За что математики не любят компьютерную арифметику

[vak]

#include <stdio.h>
int main ()
{
    if (-5.2 + 4.9 != -0.3) {
        printf("Oops!\n");
    }
    if (4.8 - 6.1 != -1.3) {
        printf("Oops!\n");
    }
    if (4.3 - 3.6 != 0.7) {
        printf("Oops!\n");
    }
}

Это мне надо было простенький тестик сварганить по работе, суммировать плавающие числа, а оно вон как боком выскочило.

Comments

nz

[nz]

Зачем так сложно, 0.1 + 0.2 != 0.3 же.

Re: nz

[vak]

Тоже верно. Упростил у себя.

C#

[dennisgorelik]

In C# floating point calculations are not precise too:

[TestMethod]
public void FloatingPointCheck()
{
	Assert.AreEqual(0.2, 0.1 + 0.1); // Passes
	Assert.AreEqual(3.0, 1.0 + 2.0); // Passes
	Assert.AreEqual(0.3, 0.1 + 0.2); // Fails
}

Re: C#

[vak]

Calculations are OK. Decimal representations are imprecise.

Re: C#

[dennisgorelik]

Is there a convenient way to use more precise representation?

Re: C#

[ircicq]

Type `decimal` in many programming languages including C#
Although it is slower

Re: C#

[dennisgorelik]

Oh.
vak meant "data representation in computer memory" (floating-point double vs decimal).
I thought he meant representation/conversion of data to decimal string.

In any case, the primary way to deal with imprecision is not increasing precision, but replacing number equality comparison with [epsilon] range comparison.

[vak]

Hex literals work fine. Not much human friendly though.

#include <stdio.h>
int main ()
{
    printf("%a\n", -0x5.2p0 + 0x4.9p0);
    printf("%a\n", 0x4.8p0 - 0x6.1p0);
    printf("%a\n", 0x4.3p0 - 0x3.6p0);
}

Результат:

-0x1.2p-1
-0x1.9p+0
0x1.ap-1

[dennisgorelik]

> Hex literals work fine.

Hex interface is a good workaround.
But it may be hard to convince end users to use hex interface, right?