Re: Floating point error

Em 24/02/2013 23:26, John R Pierce escreveu:
> On 2/24/2013 6:20 PM, John R Pierce wrote:
>> On 2/24/2013 6:13 PM, Tom Duffey wrote:
>>> - The Java app on production shows "10.3884573" while the test app
>>> shows "10.3885"
>>
>> 'real' is single precision, which is only about 6 digits of decimal
>> accuracy.   if your java variables were double precision, you
>> probably should have used double in postgres too.
>>
>> 0.1 decimal is a repeating fraction in binary.
>
> ooops, missed my third point.    if precise decimal numbers are
> important, use a decimal data type, which is NUMERIC in SQL (I'm not
> familiar enough with Java to know if it has a BCD or other decimal type).
>
Java BigDecimal is the best fit for Numeric in PostgreSQL

It requires a bit more programming effort, but is the most precise type
for money work.

For instance to sum 2.00 plus 2.00 program:

public static void main(String [] args) {
BigDecimal num1 = new BigDecimal("2.00");
BigDecimal num2 = new BigDecimal("2.00");
BigDecimal sum = num1.add(num2);

if(sum.compareTo(new BigDecimal("4.00"))==0) {
   System.out.println("Matches: 2.00+2.00 == 4.00");
} else {
   System.out.println("Uau, how did you get here?");
}
}

You must use "compareTo" method to compare two values, and can't use
BigDecimal "equals" to compare values, because 2.0 is different than
2.00 (scale differ)...


Regards,

Edson