Perl-Compatible Regular Expressions

Perl has long been considered the benchmark for powerful regular expressions. PHP uses a C library called pcre to provide almost complete support for Perl’s arsenal of regular expression features. Perl regular expressions include the POSIX classes and anchors described earlier. A POSIX-style character class in a Perl regular expression works and understands non-English characters using the Unix locale system. Perl regular expressions act on arbitrary binary data, so you can safely match with patterns or strings that contain the NUL-byte (\x00).

Delimiters

Perl-style regular expressions emulate the Perl syntax for patterns, which means that each pattern must be enclosed in a pair of delimiters . Traditionally, the slash (/) character is used; for example, / pattern /. However, any non-alphanumeric character other than the backslash character (\) can be used to delimit a Perl-style pattern. This is useful when matching strings containing slashes, such as filenames. For example, the following are equivalent:

    preg_match('/\/usr\/local\//', '/usr/local/bin/perl'); // returns true
    preg_match('#/usr/local/#', '/usr/local/bin/perl');    // returns true

Parentheses (( )), curly braces ({}), square brackets ([]), and angle brackets (<>) can be used as pattern delimiters:

    preg_match('{/usr/local/}', '/usr/local/bin/perl');    // returns true

The section "Trailing Options" discusses the single-character modifiers you can put after the closing delimiter to modify the behavior of the regular expression engine. A very useful one is x, which makes the regular expression engine strip whitespace and #-marked comments from the regular expression before matching. These two patterns are the same, but one is much easier to read:

    '/([[:alpha:]]+)\s+\1/'
    '/(               # start capture
        [[:alpha:]]+  #   a word
        \s+           #   whitespace
        \1            #   the same word again
      )               # end capture
    /x'

Match Behavior

While Perl’s regular expression syntax includes the POSIX constructs we talked about earlier, some pattern components have a different meaning in Perl. In particular, Perl’s regular expressions are optimized for matching against single lines of text (although there are options that change this behavior).

The period (.) matches any character except for a newline (\n). The dollar sign ($) matches at the end of the string or, if the string ends with a newline, just before that newline:

    preg_match('/is (.*)$/', "the key is in my pants", $captured);
    // $captured[1] is 'in my pants'

Character Classes

Perl-style regular expressions support the POSIX character classes but also define some of their own, as shown in Table 4-9.

Table 4-9. Perl-style character classes

Character class

Meaning

Expansion

\s

Whitespace

[\r\n \t]

\S

Non-whitespace

[^\r\n \t]

\w

Word (identifier) character

[0-9A-Za-z_]

\W

Non-word (identifier) character

[^0-9A-Za-z_]

\d

Digit

[0-9]

\D

Non-digit

[^0-9]

Anchors

Perl-style regular expressions also support additional anchors , as listed in Table 4-10.

Table 4-10. Perl-style anchors

Assertion

Meaning

\b

Word boundary (between \w and \W or at start or end of string)

\B

Non-word boundary (between \w and \w, or \W and \W)

\A

Beginning of string

\Z

End of string or before \n at end

\z

End of string

^

Start of line (or after \n if /m flag is enabled)

$

End of line (or before \n if /m flag is enabled)

Quantifiers and Greed

The POSIX quantifiers , which Perl also supports, are always greedy . That is, when faced with a quantifier, the engine matches as much as it can while still satisfying the rest of the pattern. For instance:

    preg_match('/(<.*>)/', 'do <b>not</b> press the button', $match);
    // $match[1] is '<b>not</b>'

The regular expression matches from the first less-than sign to the last greater-than sign. In effect, the .* matches everything after the first less-than sign, and the engine backtracks to make it match less and less until finally there’s a greater-than sign to be matched.

This greediness can be a problem. Sometimes you need minimal (non-greedy) matching—that is, quantifiers that match as few times as possible to satisfy the rest of the pattern. Perl provides a parallel set of quantifiers that match minimally. They’re easy to remember, because they’re the same as the greedy quantifiers , but with a question mark (?) appended. Table 4-11 shows the corresponding greedy and non-greedy quantifiers supported by Perl-style regular expressions.

Table 4-11. Greedy and non-greedy quantifiers in Perl-compatible regular expressions

Greedy quantifier

Non-greedy quantifier

?

??

*

*?

+

+?

{m}

{m}?

{m,}

{m,}?

{m,n}

{m,n}?

Here’s how to match a tag using a non-greedy quantifier:

    preg_match('/(<.*?>)/', 'do <b>not</b> press the button', $match);
    // $match[1] is '<b>'

Another, faster way is to use a character class to match every non-greater-than character up to the next greater-than sign:

    preg_match('/(<[^>]*>)/', 'do <b>not</b> press the button', $match);
    // $match[1] is '<b>'

Non-Capturing Groups

If you enclose a part of a pattern in parentheses, the text that matches that subpattern is captured and can be accessed later. Sometimes, though, you want to create a subpattern without capturing the matching text. In Perl-compatible regular expressions, you can do this using the (?: subpattern ) construct:

    preg_match('/(?:ello)(.*)/', 'jello biafra', $match);
    // $match[1] is ' biafra'

Backreferences

You can refer to text captured earlier in a pattern with a backreference: \1 refers to the contents of the first subpattern, \2 refers to the second, and so on. If you nest subpatterns, the first begins with the first opening parenthesis, the second begins with the second opening parenthesis, and so on.

For instance, this identifies doubled words:

    preg_match('/([[:alpha:]]+)\s+\1/', 'Paris in the the spring', $m);
    // returns true and $m[1] is 'the'

You can’t capture more than 99 subpatterns.

Trailing Options

Perl-style regular expressions let you put single-letter options (flags) after the regular expression pattern to modify the interpretation, or behavior, of the match. For instance, to match case-insensitively, simply use the i flag:

    preg_match('/cat/i', 'Stop, Catherine!');         // returns true

Table 4-12 shows the modifiers from Perl that are supported in Perl-compatible regular expressions.

Table 4-12. Perl flags

Modifier

Meaning

/ regexp /i

Match case-insensitively.

/ regexp /s

Make period (.) match any character, including newline (\n).

/ regexp /x

Remove whitespace and comments from the pattern.

/ regexp /m

Make caret (^) match after, and dollar sign ($) match before, internal newlines (\n).

/ regexp /e

If the replacement string is PHP code, eval( ) it to get the actual replacement string.

PHP’s Perl-compatible regular expression functions also support other modifiers that aren’t supported by Perl, as listed in Table 4-13.

Table 4-13. Additional PHP flags

Modifier

Meaning

/ regexp /U

Reverses the greediness of the subpattern; * and + now match as little as possible, instead of as much as possible

/ regexp /u

Causes pattern strings to be treated as UTF-8

/ regexp /X

Causes a backslash followed by a character with no special meaning to emit an error

/ regexp /A

Causes the beginning of the string to be anchored as if the first character of the pattern were ^

/ regexp /D

Causes the $ character to match only at the end of a line

/ regexp /S

Causes the expression parser to more carefully examine the structure of the pattern, so it may run slightly faster the next time (such as in a loop)

It’s possible to use more than one option in a single pattern, as demonstrated in the following example:

    $message = <<< END
    To: you@youcorp
    From: me@mecorp
    Subject: pay up

    Pay me or else!
    END;
    preg_match('/^subject: (.*)/im', $message, $match);
    // $match[1] is 'pay up'

Inline Options

In addition to specifying pattern-wide options after the closing pattern delimiter, you can specify options within a pattern to have them apply only to part of the pattern. The syntax for this is:

    (?flags:subpattern)

For example, only the word “PHP” is case-insensitive in this example:

    preg_match('/I like (?i:PHP)/', 'I like pHp');          // returns true

The i, m, s, U, x, and X options can be applied internally in this fashion. You can use multiple options at once:

    preg_match('/eat (?ix:fo    o   d)/', 'eat FoOD');     // returns true

Prefix an option with a hyphen (-) to turn it off:

    preg_match('/(?-i:I like) PHP/i', 'I like pHp');       // returns true

An alternative form enables or disables the flags until the end of the enclosing subpattern or pattern:

    preg_match('/I like (?i)PHP/', 'I like pHp');          // returns true
    preg_match('/I (like (?i)PHP) a lot/', 'I like pHp a lot', $match);
    // $match[1] is 'like pHp'

Inline flags do not enable capturing. You need an additional set of capturing parentheses do that.

Lookahead and Lookbehind

In patterns it’s sometimes useful to be able to say “match here if this is next.” This is particularly common when you are splitting a string. The regular expression describes the separator, which is not returned. You can use lookahead to make sure (without matching it, thus preventing it from being returned) that there’s more data after the separator. Similarly, lookbehind checks the preceding text.

Lookahead and lookbehind come in two forms: positive and negative. A positive lookahead or lookbehind says “the next/preceding text must be like this.” A negative lookahead or lookbehind says “the next/preceding text must not be like this.” Table 4-14 shows the four constructs you can use in Perl-compatible patterns. None of the constructs captures text.

Table 4-14. Lookahead and lookbehind assertions

Construct

Meaning

(?=subpattern)

Positive lookahead

(?!subpattern)

Negative lookahead

(?<=subpattern)

Positive lookbehind

(?<!subpattern)

Negative lookbehind

A simple use of positive lookahead is splitting a Unix mbox mail file into individual messages. The word "From" starting a line by itself indicates the start of a new message, so you can split the mailbox into messages by specifying the separator as the point where the next text is "From" at the start of a line:

    $messages = preg_split('/(?=^From )/m', $mailbox);

A simple use of negative lookbehind is to extract quoted strings that contain quoted delimiters. For instance, here’s how to extract a single-quoted string (note that the regular expression is commented using the x modifier):

    $input = <<< END
    name = 'Tim O\'Reilly';
    END;

    $pattern = <<< END
    '                   # opening quote
    (                   # begin capturing
      .*?               #   the string
      (?<! \\\\ )       #   skip escaped quotes
    )                   # end capturing
    '                   # closing quote
    END;
    preg_match( "($pattern)x", $input, $match);
    echo $match[1];
    Tim O\'Reilly

The only tricky part is that to get a pattern that looks behind to see if the last character was a backslash, we need to escape the backslash to prevent the regular expression engine from seeing “\),” which would mean a literal close parenthesis. In other words, we have to backslash that backslash: “\\).” But PHP’s string-quoting rules say that \\ produces a literal single backslash, so we end up requiring four backslashes to get one through the regular expression! This is why regular expressions have a reputation for being hard to read.

Perl limits lookbehind to constant-width expressions. That is, the expressions cannot contain quantifiers, and if you use alternation, all the choices must be the same length. The Perl-compatible regular expression engine also forbids quantifiers in lookbehind, but does permit alternatives of different lengths.

Cut

The rarely used once-only subpattern, or cut , prevents worst-case behavior by the regular expression engine on some kinds of patterns. The subpattern is never backed out of once matched.

The common use for the once-only subpattern is when you have a repeated expression that may itself be repeated:

    /(a+|b+)*\.+/

This code snippet takes several seconds to report failure:

    $p = '/(a+|b+)*\.+$/';
    $s = 'abababababbabbbabbaaaaaabbbbabbababababababbba..!';
    if (preg_match($p, $s)) {
      echo "Y";
    } else {
      echo "N";
    }

This is because the regular expression engine tries all the different places to start the match, but has to backtrack out of each one, which takes time. If you know that once something is matched it should never be backed out of, you should mark it with (?> subpattern ):

    $p = '/(?>a+|b+)*\.+$/';

The cut never changes the outcome of the match; it simply makes it fail faster.

Conditional Expressions

A conditional expression is like an if statement in a regular expression. The general form is:

    (?(condition)yespattern)
    (?(condition)yespattern|nopattern)

If the assertion succeeds, the regular expression engine matches the yespattern. With the second form, if the assertion doesn’t succeed, the regular expression engine skips the yespattern and tries to match the nopattern.

The assertion can be one of two types: either a backreference, or a lookahead or lookbehind match. To reference a previously matched substring, the assertion is a number from 1-99 (the most backreferences available). The condition uses the pattern in the assertion only if the backreference was matched. If the assertion is not a backreference, it must be a positive or negative lookahead or lookbehind assertion.

Functions

There are five classes of functions that work with Perl-compatible regular expressions : matching, replacing, splitting, filtering, and a utility function for quoting text.

Matching

The preg_match( ) function performs Perl-style pattern matching on a string. It’s the equivalent of the m// operator in Perl. The preg_match( ) function takes the same arguments and gives the same return value as the ereg( ) function, except that it takes a Perl-style pattern instead of a standard pattern:

    $found = preg_match(pattern, string [, captured ]);

For example:

    preg_match('/y.*e$/', 'Sylvie');         // returns true
    preg_match('/y(.*)e$/', 'Sylvie', $m);   // $m is array('ylvie', 'lvi')

While there’s an eregi( ) function to match case-insensitively, there’s no preg_matchi( ) function. Instead, use the i flag on the pattern:

    preg_match('y.*e$/i', 'SyLvIe');         // returns true

The preg_match_all( ) function repeatedly matches from where the last match ended, until no more matches can be made:

    $found = preg_match_all(pattern, string, matches [, order ]);

The order value, either PREG_PATTERN_ORDER or PREG_SET_ORDER, determines the layout of matches. We’ll look at both, using this code as a guide:

    $string = <<< END
    13 dogs
    12 rabbits
    8 cows
    1 goat
    END;
    preg_match_all('/(\d+) (\S+)/', $string, $m1, PREG_PATTERN_ORDER);
    preg_match_all('/(\d+) (\S+)/', $string, $m2, PREG_SET_ORDER);

With PREG_PATTERN_ORDER (the default), each element of the array corresponds to a particular capturing subpattern. So $m1[0] is an array of all the substrings that matched the pattern, $m1[1] is an array of all the substrings that matched the first subpattern (the numbers), and $m1[2] is an array of all the substrings that matched the second subpattern (the words). The array $m1 has one more elements than subpatterns.

With PREG_SET_ORDER, each element of the array corresponds to the next attempt to match the whole pattern. So $m2[0] is an array of the first set of matches ('13 dogs', '13', 'dogs'), $m2[1] is an array of the second set of matches ('12 rabbits', '12', 'rabbits'), and so on. The array $m2 has as many elements as there were successful matches of the entire pattern.

Example 4-2 fetches the HTML at a particular web address into a string and extracts the URLs from that HTML. For each URL, it generates a link back to the program that will display the URLs at that address.

Example 4-2. Extracting URLs from an HTML page
<?php
 if (getenv('REQUEST_METHOD') == 'POST') {
   $url = $_POST[url];
 } else {
   $url = $_GET[url];
 }
?>

<form action="<?php echo $PHP_SELF ?>" method="POST">
URL: <input type="text" name="url" value="<?php echo $url ?>" /><br>
<input type="submit">
</form>

<?php
 if ($url) {
   $remote = fopen($url, 'r');
   $html = fread($remote, 1048576); // read up to 1 MB of HTML
   fclose($remote);

   $urls = '(http|telnet|gopher|file|wais|ftp)';
   $ltrs = '\w';
   $gunk = '/#~:.?+=&%@!\-';
   $punc = '.:?\-';
   $any = "$ltrs$gunk$punc";

   preg_match_all("{
     \b                    # start at word boundary
     $urls        :        # need resource and a colon
     [$any] +?             # followed by one or more of any valid
                           #   characters—but be conservative
                           #   and take only what you need
     (?=                   # the match ends at
       [$punc]*            # punctuation
       [^$any]             # followed by a non-URL character
      |                    # or
      $                    # the end of the string
      )
       }x", $html, $matches);
   printf("I found %d URLs<P>\n", sizeof($matches[0]));
   foreach ($matches[0] as $u) {
     $link = $PHP_SELF . '?url=' . urlencode($u);
     echo "<A HREF='$link'>$u</A><BR>\n";
   }
 }
?>

Replacing

The preg_replace( ) function behaves like the search and replace operation in your text editor. It finds all occurrences of a pattern in a string and changes those occurrences to something else:

    $new = preg_replace(pattern, replacement, subject [, limit ]);

The most common usage has all the argument strings except for the integer limit. The limit is the maximum number of occurrences of the pattern to replace (the default, and the behavior when a limit of -1 is passed, is all occurrences).

    $better = preg_replace('/<.*?>/', '!', 'do <b>not</b> press the button');
    // $better is 'do !not! press the button'

Pass an array of strings as subject to make the substitution on all of them. The new strings are returned from preg_replace( ):

    $names = array('Fred Flintstone',
                   'Barney Rubble',
                   'Wilma Flintstone',
                   'Betty Rubble');
    $tidy  = preg_replace('/(\w)\w* (\w+)/', '\1 \2', $names);
    // $tidy is array ('F Flintstone', 'B Rubble', 'W Flintstone', 'B Rubble')

To perform multiple substitutions on the same string or array of strings with one call to preg_replace( ), pass arrays of patterns and replacements:

    $contractions = array("/don't/i", "/won't/i", "/can't/i");
    $expansions = array('do not', 'will not', 'can not');
    $string = "Please don't yell—I can't jump while you won't speak";
    $longer = preg_replace($contractions, $expansions, $string);
    // $longer is 'Please do not yell—I can not jump while you will not speak';

If you give fewer replacements than patterns, text matching the extra patterns is deleted. This is a handy way to delete a lot of things at once:

    $html_gunk = array('/<.*?>/', '/&.*?;/');
    $html = '&eacute; : <b>very</b> cute';
    $stripped  = preg_replace($html_gunk, array(  ), $html);
    // $stripped is ' : very cute'

If you give an array of patterns but a single string replacement, the same replacement is used for every pattern:

    $stripped = preg_replace($html_gunk, '', $html);

The replacement can use backreferences. Unlike backreferences in patterns, though, the preferred syntax for backreferences in replacements is $1, $2, $3, etc. For example:

    echo preg_replace('/(\w)\w+\s+(\w+)/', '$2, $1.', 'Fred Flintstone')
    Flintstone, F.

The /e modifier makes preg_replace( ) treat the replacement string as PHP code that returns the actual string to use in the replacement. For example, this converts every Celsius temperature to Fahrenheit:

    $string  = 'It was 5C outside, 20C inside';
    echo preg_replace('/(\d+)C\b/e', '$1*9/5+32', $string);
    It was 41 outside, 68 inside

This more complex example expands variables in a string:

    $name = 'Fred';
    $age  = 35;
    $string = '$name is $age';
    preg_replace('/\$(\w+)/e', '$$1', $string);

Each match isolates the name of a variable ($name, $age). The $1 in the replacement refers to those names, so the PHP code actually executed is $name and $age. That code evaluates to the value of the variable, which is what’s used as the replacement. Whew!

A variation on preg_replace( ) is preg_replace_callback( ). This calls a function to get the replacement string. The function is passed an array of matches (the zeroth element is all the text that matched the pattern, the first is the contents of the first captured subpattern, and so on). For example:

    function titlecase ($s) {
      return ucfirst(strtolower($s[0]));
    }

    $string = 'goodbye cruel world';
    $new = preg_replace_callback('/\w+/', 'titlecase', $string);
    echo $new;
    Goodbye Cruel World

Splitting

Whereas you use preg_match_all( ) to extract chunks of a string when you know what those chunks are, use preg_split( ) to extract chunks when you know what separates the chunks from each other:

    $chunks = preg_split(pattern, string [, limit [, flags ]]);

The pattern matches a separator between two chunks. By default, the separators are not returned. The optional limit specifies the maximum number of chunks to return (-1 is the default, which means all chunks). The flags argument is a bitwise OR combination of the flags PREG_SPLIT_NO_EMPTY (empty chunks are not returned) and PREG_SPLIT_DELIM_CAPTURE (parts of the string captured in the pattern are returned).

For example, to extract just the operands from a simple numeric expression, use:

    $ops = preg_split('{[+*/-]}', '3+5*9/2');
    // $ops is array('3', '5', '9', '2')

To extract the operands and the operators, use:

    $ops = preg_split('{([+*/-])}', '3+5*9/2', -1, PREG_SPLIT_DELIM_CAPTURE);
    // $ops is array('3', '+', '5', '*', '9', '/', '2')

An empty pattern matches at every boundary between characters in the string. This lets you split a string into an array of characters:

    $array = preg_split('//', $string);

Filtering an array with a regular expression

The preg_grep( ) function returns those elements of an array that match a given pattern:

    $matching = preg_grep(pattern, array);

For instance, to get only the filenames that end in .txt, use:

    $textfiles = preg_grep('/\.txt$/', $filenames);

Quoting for regular expressions

The preg_quote( ) function creates a regular expression that matches only a given string:

    $re = preg_quote(string [, delimiter ]);

Every character in string that has special meaning inside a regular expression (e.g., * or $) is prefaced with a backslash:

    echo preg_quote('$5.00 (five bucks)');
    \$5\.00 \(five bucks\)

The optional second argument is an extra character to be quoted. Usually, you pass your regular expression delimiter here:

    $to_find = '/usr/local/etc/rsync.conf';
    $re = preg_quote($filename, '/');
    if (preg_match("/$re", $filename)) {
      // found it!
    }

Differences from Perl Regular Expressions

Although very similar, PHP’s implementation of Perl-style regular expressions has a few minor differences from actual Perl regular expressions:

  • The null character (ASCII 0) is not allowed as a literal character within a pattern string. You can reference it in other ways, however (\000, \x00, etc.).

  • The \E, \G, \L, \l, \Q, \u, and \U options are not supported.

  • The (?{ some perl code }) construct is not supported.

  • The /D, /G, /U, /u, /A, and /X modifiers are supported.

  • The vertical tab \v counts as a whitespace character.

  • Lookahead and lookbehind assertions cannot be repeated using *, +, or ?.

  • Parenthesized submatches within negative assertions are not remembered.

  • Alternation branches within a lookbehind assertion can be of different lengths.

Get Programming PHP, 2nd Edition now with the O’Reilly learning platform.

O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.