How to create a byte file?

Question

I'm making a C/POSIX program that manipulates matrices written in files.

I want each matrix to be written in a different file with the following format:

1. All numbers are written as byte sequence following the endiannes of the computer
2. All numbers are written in the file row by row

So, for example in a little-endian computer with 32 bit int, the following matrix:

| 100 200 |
| 300 400 |
| 500 600 |

Would be stored in a byte sequence we can represent like this (where each number represents the decimal value of the corresponding byte):

100 0 0 0 200 0 0 0 44 1 0 0 244 1 0 0 88 2 0 0

As I have to test my program, how do I create such input files?

Answer 1

Here's my solution to the problem: I created a program (matrix_creator) that receives as arguments the path of the output file (argv[1]) and all numbers it needs to write in the output file (argv[2], argv[3],...).

I excluded all error controls to make the code easier to read:

int fd = open(path, O_RDWR|O_CREAT, 0644); // Create output file
for(int i = 2; i < argc; i++) { // Starts from argv[2]
    int t = strtol(argv[i], NULL, 0); // Parse each input number
    write(fd, &t, sizeof(t)); // Write the number in the given format
}
close(fd);

In such a way I can create different files using the same program for several tests.

To check what's inside the file I installed ghex.

Answer 2

This code:

int main(int argc, char* argv[])
{
   FILE *fp;
   fopen_s(&fp, "testfile.dat", "wb");
   int i[] = { 100, 200, 300, 400, 500, 600 };
   fwrite(i, sizeof(i), 1, fp);
   fclose(fp);
   return 0;
}

Gives me a file that looks like this in Visual Studio's binary editor:

64 00 00 00 C8 00 00 00 2C 01 00 00 90 01 00 00
F4 01 00 00 58 02 00 00

Answer 3

You simply write out the data

 int x = testvalue;
 fwrite(&x, 1, sizeof(int), fp);

However the Posix specification is awful. If the width of an int or the endianness of the computer changes, the file breaks. That seems to be the required behaviour.

Check out my github project on binary file portability

https://github.com/MalcolmMcLean/ieee754

it's much better to write out binary files so that the file, not the computer, specifies the format. And then read portably. (With floating point it is hard, with integers there are a few gotchas and it's not totally simple).