如何根据变量类型选择fscanf格式？

Question

Let's say I'm implementing a matrix and I want my struct to use scalar_t , where scalar_t can be float or double .

Then, let's say I want to scan some text file and load it into my matrix:

// Inner loop:
fscanf(mdata, format, &m->data[i * m->cols + j]);

If scalar_t * is float * , format should be "%f" . In case of double * it's "%lf" .

How do I condition the format on the type of scalar_t ? I know that I could compare sizeof but is there a general solution?

Answer 1

The designers of C99's <inttypes.h> had a similar problem. They defined types like int32_t with some specific properties, and allowed each implementation to typedef it as int or long as necessary. But then printf and scanf formats became platform-specific.

Their solution was to define a set of macros along with each type, like SCNd32 which will be "d" or "ld" to match int or long for the typedef.

Usage looks like

int32_t i;
scanf("%"SCNd32, &i);

For each type, there's a separate macro for every conversion that you might want to do with the type (so not just d but also x for hex, etc.)

You can do the same for your custom types. In the header:

#if something
typedef float scalar_t;
#define SCNfSCALAR "f"
#else
typedef double scalar_t;
#define SCNfSCALAR "lf"
#endif

And in the calling code:

scalar_t x;
scanf("%"SCNfSCALAR, &x);

Answer 2

Given that

1. You are going to need a union that has a float and a double .

2. The C standard requires a double to be a superset of a float .

3. Increasingly, platforms are setting float and double to the same thing, although they are required to be separate types, much along the lines of char , unsigned char , and signed char are always three separate types.

by far the best thing to do here is to use a double .