Number step test modulo (fmod) has precision errors with odd float numbers

Thanks, a good suggestion. PHP (and computers in general) certainly need little number massaging when dealing with odd numbers and division.

I just ran into this same problem with the 7.x-3.18 release. This statement on line 565 in numbers.inc

fmod($element['#value'] - $starting_number, $element['#step']

returns 0.0099999999999999 when $element['#value'] == 3.50, $starting_number == 0, and $element['#step'] == 0.01. I expected a result of 0.

Here's a custom function that returns correct results:

function webform_fmod($n,$b) {
  return $n-$b*floor($n/$b);
}

This issue still exists in 7.x-4.0-alpha8. Here's a patch.

Thanks @sah62! This needs to be applied everywhere, as it exists in all branches. Moving this to needs review to it will get picked up next time I'm going through the review queue.

Doh, sorry @sah62. I went through the "easy" patches the past few weeks but missed this one. Next release. :)

No prob...

I adjusted this patch to take the advice of this poster on PHP.net. This version uses ceil() or floor() based on if the number is negative for the proper modulo value when working with negative numbers.

Committed to all branches. Thanks guys!

Works for me - thanks!

Given the webform_modulo code:
return $a - $b * (($b < 0) ? ceil($a / $b) : floor($a / $b));

If I try this with:
$a = 4.22
$b = 0.01

I get webform_modulo = 0.0099999999999998
and fmod = 0.0099999999999997

This appear to be an issue with PHP which this function does not resolve.
See var_dump(4.22-4.21);

Hi,
I have the same issue with Webform 4 RC3.

There is a number field with next validation settings:

Minimum: 0
Maximum: 10
Step: 0.1

When I'm trying to submit a form with value 8.6 I'm getting an error:

[field-title] field value must be a multiple of 0.1. i.e. 0, 0.1, 0.2, 0.3, etc.

There are two sources of this error. The first is that you can't compare floating points with the equal operator because of very small floating point rounding errors. The same routine used for fp comparisons in conditionals is used instead. The second is that due to fp rounding errors, the modulo function may return the modulus (e.g. 0.1 in the example), rather than 0. Both are explicitly tested for. To implement this I moved the fp comparison routine from conditionals.inc to number.inc.

Committed to 7.x-4.x and 8.x.