Re: ROUND



kkmigas@xxxxxxxxx wrote:
On Dec 27, 12:54 am, Jerry Stuckle <jstuck...@xxxxxxxxxxxxx> wrote:

kkmi...@xxxxxxxxx wrote:

PHP code
echo "round(10.045,2) - ".round(10.045,2)."<br />";
echo "round(20.045,2) - ".round(20.045,2)."<br />";
echo "round(30.045,2) - ".round(30.045,2)."<br />";
echo "round(40.045,2) - ".round(40.045,2)."<br />";

This is very strange:

round(10.045,2) - 10.04
round(20.045,2) - 20.05
round(30.045,2) - 30.05
round(40.045,2) - 40.05

On Dec 26, 11:57 am, "kkmi...@xxxxxxxxx" <kkmi...@xxxxxxxxx> wrote:

I've tested the same function in the same hardware configuration and
the only difference is the OS, so it must be the libraries the big
problem is that i don't know witch ones to replace in order to get the
system to solve the way i want.

On Dec 22, 4:42 pm, "David T. Ashley" <d...@xxxxxxxx> wrote:

<kkmi...@xxxxxxxxx> wrote in messagenews:1166719766.801640.296050@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Can some one explain if this can be fixed using php.ini settings ?

echo "round 20.545 -".round(20.545,2)."<br>";
echo "round 20.555 -".round(20.555,2)."<br>";
echo "number_format 20.545 -".number_format(20.545, 2, ',',
'.')."<br>";
echo "number_format 20.555 -".number_format(20.555, 2, ',',
'.')."<br>";

PHP Version 4.3.0 / FreeBSD
round 20.545 -20.55
round 20.555 -20.56
number_format 20.545 -20,55
number_format 20.555 -20,55

PHP Version 4.4.4 / CENTOS
round 20.545 -20.55
round 20.555 -20.55
number_format 20.545 -20,55
number_format 20.555 -20,55

PHP Version 5.1.4 / Windows NT
round 20.545 -20.55
round 20.555 -20.56
number_format 20.545 -20,55
number_format 20.555 -20,56My only observation about the examples you chose is that the fractional

parts are infinitely repeating "radiximals" in binary.

0.545 = 545/1000 = 109/200 (irreducible)
0.555 = 555/1000 = 111/200 (irreducible)

Both of the numbers above can't be expressed as precise binary floating
point numbers because the denominator after reduction has prime factors that
are not "2".

PHP.INI may be a factor, but it may also be that something at a lower level
is occurring (i.e. machine arithmetic, how certain rounding settings on
floating-point processors or libraries are set).

In other words, you've chosen antagonistic examples.

Why don't you try 7/8 (0.875), 15/16 (0.9275), or 31/32 (0.96875) as the
fractional part of the numbers and see if you can get the same behavior.
I'm guessing that you may not.

Any fraction with a reasonable denomiator (< 2^16) that is a power of 2
would be a reasonable test case.

Note that these are all exactly representable by a typical machine.

Just a guess.This is the exact same problem you reported in comp.databases.mysql, and

the solution is exactly the same.

--
==================
Remove the "x" from my email address
Jerry Stuckle
JDS Computer Training Corp.
jstuck...@xxxxxxxxxxxxx
==================



> Has i told you on the other newsgroup
>
> This 2 servers have the same hardware configuration
>
> On Windows 2003
> http://efeito.org/info.php
>
> On Linux
> http://efeito.net/info.php
>

(Top posting fixed)

And as I and others have told you - when using floating point, results when using expressions like these are unpredictable.

It's just like when using base 10:

round((1.0/3.0) + (1.0/6.0))

can equal zero, because neither 1.0/3.0 nor 1.0/1.6 can be expressed as an exact value in base 10. They are repeating decimals.

Sure, they *should* come out to 0.5. But their actual value will be 0.499999999 to however many decimal places could be handled.

*Some* libraries will "fix" this by adding a small value (i.e. 0.000000001) to the results to "fudge" the value. But no such fudge factor is required, nor is it always good.

You are always best to use integers if you need exact values, or add your own fudge factor to the results.

It's been that way for the almost 40 years I've been programming, and it's not going to change now. That's why many processors now have a "packed decimal" or a "binary coded decimal" type internally to give exact results.

--
==================
Remove the "x" from my email address
Jerry Stuckle
JDS Computer Training Corp.
jstucklex@xxxxxxxxxxxxx
==================
.