Performance of tokenizing CSS in PHP

Question

This is a noob question from someone who hasn't written a parser/lexer ever before.

I'm writing a tokenizer/parser for CSS in PHP (please don't repeat with 'OMG, why in PHP?'). The syntax is written down by the W3C neatly here (CSS2.1) and here (CSS3, draft).

It's a list of 21 possible tokens, that all (but two) cannot be represented as static strings.

My current approach is to loop through an array containing the 21 patterns over and over again, do an if (preg_match()) and reduce the source string match by match. In principle this works really good. However, for a 1000 lines CSS string this takes something between 2 and 8 seconds, which is too much for my project.

Now I'm banging my head how other parsers tokenize and parse CSS in fractions of seconds. OK, C is always faster than PHP, but nonetheless, are there any obvious D'Oh! s that I fell into?

I made some optimizations, like checking for '@', '#' or '"' as the first char of the remaining string and applying only the relevant regexp then, but this hadn't brought any great performance boosts.

My code (snippet) so far:

$TOKENS = array(
  'IDENT' => '...regexp...',
  'ATKEYWORD' => '@...regexp...',
  'String' => '"...regexp..."|\'...regexp...\'',
  //...
);

$string = '...CSS source string...';
$stream = array();

// we reduce $string token by token
while ($string != '') {
    $string = ltrim($string, " \t\r\n\f"); // unconsumed whitespace at the
        // start is insignificant but doing a trim reduces exec time by 25%
    $matches = array();
    // loop through all possible tokens
    foreach ($TOKENS as $t => $p) {
        // The '&' is used as delimiter, because it isn't used anywhere in
        // the token regexps
        if (preg_match('&^'.$p.'&Su', $string, $matches)) {
            $stream[] = array($t, $matches[0]);
            $string = substr($string, strlen($matches[0]));
            // Yay! We found one that matches!
            continue 2;
        }
    }
    // if we come here, we have a syntax error and handle it somehow
}

// result: an array $stream consisting of arrays with
// 0 => type of token
// 1 => token content

Answer 1

Use a lexer generator.

Answer 2

The first thing I would do would be to get rid of the preg_match(). Basic string functions such as strpos() are much faster, but I don't think you even need that. It looks like you are looking for a specific token at the front of a string with preg_match(), and then simply taking the front length of that string as a substring. You could easily accomplish this with a simple substr() instead, like this:

foreach ($TOKENS as $t => $p)
{
    $front = substr($string,0,strlen($p));
    $len = strlen($p);  //this could be pre-stored in $TOKENS
    if ($front == $p) {
        $stream[] = array($t, $string);
        $string = substr($string, $len);
        // Yay! We found one that matches!
        continue 2;
    }
}

You could further optimize that by pre-calculating the length of all your tokens and storing them in the $TOKENS array, so that you don't have to call strlen() all the time. If you sorted $TOKENS into groups by length, you could reduce the number of substr() calls further as well, as you could take a substr($string) of the current string being analyzed just once for each token length, and run through all the tokens of that length before moving on to the next group of tokens.

Answer 3

the (probably) faster (but less memory friendly) approach would be to tokenize the whole stream at once, using one big regexp with alternatives for each token, like

 preg_match_all('/
       (...string...)
       |
       (@ident)
       |
       (#ident)
       ...etc
   /x', $stream, $tokens);

 foreach($tokens as $token)...parse

Answer 4

Don't use regexp, scan character by character.

$tokens = array();
$string = "...code...";
$length = strlen($string);
$i = 0;
while ($i < $length) {
  $buf = '';
  $char = $string[$i];
  if ($char <= ord('Z') && $char >= ord('A') || $char >= ord('a') && $char <= ord('z') || $char == ord('_') || $char == ord('-')) {
    while ($char <= ord('Z') && $char >= ord('A') || $char >= ord('a') && $char <= ord('z') || $char == ord('_') || $char == ord('-')) {
      // identifier
      $buf .= $char;
      $char = $string[$i]; $i ++;
    }
    $tokens[] = array('IDENT', $buf);
  } else if (......) {
    // ......
  }
}

However, that makes the code unmaintainable, therefore, a parser generator is better.

Answer 5

It's an old post but still contributing my 2 cents on this. one thing that seriously slows down the original code in the question is the following line :

$string = substr($string, strlen($matches[0]));

instead of working on the entire string, take just a part of it (say 50 chars) which are enough for all the possible regexes. then, apply the same line of code on it. when this string shrinks below a preset length, load some more data to it.