Manipulating the address of a variable to store a smaller type?

Question

This is what I get for pampering myself with high-level programming languages.

---

I have a function which writes a 32-bit value to a buffer, and a uint64_t on the stack. Is the following code a sane way to store it?

uint64_t size = 0;
// ...
getBytes((uint32_t*)&size+0x1);

I'm assuming that this would be the canonical, safe style:

uint64_t size = 0;
// ...
uint32_t smallSize;
getBytes(&smallSize);
size = smallSize;

Answer 1

No. It works correctly only on big-endian machines. And assuming a particular byte order - without even checking it first - is not sane.

Even if you are sure that your program runs only on big-endian machines right now, you'll never know whether it might have to run on a little-endian machine in the future. (I'm writing this on a computer made by a company which used big-endian processors for decades, then switched to little-endian processors a couple years ago, and is now also quite successful with bi-endian processors in certain devices ;-))

Answer 2

Why not make getBytes() to return uint64_t? And use the argument (e.g. int *) to return error code if any.

From my personal experience, if you really want to unify the two code paths, then use uint64_t in both.

Also note that "(uint32_t*)&size" breaks C99 strict aliasing rules (and e.g. in GCC one would have to disable the optimization).