fmod() does not mirror a calculator's mod function. For example, fmod(.25, .05) will return .05 instead of 0 due to floor(). Using the aforementioned example, you may get 0 by replacing floor() with round() in a custom fmod().

<?
function fmod_round($x, $y) {
    $i = round($x / $y);
    return $x - $i * $y;
}

var_dump(fmod(.25, .05)); // float(0.05)
var_dump(fmod_round(.25, .05)); // float(0)
?>

Note that fmod does not behave like a similar function written in PHP itself does due to the lack of fixing floating point representation errors.

Have a look at this:
<?php
var_dump(10 / (10 / 3) === 3.0); // bool(true)
var_dump(fmod(10, 10 / 3)); // float(3.3333333333333)
var_dump(fmod(10, 10 / 3) < 10 / 3); // bool(true)
?>

Internally there is no way of exactly representing the result of 10 / 3, so it will always be a bit above or below the actual result. In this case, the example proves it being a bit above the actual result.

PHP seems quite good at auto-fixing floating point representation errors so they behave like the user would expect it. That's why the first line yields true, although the result is slightly below 3 (like 2.9999999999[something]). I failed to trick PHP into rounding or cropping the result to 2.

However, fmod seems to not apply these fixes during calculations. From 10 / 3 it gets a value slightly below 3, floors it to 2 and returns 10 - 2 * 10 / 3, which is slightly less than the actual result of 10 / 3, but looks like 10 / 3 (third line).

Unfortunately, this is not the expected result. See other notes for high quality fixes.

This class ran through several unit tests and fixes all failures found in bugs.php.net

<?php
abstract class MyNumber {
    public static function isZero($number, $precision = 0.0000000001)
    {
        $precision = abs($precision);
        return -$precision < (float)$number && (float)$number < $precision;
    }
    public static function isEqual($number1, $number2)
    {
        return self::isZero($number1 - $number2);
    }
    public static function fmod($number1, $number2)
    {
        $rest = fmod($number1, $number2);
        if (self::isEqual($rest, $number2)) {
            return 0.0;
        }
        if (mb_strpos($number1, ".") === false) {
            $decimals1 = 0;
        } else {
            $decimals1 = mb_strlen($number1) - mb_strpos($number1, ".") - 1;
        }
        if (mb_strpos($number2, ".") === false) {
            $decimals2 = 0;
        } else {
            $decimals2 = mb_strlen($number2) - mb_strpos($number2, ".") - 1;
        }
        return (float)round($rest, max($decimals1, $decimals2));
    }
}
?>

I don't believe that is correct.

Try this out using your patch:
<?php

echo duration( mktime(0, 0, 0, 1, 0, 2006)-time() );

?>

As of right now, this will read:
1 month, 22 days, 24 hours, 49 minutes, 15 seconds

Which is completely incorrect. Seeing as how it is the 9th of December.

The real real flaw here is how the 'year' and 'month' periods are calculated. As most months vary in length...

Thank you very much SnakeEater251 for pointing this out.

The quickest way to get slightly more accurate results, is to use averages based on one "true" year, which is 365.25 days.

Change the year and month to:
      'year'       => 31557600, // one 'true year' (365.25 days)
      'month'    => 2629800, // one 'true year' divided by 12 :-)

I will work on developing a true fix, for pin-point accuracy. ;-)

 - Cory Christison

Integer Modulo

If you want the remainder of the division of two Integers rather than Floats, use "%"; eg:

<?php
$a = 4;
$b = 3;

print($a % $b);
?>

Will output "1".

NAN (.net Equivalent = Double.NaN) means "Not-a-Number".
Some ways to get NaN are modulo 0, and square root of 0.

WARNING: Due to how floating point numbers work, fmod() and any simple alternatives are problematic when there is either a massive orders of magnitude different between the input $x and $y, or the input and output values. If you need to work with large numbers or arbitrary precision, it is best to work with something like BC Math or GMP.

When working around fmod()'s problems, remember that floor() always goes towards -INF, not 0. This causes a commonly proposed fmod() alternative to only work with positive numbers:
<?php 
function fmod_positive_only($x, $y) {
    return $x - floor($x/$y) * $y;
}
?>
Given these simplistic input values:
fmod_positive_only(-5, 3) = 1 (wrong)
-5 % 3 = -2 (correct)

Correctly removing the decimal part of the quotient can be achieved with either casting to an int (always goes towards zero) or dynamically choosing ceil() or floor(). Dynamically choosing floor or ceil in an attempt to keep precision is overkill. If your $x and $y values are so different that it suffers from an overflow problem when casting, it was probably going to have precision problems anyway (see warnings below).

<?php 
function fmod_overkill($x, $y) {
    if (!$y) { return NAN; }
    $q = $x / $y;
    $f = ($q < 0 ? 'ceil' : 'floor');
    return $x - $f($q) * $y;
}
?>

This is the "best" alternative for fmod() when given "normal" numbers.
<?php
function fmod_alt($x, $y) {
    if (!$y) { return NAN; }
    return floatval($x - intval($x / $y) * $y);
}
?>

WARNING: Even when you get a non-zero response, know your input numbers and when fmod() can go wrong. For large values or depending on your input variable types, float still may not contain enough precision to get back the correct answer. Here are a few problems with fmod() and their alternatives.

PHP_INT_MAX = 9223372036854775807
fmod(PHP_INT_MAX, 2) = 0 (wrong)
fmod_alt(PHP_INT_MAX, 2) = 0 (wrong)
PHP_INT_MAX % 2 = 1 (correct)

fmod(PHP_INT_MAX, PHP_INT_MAX - 1) = 0 (wrong)
fmod_alt(PHP_INT_MAX, PHP_INT_MAX - 1) = 1 (correct)
fmod_alt(PHP_INT_MAX, PHP_INT_MAX - 1.0) = 0 (wrong)
PHP_INT_MAX % (PHP_INT_MAX - 1) = 1 (correct)
PHP_INT_MAX % (PHP_INT_MAX - 1.0) = 9223372036854775807 (wrong)

fmod(PHP_INT_MAX, 131) =  98 (wrong)
fmod_alt(PHP_INT_MAX, 131) = 359 (wrong)
fmod_positive_only(PHP_INT_MAX, 131) = 0 (wrong)
PHP_INT_MAX % 131 = 97 (correct)