I think the example given here using classes is convoluting things too much to demonstrate what this function does.

array_udiff() will walk through array_values($a) and array_values($b) and compare each value by using the passed in callback function.

To put it another way, array_udiff() compares $a[0] to $b[0], $b[1], $b[2], and $b[3] using the provided callback function.  If the callback returns zero for any of the comparisons then $a[0] will not be in the returned array from array_udiff().  It then compares $a[1] to $b[0], $b[1], $b[2], and $b[3].  Then, finally, $a[2] to $b[0], $b[1], $b[2], and $b[3].

For example, compare_ids($a[0], $b[0]) === -5 while compare_ids($a[1], $b[1]) === 0.  Therefore, $a[1] is not returned from array_udiff() since it is present in $b.

<?
$a = array(
        array(
                'id' => 10,
                'name' => 'John',
                'color' => 'red',
        ),
        array(
                'id' => 20,
                'name' => 'Elise',
                'color' => 'blue',
        ),
        array(
                'id' => 30,
                'name' => 'Mark',
                'color' => 'red',
        ),
);

$b = array(
        array(
                'id' => 15,
                'name' => 'Nancy',
                'color' => 'black',
        ),
        array(
                'id' => 20,
                'name' => 'Elise',
                'color' => 'blue',
        ),
        array(
                'id' => 30,
                'name' => 'Mark',
                'color' => 'red',
        ),
        array(
                'id' => 40,
                'name' => 'John',
                'color' => 'orange',
        ),
);

function compare_ids($a, $b)
{
    return ($a['id'] - $b['id']);
}
function compare_names($a, $b)
{
    return strcmp($a['name'], $b['name']);
}

$ret = array_udiff($a, $b, 'compare_ids');
var_dump($ret);

$ret = array_udiff($b, $a, 'compare_ids');
var_dump($ret);

$ret = array_udiff($a, $b, 'compare_names');
var_dump($ret);
?>

Which returns the following.

In the first return we see that $b has no entry in it with an id of 10.
<?
array(1) {
  [0]=>
  array(3) {
    ["id"]=>
    int(10)
    ["name"]=>
    string(4) "John"
    ["color"]=>
    string(3) "red"
  }
}
?>

In the second return we see that $a has no entry in it with an id of 15 or 40.
<?
array(2) {
  [0]=>
  array(3) {
    ["id"]=>
    int(15)
    ["name"]=>
    string(5) "Nancy"
    ["color"]=>
    string(5) "black"
  }
  [3]=>
  array(3) {
    ["id"]=>
    int(40)
    ["name"]=>
    string(4) "John"
    ["color"]=>
    string(6) "orange"
  }
}
?>

In third return we see that all names in $a are in $b (even though the entry in $b whose name is 'John' is different, the anonymous function is only comparing names).
<?
array(0) {
}
?>

Note that the compare function is used also internally, to order the arrays and choose which element compare against in the next round.

If your compare function is not really comparing (ie. returns 0 if elements are equals, 1 otherwise), you will receive an unexpected result.

Re: "convoluted"

I think the point being made is that array_udiff() can be used not only for comparisons between homogenous arrays, as in your example (and definitely the most common need), but it can be used to compare heterogeneous arrays, too.

Consider:

<?php
function compr_1($a, $b) {
    $aVal = is_array($a) ? $a['last_name'] : $a;
    $bVal = is_array($b) ? $b['last_name'] : $b;
    return strcasecmp($aVal, $bVal);
}

$aEmployees = array(
    array('last_name'  => 'Smith',
            'first_name' => 'Joe',
            'phone'      => '555-1000'),
    array('last_name'  => 'Doe',
            'first_name' => 'John',
            'phone'      => '555-2000'),
    array('last_name'  => 'Flagg',
            'first_name' => 'Randall',
            'phone'      => '666-1000')
    );

$aNames = array('Doe', 'Smith', 'Johnson');
    
$result = array_udiff($aEmployees, $aNames, "compr_1");

print_r($result);
?>

Allowing me to get the "employee" that's not in the name list:

Array ( [2] => Array ( [last_name] => Flagg [first_name] => Randall [phone] => 666-1000 ) )

Something interesting to note, is that the two arguments to the compare function don't correspond to array1 and array2.  That's why there has to be logic in it to handle that either of the arguments might be pointing to the more complex employee array.  (Found this out the hard way.)

It is not stated, by this function also diffs array1 to itself, removing any duplicate values...

Quick example for using array_udiff to do a multi-dimensional diff

Returns values of $arr1 that are not in $arr2

<?php
$arr1 = array( array('Bob', 42), array('Phil', 37), array('Frank', 39) );
        
$arr2 = array( array('Phil', 37), array('Mark', 45) );
        
$arr3 = array_udiff($arr1, $arr2, create_function(
    '$a,$b',
    'return strcmp( implode("", $a), implode("", $b) ); ')
    );
        
print_r($arr3);
?>

Output:

Array
(
    [0] => Array
        (
            [0] => Bob
            [1] => 42
        )
 
    [2] => Array
        (
            [0] => Frank
            [1] => 39
        )
 
)
1

Hope this helps someone

I find it that this is an ideal place to apply the spaceship operator, but it was not used in the examples. 

Here is Example#1 using the spaceship operator in the comparison function. 

    <?php
    // Arrays to compare
    $array1 = array(new stdclass, new stdclass,
                    new stdclass, new stdclass,
                   );

    $array2 = array(
                    new stdclass, new stdclass,
                   );

    // Set some properties for each object
    $array1[0]->width = 11; $array1[0]->height = 3;
    $array1[1]->width = 7;  $array1[1]->height = 1;
    $array1[2]->width = 2;  $array1[2]->height = 9;
    $array1[3]->width = 5;  $array1[3]->height = 7;

    $array2[0]->width = 7;  $array2[0]->height = 5;
    $array2[1]->width = 9;  $array2[1]->height = 2;

    function compare_by_area($a, $b) {
        $areaA = $a->width * $a->height;
        $areaB = $b->width * $b->height;
    
        return $areaA <=> $areaB;
    }

    print_r(array_udiff($array1, $array2, 'compare_by_area'));
    ?>

The output is:
    Array
    (
        [0] => stdClass Object
            (
                [width] => 11
                [height] => 3
            )
    
        [1] => stdClass Object
            (
                [width] => 7
                [height] => 1
            )
    
    )

I find it is pretty awesome you can substitute all of these lines:
    if ($areaA < $areaB) {
        return -1;
    } elseif ($areaA > $areaB) {
        return 1;
    } else {
        return 0;
    }

with just:

    return $areaA <=> $areaB;

Neat!

Very easy way of achieving a case-insensitive version of array_diff (or indeed array_diff_assoc, array_intersect or any of these types of functions which have a similar function that takes a callback function as one of their parameters):

array_udiff($array1, $array2, 'strcasecmp');

This works because strcasecmp() compares two strings case-insensitively, as compared to the array_diff() which compares two strings by using the == operator, which is case-sensitive.

Note that php does the string conversion *before* sending the values to the callback function.

This functionality is now implemented in the PEAR package PHP_Compat.

More information about using this function without upgrading your version of PHP can be found on the below link:

http://pear.php.net/package/PHP_Compat