Auto merge of #181 - servo:skip_whitespace, r=nox

Add Parser::skip_whitespace

(and use it it rule parsing to make it hopefully more efficient.)

<!-- Reviewable:start -->
---
This change is [<img src="https://reviewable.io/review_button.svg" height="34" align="absmiddle" alt="Reviewable"/>](https://reviewable.io/reviews/servo/rust-cssparser/181)
<!-- Reviewable:end -->

Author: bors-servo

Cargo.toml

@@ -1,6 +1,6 @@
 [package]
 name = "cssparser"
-version = "0.19.2"
+version = "0.19.3"
 authors = [ "Simon Sapin <simon.sapin@exyr.org>" ]
 
 description = "Rust implementation of CSS Syntax Level 3"

src/parser.rs

@@ -288,6 +288,32 @@ impl<'i: 't, 't> Parser<'i, 't> {
         }
     }
 
+    /// Advance the input until the next token that’s not whitespace or a comment.
+    #[inline]
+    pub fn skip_whitespace(&mut self) {
+        // If we just consumed `{`, `[`, `(`, or `function(`, leave whitespace
+        // or comments inside the block or function up to the nested parser.
+        if self.at_start_of.is_some() {
+            return
+        }
+
+        self.input.tokenizer.skip_whitespace()
+    }
+
+    #[inline]
+    pub(crate) fn skip_cdc_and_cdo(&mut self) {
+        self.input.tokenizer.skip_cdc_and_cdo()
+    }
+
+    #[inline]
+    pub(crate) fn next_byte(&self) -> Option<u8> {
+        let byte = self.input.tokenizer.next_byte();
+        if self.stop_before.contains(Delimiters::from_byte(byte)) {
+            return None
+        }
+        byte
+    }
+
     /// Restore the internal state of the parser (including position within the input)
     /// to what was previously saved by the `Parser::position` method.
     ///
@@ -364,14 +390,8 @@ impl<'i: 't, 't> Parser<'i, 't> {
     ///
     /// This only returns a closing token when it is unmatched (and therefore an error).
     pub fn next(&mut self) -> Result<&Token<'i>, BasicParseError<'i>> {
-        loop {
-            match self.next_including_whitespace_and_comments() {
-                Err(e) => return Err(e),
-                Ok(&Token::WhiteSpace(_)) | Ok(&Token::Comment(_)) => {},
-                _ => break
-            }
-        }
-        Ok(self.input.cached_token_ref())
+        self.skip_whitespace();
+        self.next_including_whitespace_and_comments()
     }
 
     /// Same as `Parser::next`, but does not skip whitespace tokens.
@@ -459,6 +479,7 @@ impl<'i: 't, 't> Parser<'i, 't> {
     where F: for<'tt> FnMut(&mut Parser<'i, 'tt>) -> Result<T, ParseError<'i, E>> {
         let mut values = vec![];
         loop {
+            self.skip_whitespace();  // Unnecessary for correctness, but may help try() in parse_one rewind less.
             values.push(self.parse_until_before(Delimiter::Comma, &mut parse_one)?);
             match self.next() {
                 Err(_) => return Ok(values),

src/rules_and_declarations.rs

@@ -342,15 +342,29 @@ where P: QualifiedRuleParser<'i, QualifiedRule = R, Error = E> +
 
     fn next(&mut self) -> Option<Result<R, PreciseParseError<'i, E>>> {
         loop {
+            if self.is_stylesheet {
+                self.input.skip_cdc_and_cdo()
+            } else {
+                self.input.skip_whitespace()
+            }
             let start = self.input.state();
-            // FIXME: remove intermediate variable when lifetimes are non-lexical
-            let at_keyword = match self.input.next_including_whitespace_and_comments() {
-                Ok(&Token::WhiteSpace(_)) | Ok(&Token::Comment(_)) => continue,
-                Ok(&Token::CDO) | Ok(&Token::CDC) if self.is_stylesheet => continue,
-                Ok(&Token::AtKeyword(ref name)) => Some(name.clone()),
-                Ok(_) => None,
-                Err(_) => return None,
-            };
+
+            let at_keyword;
+            match self.input.next_byte() {
+                Some(b'@') => {
+                    match self.input.next_including_whitespace_and_comments() {
+                        Ok(&Token::AtKeyword(ref name)) => at_keyword = Some(name.clone()),
+                        _ => at_keyword = None,
+                    }
+                    // FIXME: move this back inside `match` when lifetimes are non-lexical
+                    if at_keyword.is_none() {
+                        self.input.reset(&start)
+                    }
+                }
+                Some(_) => at_keyword = None,
+                None => return None
+            }
+
             if let Some(name) = at_keyword {
                 let first_stylesheet_rule = self.is_stylesheet && !self.any_rule_so_far;
                 self.any_rule_so_far = true;
@@ -362,7 +376,6 @@ where P: QualifiedRuleParser<'i, QualifiedRule = R, Error = E> +
                 }
             } else {
                 self.any_rule_so_far = true;
-                self.input.reset(&start);
                 return Some(parse_qualified_rule(self.input, &mut self.parser)
                             .map_err(|e| PreciseParseError {
                                 error: e,
@@ -400,20 +413,27 @@ pub fn parse_one_rule<'i, 't, R, P, E>(input: &mut Parser<'i, 't>, parser: &mut
 where P: QualifiedRuleParser<'i, QualifiedRule = R, Error = E> +
          AtRuleParser<'i, AtRule = R, Error = E> {
     input.parse_entirely(|input| {
-        loop {
-            let start = input.state();
-            // FIXME: remove intermediate variable when lifetimes are non-lexical
-            let at_keyword = match *input.next_including_whitespace_and_comments()? {
-                Token::WhiteSpace(_) | Token::Comment(_) => continue,
-                Token::AtKeyword(ref name) => Some(name.clone()),
-                _ => None
-            };
-            if let Some(name) = at_keyword {
-                return parse_at_rule(&start, name, input, parser).map_err(|e| e.error)
-            } else {
-                input.reset(&start);
-                return parse_qualified_rule(input, parser)
+        input.skip_whitespace();
+        let start = input.state();
+
+        let at_keyword;
+        if input.next_byte() == Some(b'@') {
+            match *input.next_including_whitespace_and_comments()? {
+                Token::AtKeyword(ref name) => at_keyword = Some(name.clone()),
+                _ => at_keyword = None,
+            }
+            // FIXME: move this back inside `match` when lifetimes are non-lexical
+            if at_keyword.is_none() {
+                input.reset(&start)
             }
+        } else {
+            at_keyword = None
+        }
+
+        if let Some(name) = at_keyword {
+            parse_at_rule(&start, name, input, parser).map_err(|e| e.error)
+        } else {
+            parse_qualified_rule(input, parser)
         }
     })
 }

src/tokenizer.rs

@@ -411,6 +411,76 @@ impl<'a> Tokenizer<'a> {
     fn starts_with(&self, needle: &[u8]) -> bool {
         self.input.as_bytes()[self.position..].starts_with(needle)
     }
+
+    pub fn skip_whitespace(&mut self) {
+        while !self.is_eof() {
+            match_byte! { self.next_byte_unchecked(),
+                b' ' | b'\t' => {
+                    self.advance(1)
+                },
+                b'\n' | b'\x0C' => {
+                    self.advance(1);
+                    self.seen_newline(false);
+                },
+                b'\r' => {
+                    self.advance(1);
+                    self.seen_newline(true);
+                },
+                b'/' => {
+                    if self.starts_with(b"/*") {
+                        consume_comment(self);
+                    } else {
+                        return
+                    }
+                }
+                _ => {
+                    return
+                }
+            }
+        }
+    }
+
+    pub fn skip_cdc_and_cdo(&mut self) {
+        while !self.is_eof() {
+            match_byte! { self.next_byte_unchecked(),
+                b' ' | b'\t' => {
+                    self.advance(1)
+                },
+                b'\n' | b'\x0C' => {
+                    self.advance(1);
+                    self.seen_newline(false);
+                },
+                b'\r' => {
+                    self.advance(1);
+                    self.seen_newline(true);
+                },
+                b'/' => {
+                    if self.starts_with(b"/*") {
+                        consume_comment(self);
+                    } else {
+                        return
+                    }
+                }
+                b'<' => {
+                    if self.starts_with(b"<!--") {
+                        self.advance(4)
+                    } else {
+                        return
+                    }
+                }
+                b'-' => {
+                    if self.starts_with(b"-->") {
+                        self.advance(4)
+                    } else {
+                        return
+                    }
+                }
+                _ => {
+                    return
+                }
+            }
+        }
+    }
 }
 
 /// A position from the start of the input, counted in UTF-8 bytes.
@@ -514,9 +584,7 @@ fn next_token<'a>(tokenizer: &mut Tokenizer<'a>) -> Result<Token<'a>, ()> {
         }
         b'/' => {
             if tokenizer.starts_with(b"/*") {
-                let contents = consume_comment(tokenizer);
-                check_for_source_map(tokenizer, contents);
-                Comment(contents)
+                Comment(consume_comment(tokenizer))
             } else {
                 tokenizer.advance(1);
                 Delim('/')
@@ -627,7 +695,9 @@ fn consume_comment<'a>(tokenizer: &mut Tokenizer<'a>) -> &'a str {
                 tokenizer.advance(1);
                 if tokenizer.next_byte() == Some(b'/') {
                     tokenizer.advance(1);
-                    return tokenizer.slice(start_position..end_position)
+                    let contents = tokenizer.slice(start_position..end_position);
+                    check_for_source_map(tokenizer, contents);
+                    return contents
                 }
             }
             b'\n' | b'\x0C' => {
@@ -643,7 +713,9 @@ fn consume_comment<'a>(tokenizer: &mut Tokenizer<'a>) -> &'a str {
             }
         }
     }
-    tokenizer.slice_from(start_position)
+    let contents = tokenizer.slice_from(start_position);
+    check_for_source_map(tokenizer, contents);
+    contents
 }
 
 fn consume_string<'a>(tokenizer: &mut Tokenizer<'a>, single_quote: bool) -> Token<'a> {