scanf() and floating point numbers in Turbo C

Question

I created a structure as:

typedef struct {
  float real, img;
} cmplx;

And I created a function as

void input(cmplx *a) {
  scanf("%f + %f i", &a->real, &a->img); 
}

and called the function from main as:

cmplx a;
input(&a);

The execution stops when the scanf is reached. If floats are replaced by ints, the problem is solved. What is this behaviour? Is there a way I can use floats for my problem?

The program was compiled in Turbo C, in Windows XP.

Answer 1

It is hard to answer without knowing the error message that your program's execution stops with, but from your comment "is there any code i can write to tell compiler to link floating point library", I suspect it may be this issue:

"Floating point formats not linked" is a Borland run-time error (Borland C or C++, Turbo C or C++). Borland's compilers try to be smart and not link in the floating- point (f-p) library unless you need it. Alas, they all get the decision wrong. One common case is where you don't call any f-p functions, but you have %f or other f-p formats in scanf() or printf() calls. The cure is to call an f-p function, or at least force one to be present in the link.

To do that, define this function somewhere in a source file but don't call it:

static void forcefloat(float *p)
{ 
    float f = *p;
    forcefloat(&f);
}

It doesn't have to be in the module with the main program, as long as it's in a module that will be included in the link.

If you have Borland C++ 3.0, the README file documents a slightly less ugly work-around. Insert these statements in your program:

extern unsigned _floatconvert;
#pragma extref _floatconvert

Using this workaround, or a more modern compiler, will probably fix your issue.