从标准输入读取双数，使用 C read()

Question

According to the Linux manual ,

#include <unistd.h>
ssize_t read(int fd, void *buf, size_t count);

attempts to read up to count bytes from file descriptor fd into the buffer starting at buf.

I am writing a C program that uses "read" (instead of scanf) to read a double from stdin. Below is what I get. I typed inputs such as 42.0, 47.9, etc from the console. It turns out that the program produces random outputs. Can anyone help?

#include <stdio.h>
#include <unistd.h>

int main(){
    double x;
    read(0, &x, sizeof(x));
    printf ("Results = %g\n", x);
}

Answer 1

Input on standard input (please use STDIN_FILENO instead of the "magic number" 0 ) is usually text .

You attempt to read the input as a raw binary representation of the value.

Unless you know that the input will be in raw binary form (for example from redirection of a binary file or from a pipe where the other end is writing raw binary data) I suggest you read into a character buffer, make sure the buffer is null-terminated (you have to do it yourself, the position is what read return if successful), and use eg sscanf or strtod to convert the string.

Answer 2

As the manual says, attempts to read up to count bytes from file descriptor fd and that means for a double for example to read sizeof (double) bytes with the internal format representation of a double value. As it says it reads bytes, those bytes can be generated by a corresponding write(2) of a double value, but are only understandable by the machine, so entering 24.2 is not valid, as that is a string representation of a number in ascii decimal form.

There's a standard binary representation for internal use on extended use, described in the IEEE-752 standard. It describes internal binary format representations for float , double and extended long doubles (128bit representation of a floating point number, normally not used in current architectures, but only a small set) If you know it, probably you can produce the exact representation needed to be able to input correctly a specific number. You should read that document in order to understand why the sequence of chars 2 , 4 , . , 2 , that form the string 24.2 is not the same as the internal binary representation of that number, which is what is written by write(2) .

Try to write(2) that first, with:

double number = 3.1415926535897932;
write(1, &number, sizeof number);

and see what happens in your output. Next try to filter up with a hex reader, as in

$ new_program | hd

and you'll see the exact values of the bytes that pi produces. Then, try to feed that to your original program, as in

$ new_program | your_original_program

and print them again with

read(0, &my_double, sizeof my_double);
printf("%g", my_double);

you'll see that what was read is what you wrote, but the binary representation of a double value is, in general, misunderstandable for humans (well, there's people so used to binary that understand hex dumps up to this level, but I'm not one of them)