Size of structures is messed up

Code:

---

struct little{
    unsigned char ident;
    unsigned short int data;
};

---

Code:

---

    char debug[256];
    sprintf(debug,"%i+%i=%i",sizeof(unsigned char),sizeof(unsigned short int),sizeof(little));
    MessageBox(0,debug,"Debug",MB_OK);

---

The messagebox says 1+2=4. How is this possible?

Structures have internal and trailing padding to preserve alignment contraints.
This has been discussed many times before.

How can I get around this "padding"? I want to save it into a binary file as data, but it adds some bytes with random values.

http://en.wikipedia.org/wiki/Data_structure_alignment

I just used good old char array. No padding problems. :)
So that instead of:

Code:

---

boke[count].ident=i+1;
boke[count].data=number;

---

I have:

Code:

---

boke[count*3]=i+1;
boke[count*3+1]=number/256;
boke[count*3+2]=number%256;

---

Low-level stuff always helps out.

try to reorder from longest types to shortest...

> but it adds some bytes with random values.
So long as you use the same structure for writing and reading, the content of the padding bytes doesn't matter.

If you're paranoid about it, memset the whole structure to 0 before assigning individual fields.

Or use a non-portable "#pragma" of some sort to force the compiler to squeeze out the empty space (at the expense of more complicated code)

I suppose this should work too:

Code:

---

WORD number=61001;
char* boke=new char[5*MAXPARTS];
WORD* arr;
boke[count*3]=i+1;
arr=(WORD*)((int)boke+count*3+1);
*arr=number;
//do something with it
delete[] boke;

---

Same with integer:

Code:

---

int number=548841;
char* boke=new char[5*MAXPARTS];
int* arr;
boke[count*5]=i+1;
arr=(WORD*)((int)boke+count*5+1);
*arr=number;
//do something with it
delete[] boke;

---

> int* arr;
> boke[count*5]=i+1;
> arr=boke+count*5+1;
> *arr=number;
Nope - that will get you a bus error (or unaligned access error) on some machines.
Did you get any warnings when you compiled it?

If you must pack ints into a char array, then you need to use memcpy() to be sure of it working.

No warnings - work excellenty (I changed the code a bit so it won't give me "couldn't convert" errors)
I've done exactly the same thing in assembly (where it is normal) before.
I think this is a very logical thing and I see no problem why it shouldn't work (integers don't always have to start at even memory addresses so I see no reason to get unaligned access error).

Good solid "works for me" logic - keep it up :rolleyes:

It seems as logical as 2+2=4 to me that it works.

So the gcc option of -Wcast-align is basically a waste of time in your opinion then, based on your limited experience?

Quote:

---

Originally Posted by Salem

So the gcc option of -Wcast-align is basically a waste of time in your opinion then, based on your limited experience?

---

I don't think that aligns memory... but I might be wrong.