Divide a float into single bytes

Question

I am using RakNet for networking and I need to transmit float values. The problem is, that I can only send single bytes. Since float is composed out of 4 bytes, I thought it must be possible, to tear the float down into these four bytes and send them. After that, I want to compose them together again to a float.

How do I do that?

pointers and pointer math... get the address of the float, then start sucking bytes. — Marc B, Dec 24 '13 at 17:20
You could serialize it: http://stackoverflow.com/questions/1786137/c-serialization-of-the-floating-point-numbers-floats-doubles — Hariprasad, Dec 24 '13 at 17:28

AnT · Accepted Answer · 2013-12-24T17:36:26.293

6

memcpy the original float value to an appropriately sized array of unsigned chars

float f = ...;

unsigned char c[sizeof f];
memcpy(c, &f, sizeof f);
// use `c[i]` to access the bytes

Reinterpret the original pointer as an array of unsigned chars using reinterpret_cast

float f = ...;

unsigned char *c = reinterpret_cast<unsigned char *>(&f);
// use `c[i]` to access the bytes

Use a union to perform the same reinterpretation

union FC {
  float f;
  unsigned char c[sizeof FC::f];
};

float f = ...;

FC u = { f };
// use `u.c[i]` to access the bytes

Just keep in mind a few extra details: the last approach (with union) is now legal in C, but is still formally illegal in C++ - you are not allowed to read "inactive" members of unions in C++. Observe the byte ordering on little- vs. big-endian platforms. If you use the second approach, also reinterpret the recipient float as an array of unsigned chars and not the other way around to avoid alignment issues.

edited Dec 24 '13 at 17:36

answered Dec 24 '13 at 17:30

AnT

302,239
39
506
752

I wouldn't use memcpy function because it is a very risky function from the point of view of security. ;) – NKN Dec 24 '13 at 17:41
@NKN: There's nothing risky about it if you know what you are doing. "Risky" functions are the functions that do not offer you any chance to prevent problems (like `gets` or `atoi`, for example). `memcpy` does not belong to that category. – AnT Dec 24 '13 at 17:42
Thanks for your answer. The first method sounds like the easiest. But how do you make a float again out of these bytes? – Dec 24 '13 at 17:45
@Liess Jemai: In the first method, you just `memcpy` it back from char array to a `float` object. In the second method... well, you have to be careful: reinterpret the target `float` object as an array of char and write the data there. In the third method, you simply do it backwards: store the bytes into `u.c` and then read `u.f`. – AnT Dec 24 '13 at 22:58
1

@zhiayang: Reinterpreting char array as a float is not a good idea: it does not guarantee that the alignment requirements are met and it is generally illegal. – AnT Dec 24 '13 at 22:59

Divide a float into single bytes

1 Answers1