从C中的浮点数转换整数的舍入不一致

Question

Been tearing my hair out all day over this little problem I have that is proving hard to fix and I'm fairly sure there are things going on under the surface that I am not aware of. Any feedback or help is hugely welcomed.

I have a function which takes in a float value and returns a string of ASCII characters for output onto an LCD screen. The function is thus:

char returnString(float x, bool isTriggTemp)
{
    char numberString[3];
    char bufferString[2];
    char bufferStringDec[7];

    int integerVal = (int)x;

    if(isTriggTemp == false)
    {
        int decimalVal = (x-integerVal)*10000;
        sprintf(numberString, "%s.%s", itoa(bufferString,integerVal,10),
             itoa(bufferStringDec,decimalVal,10));
    }
    else
    {
        int decimalVal = (x-integerVal)*1000; 

        sprintf(numberString, "%s.%s", itoa(bufferString,integerVal,10),
             itoa(bufferStringDec,decimalVal,10));
    }

    return numberString;
}

The reason for the If statement and the bool is that there are two floats that can be passed to the function. One with 1 decimal place, and one with up to 5. I'm not to fussed about the 5 decimal place number as I am comparing the two floats later in the program and the relevant decimals seem roughly consistent.

My query stems from this line:

 int decimalVal = (x-integerVal)*10;

When used like this, which is what I would expect to have to use given the above logic, the outputted string reads "X.0" regardless of the X value.

Increasing it to 100:

 int decimalVal = (x-integerVal)*100;

Gives me the correct value for only even numbers. (X.2 is fine when the float is X.2), but with odd numbers I seem to get a rounded down version (X.3 float prints X.2, X.5 -> X.4) etc.

When I increase it to 1000, I begin to see the start of the problem:

int decimalVal = (x-integerVal)*1000;

For even values, X.4 for example, I get X.40 printed. And with odd numbers I get 0.01 less than the original float. EgX5 - > X.49.

Obviously there's something amiss here and non-exact decimals are being cut off. A) how can I fix this? and B) Given the arithmetic I would have guessed that *10 should be used but *100 is the order of 10 that gives me the closest to the correct output.

Any help is much appreciated.

Answer 1

You can write the precision you want for floating-point numbers with %f. And even do the precision level dynamically:

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

// Use malloc (drawback: need to track return and deallocate to prevent memory leak.)
char * returnString( float x, int isTriggTemp )
{
    char * buffer = malloc( 128 );
    if( buffer )
        sprintf( buffer, "%0.*f", isTriggTemp ? 4 : 5, x );
    return buffer;
}

// Use static buffer (drawback: non-reentrant)
char * returnStringStatic( float x, int isTriggTemp )
{
    static char buffer[128];
    sprintf( buffer, "%0.*f", isTriggTemp ? 4 : 5, x );
    return buffer;
}

// Use given buffer (drawback: caller needs to be aware of buffer size needs and additional parameters are involved)
char * returnStringGivenBuffer( char * buffer, float x, int isTriggTemp )
{
    sprintf( buffer, "%0.*f", isTriggTemp ? 4 : 5, x );
    return buffer;
}

int main( int argc, char ** argv )
{
    float val = 3.14159;
    int highprecision = 0;

    printf( "Using sprintf\n" );
    printf( "%0.1f\n", val );
    printf( "%0.5f\n", val );
    printf( "%0.*f\n", ( highprecision ? 5 : 1 ), val );
    highprecision = 1;
    printf( "%0.*f\n", ( highprecision ? 5 : 1 ), val );

    printf( "\nUsing sprintf\n" );
    char buffer[128];
    sprintf( buffer, "%0.1f", val );
    printf( "%s\n", buffer );
    sprintf( buffer, "%0.5f", val );
    printf( "%s\n", buffer );
    sprintf( buffer, "%0.*f", ( highprecision ? 5 : 1 ), val );
    printf( "%s\n", buffer );
    highprecision = 1;
    sprintf( buffer, "%0.*f", ( highprecision ? 5 : 1 ), val );
    printf( "%s\n", buffer );

    printf( "\nUsing dynamic allocation\n" );
    char * fval = returnString( val, 0 );
    printf( "%s\n", fval ? fval : "" );
    if( fval ) free( fval );
    fval = returnString( val, 1 );
    printf( "%s\n", fval ? fval : "" );
    if( fval ) free( fval );

    printf( "\nUsing static buffer\n" );
    char * ptr = returnStringStatic( val, 0 );
    printf( "%s\n", ptr );
    ptr = returnStringStatic( val, 1 );
    printf( "%s\n", ptr );

    printf( "\nUsing given buffer\n" );
    ptr = returnStringGivenBuffer( buffer, val, 0 );
    printf( "%s\n", ptr );
    ptr = returnStringGivenBuffer( buffer, val, 1 );
    printf( "%s\n", ptr );

    return 0;
}

Results:

Using sprintf
3.1
3.14159
3.1
3.14159

Using sprintf
3.1
3.14159
3.14159
3.14159

Using dynamic allocation
3.14159
3.1416

Using static buffer
3.14159
3.1416

Using given buffer
3.14159
3.1416

Answer 2

OP approach to printing a fraction fails under various situations

1. fraction is close to 1.0 . returnString(0.999999, isTriggTemp)

2. x value greater than INT_MAX .

3. Negative numbers.

A more reliable method to to scale first and then break into integer/fraction parts.

char *returnString(float x, bool isTriggTemp) {
   float scale = 10000;
   x = roundf(x * scale);
   float ipart = x/scale;
   float dpart = fmodf(fabsf(x), scale);
   itoa(bufferString,ipart,10);
   itoa(bufferStringDec,dpart,10);  // May need to add leading zeros
   ...

[Edit]

An easy way to have leading zeros:

   static char bufferString[20+7];
   char bufferStringDec[7];

   itoa(bufferStringDec,dpart + scale,10);
   bufferStringDec[0] = '.';  // Overwrite leading `1`.
   return strcat(bufferStringDec, bufferStringDec);

Of course code could use

void returnString(char *numberString, size_t size, float x, bool isTriggTemp) {
  snprintf(numberString, sizeof numberString, "%.*f", isTriggTemp ? 3 : 4, x);
}

---

Insure buffers are static or otherwise available after the function completes.

Check return type.

Answer 3

Are you sure this code doesn't crash? Your buffers are so small I think you might be overflowing your buffers. Anyway, the reason evens work and odds don't may be because when you do things like int decimalVal = (x-integerVal)*10000; things get rounded down. For example, int x = (int)((2.29 - 2)*10) will give you 2, not 3. In this example, the fix is to add .05 before you multiply by 10. I'm sure you can figure out the general rule.

Answer 4

Mr. Houpis answer is the best thing to pursue, but for your edification, you probably need to do the calculations as floats:

int decimalVal = (x - (float) integerVal) * 10000.0;

In addition, your declaration for returnString should be char* rather than char.