Double addition error

[jimstutt]

ghci>0.1+0.2 ghci>3.000000000000004

ghc produces the same result with:

foldl (+) [0.1,0.2]

[jimstutt]

Sorry foldl (+) 0 [0.1,0.2]

[ross]

It's an artifact of fixed-precision floating point not being able to represent decimal fractions exactly; you get the same from C:

#include <stdio.h>

int main() {
        printf("%.17f\n", 0.1+0.2);
        return 0;
}

[simonpj]

Well, if 0.1+0.2 really gave result 3.0000000004, as reported, THAT would be a bug. But it doesn't. It gives 0.3000000004. Wich as Ross says is fine.

[jimstutt]

Embarrassment. Didn't mean to waste time - didn't know or realise ghc stuck to IEEE representation nor that addition has rounding errors, not just division. Back to "what every scientist should know about floating point..". And too blind to see 3.0 not .3. Apologies.

[maeder]

ghc is not stuck to IEEE (it only defaults to Double) but you can do

Prelude> 0.1 + 0.2 :: Rational
3 % 10

The output is not printed nicely, but how should it print 1/3 as decimal?

Prelude> 1 / 3 :: Rational
1 % 3