float multiplication and subtraction

i know this has been mentioned over and over (and over) again in this forum but i would like to ask specifically on a function i've been playing around:

Code:

#include <stdio.h>
#include <stdlib.h>

char *ftoi(float f){
    char str[20];
    char *fstr;
    float a;
   
    fstr= malloc(100);
       
    sprintf(str,"%i",(int)f); // get the int part and make it string
    strcat(fstr,str);  // store this part inside the main string
    strcat(fstr,"."); // oh, put the decimal place too
       
    f-=(int)f; // keep the part after the decimal place
   
    while(f>0){
        printf("%f\n",f); 
        f*=10;  // move the number 1 decimal place left
        printf("%f\n\n",f);
        sprintf(str,"%i",(int)f);  // make int part -> string
        strcat(fstr,str);  // store it
        f-=(int)f;      // keep only decimal 
    }
   
    printf("%s\n",fstr);   

     return fstr;
}

obviously that i want to convert float -> string.

but then all hell breaks loose. say for f=2,55 my printf's are the following:

Code:

0.550000  //ok let's see
5.500000  //so far so good

0.500000  //...ok...
4.999995  // nooooooooooo....

0.999995 //&^*&^%&^&*ing ^$%#^%
9.999952

0.999952
9.999523

0.999523
9.995232

0.995232
9.952316

0.952316
9.523163

0.523163
5.231628

0.231628
2.316284

0.316284
3.162842

0.162842
1.628418

0.628418
6.284180

0.284180
2.841797

0.841797
8.417969

0.417969
4.179688

0.179688
1.796875

0.796875
7.968750

0.968750
9.687500

0.687500
6.875000

0.875000
8.750000

0.750000
7.500000

0.500000
5.000000   // now that's funny, no 4.99 i guess it didn't have any other blocks to                    read 

2.5499999523162841796875  // :-/

most probably the problem is at the multiplication. i even tried fdimf()

Code:

fdimf(f,10,0) or fdimf(f, 10.0, 0)

because of the so called "infinite precision" but i saw no change at all...
played with doubles a bit too but nothing again...

any suggestions on what should i do (except from messing with floats unless i'm using matlab) ?

thank's in advance guys.

Well, to be honest, MATLAB thinks that 2.55 is 2.5499999523162841796875 too, since floats are the same under the hood in both. (Unless MATLAB has changed since the last time I used it, which was a while ago.)

I don't know where you're getting your numbers from; I would recommend the %g format specifier (and sprintf can print directly into a string/char array, which is handy). If you know that your numbers are supposed to be nice fractions, you could build a fraction data type. If you know a maximum number of decimal places, you could build a fixed-point data type.

Originally Posted by tabstop

If you know that your numbers are supposed to be nice fractions, you could build a fraction data type.

nice fractions?

If you know a maximum number of decimal places, you could build a fixed-point data type.

the only thing i want to assume is float range and positive values...

Without knowing more about what the numbers are, I can't be more specific than "nice fractions".

Float range is a tricky thing. As you've seen, even 2.55 is not a valid floating-point value. Again, my suggestion is to use sprintf and %g.

Your ftoi function if flawed in so many ways:

1. By the name "ftoi", most would immediately assume it converts a float to an int.
2. The design is flawed in the it will almost certainly result in memory leaks due to forgetting to free the return result. (Better to pass the buffer in instead)
3. The function does not check the return result of malloc for NULL.
4. Floats can store larger values than ints. Your function will fail for numbers over 11 decimal digits, which as it happens there are far more of than there are under 11 digits. E.g. 1E50
5. Floats can store values far below 0.0000000001 it will again fail for these and produce zero. E.g. 1E-50
6. The function fails for negative values.
7. There are already way provided to do this which actually works. printf is one way.

There's nothing wrong with playing around with this, but make no mistake that it should only be for learning experience. The above comments are simply to help you along in that learning.

I happen to already have a fraction class. It can be found in the Useful Classes section of the website linked in my signature. It will even convert a float to a fraction, with pretty good accuracy.