Question

I have a regular expression that Im using in php:

$word_array = preg_split(
    '/(\/|\.|-|_|=|\?|\&|html|shtml|www|php|cgi|htm|aspx|asp|index|com|net|org|%|\+)/',
    urldecode($path), NULL, PREG_SPLIT_NO_EMPTY
);

It works great. It takes a chunk of url paramaters like:

/2009/06/pagerank-update.html

and returns an array like:

array(4) {
  [0]=>
  string(4) "2009"
  [1]=>
  string(2) "06"
  [2]=>
  string(8) "pagerank"
  [3]=>
  string(6) "update"
}

The only thing I need is for it to also not return strings that are less than 3 characters. So the "06" string is garbage and I’m currently using an if statement to weed them out.

Answer 1

The magic of the split. My original assumption was technically not correct (albeit a solution easier to come to). So let’s check your split pattern:

(\/|\.|-|_|=|\?|\&|html|shtml|www|php|cgi|htm|aspx|asp|index|com|net|org|%|\+)

I re-arranged it a bit. The outer parenthesis is not necessary and I moved the single characters into a character class at the end:

 html|shtml|www|php|cgi|htm|aspx|asp|index|com|net|org|[\/._=?&%+-]

That for some sorting upfront. Let’s call this pattern the split pattern, s in short and define it.

You want to match all parts that are not of those characters from the split-at pattern and at minimum three characters.

I could achieve this with the following pattern, including support of the correct split sequences and unicode support.

$pattern    = '/
    (?(DEFINE)
        (?<s> # define subpattern which is the split pattern
            html|shtml|www|php|cgi|htm|aspx|asp|index|com|net|org|
            [\\/._=?&%+-] # a little bit optimized with a character class
        )
    )
    (?:(?&s))          # consume the subpattern (URL starts with \/)
    \K                 # capture starts here
    (?:(?!(?&s)).){3,} # ensure this is not the skip pattern, take 3 characters minimum
/ux';

Or in smaller:

$path       = '/2009/06/pagerank-update.htmltesthtmltest%C3%A4shtml';
$subject    = urldecode($path);
$pattern    = '/(?(DEFINE)(?<s>html|shtml|www|php|cgi|htm|aspx|asp|index|com|net|org|[\\/._=?&%+-]))(?:(?&s))\K(?:(?!(?&s)).){3,}/u';
$word_array = preg_match_all($pattern, $subject, $m) ? $m[0] : [];
print_r($word_array);

Result:

Array
(
    [0] => 2009
    [1] => pagerank
    [2] => update
    [3] => test
    [4] => testä
)

The same principle can be used with preg_split as well. It’s a little bit different:

$pattern = '/
    (?(DEFINE)       # define subpattern which is the split pattern
        (?<s>
    html|shtml|www|php|cgi|htm|aspx|asp|index|com|net|org|
    [\/._=?&%+-]
        )
    )
    (?:(?!(?&s)).){3,}(*SKIP)(*FAIL)       # three or more is okay
    |(?:(?!(?&s)).){1,2}(*SKIP)(*ACCEPT)   # two or one is none
    |(?&s)                                 # split @ split, at least
/ux';

Usage:

$word_array = preg_split($pattern, $subject, 0, PREG_SPLIT_NO_EMPTY);

Result:

Array
(
    [0] => 2009
    [1] => pagerank
    [2] => update
    [3] => test
    [4] => testä
)

These routines work as asked for. But this does have its price with performance. The cost is similar to the old answer.

Related questions:

• Antimatch with Regex
• Split string by delimiter, but not if it is escaped

---

Old answer, doing a two-step processing (first splitting, then filtering)

Because you are using a split routine, it will split – regardless of the length.

So what you can do is to filter the result. You can do that again with a regular expression (preg_filter), for example one that is dropping everything smaller three characters:

$word_array = preg_filter(
    '/^.{3,}$/', '$0', 
    preg_split(
        '/(\/|\.|-|_|=|\?|\&|html|shtml|www|php|cgi|htm|aspx|asp|index|com|net|org|%|\+)/',
        urldecode($path), 
        NULL, 
        PREG_SPLIT_NO_EMPTY
    )
);

Result:

Array
(
    [0] => 2009
    [2] => pagerank
    [3] => update
)