    kind: FullCodeCharKind,
+}
+
+impl<'a> LineClasses<'a> {
+    pub fn new(s: &'a str) -> Self {
+        LineClasses {
+            base: CharClasses::new(s.chars()).peekable(),
+            kind: FullCodeCharKind::Normal,
+        }
+    }
+}
+
+impl<'a> Iterator for LineClasses<'a> {
+    type Item = (FullCodeCharKind, String);
+
+    fn next(&mut self) -> Option<Self::Item> {
+        self.base.peek()?;
+
+        let mut line = String::new();
+
+        let start_class = match self.base.peek() {
+            Some((kind, _)) => *kind,
+            None => FullCodeCharKind::Normal,
+        };
+
+        while let Some((kind, c)) = self.base.next() {
+            if c == '\n' {
+                self.kind = match (start_class, kind) {
+                    (FullCodeCharKind::Normal, FullCodeCharKind::InString) => {
+                        FullCodeCharKind::StartString
+                    }
+                    (FullCodeCharKind::InString, FullCodeCharKind::Normal) => {
+                        FullCodeCharKind::EndString
+                    }
+                    _ => kind,
+                };
+                break;
+            } else {
+                line.push(c);
+            }
+        }
+
+        Some((self.kind, line))
+    }
+}
+
 /// Iterator over functional and commented parts of a string. Any part of a string is either
 /// functional code, either *one* block comment, either *one* line comment. Whitespace between
 /// comments is functional code. Line comments contain their ending newlines.
@@ -1043,13 +1444,41 @@ pub fn recover_comment_removed(
 ) -> Option<String> {
     let snippet = context.snippet(span);
     if snippet != new && changed_comment_content(snippet, &new) {
-        // We missed some comments. Keep the original text.
+        // We missed some comments. Warn and keep the original text.
+        if context.config.error_on_unformatted() {
+            context.report.append(
+                context.source_map.span_to_filename(span).into(),
+                vec![FormattingError::from_span(
+                    span,
+                    &context.source_map,
+                    ErrorKind::LostComment,
+                )],
+            );
+        }
         Some(snippet.to_owned())
     } else {
         Some(new)
     }
 }
 
+pub fn filter_normal_code(code: &str) -> String {
+    let mut buffer = String::with_capacity(code.len());
+    LineClasses::new(code).for_each(|(kind, line)| match kind {
+        FullCodeCharKind::Normal
+        | FullCodeCharKind::StartString
+        | FullCodeCharKind::InString
+        | FullCodeCharKind::EndString => {
+            buffer.push_str(&line);
+            buffer.push('\n');
+        }
+        _ => (),
+    });
+    if !code.ends_with('\n') && buffer.ends_with('\n') {
+        buffer.pop();
+    }
+    buffer
+}
+
 /// Return true if the two strings of code have the same payload of comments.
 /// The payload of comments is everything in the string except:
 ///     - actual code (not comments)
@@ -1100,6 +1529,7 @@ fn new(comment: &'a str) -> CommentReducer<'a> {
 
 impl<'a> Iterator for CommentReducer<'a> {
     type Item = char;
+
     fn next(&mut self) -> Option<Self::Item> {
         loop {
             let mut c = self.iter.next()?;
@@ -1141,8 +1571,7 @@ fn remove_comment_header(comment: &str) -> &str {
 
 #[cfg(test)]
 mod test {
-    use super::{contains_comment, rewrite_comment, CharClasses, CodeCharKind, CommentCodeSlices,
-                FindUncommented, FullCodeCharKind};
+    use super::*;
     use shape::{Indent, Shape};
 
     #[test]
@@ -1202,28 +1631,31 @@ fn comment_code_slices_three() {
     }
 
     #[test]
-    #[cfg_attr(rustfmt, rustfmt_skip)]
-    fn format_comments() {
-        let mut config: ::config::Config = Default::default();
-        config.set().wrap_comments(true);
-        config.set().normalize_comments(true);
+    #[rustfmt::skip]
+    fn format_doc_comments() {
+        let mut wrap_normalize_config: ::config::Config = Default::default();
+        wrap_normalize_config.set().wrap_comments(true);
+        wrap_normalize_config.set().normalize_comments(true);
+
+        let mut wrap_config: ::config::Config = Default::default();
+        wrap_config.set().wrap_comments(true);
 
         let comment = rewrite_comment(" //test",
                                       true,
                                       Shape::legacy(100, Indent::new(0, 100)),
-                                      &config).unwrap();
+                                      &wrap_normalize_config).unwrap();
         assert_eq!("/* test */", comment);
 
         let comment = rewrite_comment("// comment on a",
                                       false,
                                       Shape::legacy(10, Indent::empty()),
-                                      &config).unwrap();
+                                      &wrap_normalize_config).unwrap();
         assert_eq!("// comment\n// on a", comment);
 
         let comment = rewrite_comment("//  A multi line comment\n             // between args.",
                                       false,
                                       Shape::legacy(60, Indent::new(0, 12)),
-                                      &config).unwrap();
+                                      &wrap_normalize_config).unwrap();
         assert_eq!("//  A multi line comment\n            // between args.", comment);
 
         let input = "// comment";
@@ -1232,14 +1664,55 @@ fn format_comments() {
         let comment = rewrite_comment(input,
                                       true,
                                       Shape::legacy(9, Indent::new(0, 69)),
-                                      &config).unwrap();
+                                      &wrap_normalize_config).unwrap();
         assert_eq!(expected, comment);
 
         let comment = rewrite_comment("/*   trimmed    */",
                                       true,
                                       Shape::legacy(100, Indent::new(0, 100)),
-                                      &config).unwrap();
+                                      &wrap_normalize_config).unwrap();
         assert_eq!("/* trimmed */", comment);
+
+        // check that different comment style are properly recognised
+        let comment = rewrite_comment(r#"/// test1
+                                         /// test2
+                                         /*
+                                          * test3
+                                          */"#,
+                                      false,
+                                      Shape::legacy(100, Indent::new(0, 0)),
+                                      &wrap_normalize_config).unwrap();
+        assert_eq!("/// test1\n/// test2\n// test3", comment);
+
+        // check that the blank line marks the end of a commented paragraph
+        let comment = rewrite_comment(r#"// test1
+
+                                         // test2"#,
+                                      false,
+                                      Shape::legacy(100, Indent::new(0, 0)),
+                                      &wrap_normalize_config).unwrap();
+        assert_eq!("// test1\n\n// test2", comment);
+
+        // check that the blank line marks the end of a custom-commented paragraph
+        let comment = rewrite_comment(r#"//@ test1
+
+                                         //@ test2"#,
+                                      false,
+                                      Shape::legacy(100, Indent::new(0, 0)),
+                                      &wrap_normalize_config).unwrap();
+        assert_eq!("//@ test1\n\n//@ test2", comment);
+
+        // check that bare lines are just indented but left unchanged otherwise
+        let comment = rewrite_comment(r#"// test1
+                                         /*
+                                           a bare line!
+
+                                                another bare line!
+                                          */"#,
+                                      false,
+                                      Shape::legacy(100, Indent::new(0, 0)),
+                                      &wrap_config).unwrap();
+        assert_eq!("// test1\n/*\n a bare line!\n\n      another bare line!\n*/", comment);
     }
 
     // This is probably intended to be a non-test fn, but it is not used. I'm
@@ -1298,4 +1771,27 @@ fn check(haystack: &str, needle: &str, expected: Option<usize>) {
         check("\"/* abc */\"", "abc", Some(4));
         check("\"/* abc", "abc", Some(4));
     }
+
+    #[test]
+    fn test_remove_trailing_white_spaces() {
+        let s = "    r#\"\n        test\n    \"#";
+        assert_eq!(remove_trailing_white_spaces(&s), s);
+    }
+
+    #[test]
+    fn test_filter_normal_code() {
+        let s = r#"
+fn main() {
+    println!("hello, world");
+}
+"#;
+        assert_eq!(s, filter_normal_code(s));
+        let s_with_comment = r#"
+fn main() {
+    // hello, world
+    println!("hello, world");
+}
+"#;
+        assert_eq!(s, filter_normal_code(s_with_comment));
+    }
 }