More fun with floating points

Mithious

By coincidence this is on a similar topic to today's CodeSOD. I present to you this rounding algorithm copy-pasted 100s of times through our codebase.

    roundIt = 1;
    sprintf_s(rndx, "%50.12f", givenDoubleValue);

    if (rndx[strlen(rndx) - 12] == '0')
      if (rndx[strlen(rndx) - 11] == '0')
        roundIt = 0;

    if (roundIt)
    {
      rndx[strlen(rndx) - 12] = '0';
      rndx[strlen(rndx) - 11] = '0';
    }

    memset (x, '\0', sizeof x);

    strncpy_s(x, rndx, (strlen(rndx) - 10));

    givenDoubleValue = atof(x);
    if (roundIt)
      givenDoubleValue += 1.00;

DrPepper

By about the 3rd line of this code, I'd be thinking -- there must be a better way of doing this. Converting a numeric value into a string for further processing is almost always a code smell.

konnichimade

@Mithious said:

... copy-pasted 100s of times through our codebase.

D:

jnz

So 1.01 rounds to 2? It is similar to truncating to two decimal places and then running through ceil().

Fun things happen to negative numbers. -0.99 rounds to 1!

And hopefully you never have large numbers (like 1e50) or it will abort your program.

PJH

@jnz said:

And hopefully you never have large numbers (like 1e50) or it will abort your program.

%f would (should) never return exponential formatted numbers - you need %g for that:

[pjh@pjh tmp]$ cat ./format.c
#include <stdio.h>
int main(void){
        double d =1e100;
        printf("%.12f\n", d);
        printf("%.12g\n", d);
        return 0;
}
[pjh@pjh tmp]$ make format
cc     format.c   -o format
[pjh@pjh tmp]$ ./format
10000000000000000159028911097599180468360808563945281389781327557747838772170381060813469985856815104.000000000000
1e+100
[pjh@pjh tmp]$

Mithious

@jnz said:

Fun things happen to negative numbers. -0.99 rounds to 1!

I was wondering if anyone would notice that one. :)

DrakeSmith

I thought you were supposed to always round down and then deposit the extra into your own bank account - after beating the crap out of your copier, of course.

Or am I 'jumping to conclusions'?

antiquarian

@DrPepper said:

By about the 3rd line of this code, I'd be thinking -- there must be a better way of doing this. Converting a numeric value into a string for further processing is almost always a code ~~smell~~ stench.

FTFY

dkf

@jnz said:

So 1.01 rounds to 2? It is similar to truncating to two decimal places and then running through ceil().

Actually, it's just like running through ceil(). And then adding 1 if the original number was negative, just for giggles.
@jnz said:

And hopefully you never have large numbers (like 1e50) or it will abort your program.

You can't tell that for sure; it depends on the buffer sizes and we were never told them. (Most likely true though.)