>> I do wonder why you [the programmer] would choose one >> rounding rule over another [instead of allowing for whatever >> was specified in the application spec].=20
"Dmitriy Nagirnyak" <nagir_argocomputing_com_au> wrote
> I prefer arithmetic rounding because if it is traditional and=20 > most common method (Symmetric Arithmetic Rounding).
Dmitriy,=20
Just to help clarify the argument, are you referring to=20 rounding of decimal fraction variables like varDecimal or BCD? Or binary fraction variables like single, double, and extended?
I am not sure why you would want to use the "most common"=20 method unless your application is a "most common" application.
> It is used in mathematics, physics. ... > http://en.wikipedia.org/wiki/Rounding
Ditto for most of the other rounding methods. I bet that=20 I can even find an application of "statistical rounding".
> The default .NET implementation of Math.Round implements > bankers rounding and might be used for banking calculations.
> Also ROUND functions in the databases (including mssql)=20 > also use [symmetric] arithmetic rounding.
Note SQL Server can also use "floor" and "ceil" rounding,=20 although they only round to whole number values.=20
Discussion of rounding and a source of names for different=20 rounding methods: http://support.microsoft.com/kb/196652
Discussion of role of accountant in making some rounding=20 decisions: http://bytes.com/forum/thread374225.html
> So I think ECO (if it supports round in PS) will probably > fail the test using OclService and OclPsService like=20 > "self.Transactions->select(Amount.round>4)" > If there will be Transactions with Amounts 3.5 and 4.5 > (they will round to 4 and 4 correspondingly).
> [Symmetric arithmetic rounding] is used everywhere ...
I doubt that! =20
> Math.Round in .NET is just wrong.=20
To which of the overloaded Math.Round methods are you referring? http://msdn.microsoft.com/en-us/library/system.math.round.aspx
> I'll tell the story about it. I was writing a diploma work.=20 > I wrote system that visualises Pascal program execution. =20 > So I had to interpret Pascal language (some ANSI standard > of it). Of course round has to be supported there. There > were no problems, I just call Math.Round to perform it. I > didn't know that it's bankers round those days (all languages=20 > did arithmetic, except .NET platform, as it turned out). > I was told that my system was working incorrectly. I was > very surprised when I saw this: > int :=3D Round(4.5) =3D=3D=3DVisualised as=3D=3D=3D>4 > And it was really incorrect. So I had to replace Math.Round > with my own method. =20
Did you extend your program to do rounding to a specified=20 number of decimal fraction digits? Did you program use decimal fraction variables like varDecimal=20 or BCD? Or binary fraction variables like single and double? =20
>> I would presume that the application should dictate the >> rounding requirements and that it not be left to the >> programmer's choice. =20 > ... but default rounding method should be [symmetric > arithmetic rounding], not bankers.
I wonder how you would do that within the FPU? <a challenge>
> If you need non-standard rounding - implement it.
That is easier said than done, especially if you have to do=20 it using floating binary point variables, like single, double,=20 and extended, to represent decimal fraction values.
Regards, JohnH