Structural Padding??

Dear All,
I am writing program to avoid structural padding as follows using attributes
but i am getting error in this code...

Code:

---


#include "stdlib.h"
int main()
{
  struct data
  {
    int a;
    char ch;
    float s;
  }__attrubute__((packed));

  struct data e;
  printf("%u %u %u\n", &e.a, &e.ch,&e.s);
  printf("%d\n", sizeof(e));
  return 0;
  }

---

errors are:-

1. error C2065: 'packed' : undeclared identifier
2. error C2440: 'initializing' : cannot convert from 'int' to 'struct main::data'

If it's like you other question, you're using the wrong compiler.

> I am writing program to avoid structural padding as follows using attributes
Why?

Bearing in mind that most such reasons are ultimately flawed approaches to the problem.
For example
- mapping the header of a file such as BMP
- mapping to a hardware register
etc etc

Microsoft compilers use #pragma pack if I'm not mistaken.

Yeah, GNU C compilers do no support #pragma. What you can do is

Code:

---


 struct data
  {
    int a;
    char ch;
    float s;
  }__attrubute__((_packed_));

---

Or you can work with your data directly within the struct, but you must know your compiler in how it will operate within the word boundary of your target system.

Or:

Code:

---

struct data
  {
    int a;
    char ch;
    float s;
  }
#ifdef __GNUC__
    __attrubute__((_packed_));
#elif defined(MSC)  /* not sure exactly what this would be */
    ...
#else
    #error "Cannot eliminate structure padding"
#endif
;

---

Or something like that. (I think MSVC's #pragma has to go before the structure.)

Code:

---

#ifdef _MSC_VER
#pragma pack(push, r1, 1)
#endif
struct data
{
    int a;
    char ch;
    float s;
}
#ifdef __GNUC__
    __attrubute__((_packed_));
#elif _MSC_VER
#pragma pack(pop, r1)
#else
    #error "Cannot eliminate structure padding"
#endif
;

---

I think that's how it's done.

Code:

---

#elif _MSC_VER

---

Bad idea -- you'll get an error if _MSC_VER is not defined.

Code:

---

#elif defined(_MSC_VER)

---

is better, or, if you're really testing the truth of _MSC_VER (which I doubt you want),

Code:

---

#elif defined(_MSC_VER) && _MSC_VER

---

Other than that . . . it looks like it might work.

Ah, of course. But _MSC_VER is guaranteed to be defined if it's a Microsoft compiler so you only really need to check if it's defined.

What if it's a non-MS (and non-GCC) compiler? You'll get an error like "_MSC_VER not defined" instead of my nice #error. :)

BTW: this works on GCC. That is, it prints 2. (Care to test it on MSVC?)

Code:

---

#ifdef _MSC_VER
#pragma pack(push, r1, 1)
#endif
struct data
{
    char one, two;
}
#ifdef __GNUC__
    __attribute__((__packed__));
#elif defined(_MSC_VER)
#pragma pack(pop, r1)
#else
    #error "Cannot eliminate structure padding"
#endif
;

#include <stdio.h>

int main() {
    printf("&#37;d\n", (int)sizeof(struct data));
    return 0;
}

---

Note how I had to change the GCC line:

Code:

---

    __attribute__((__packed__));

---

attribute and double underscores surrounding packed.

[edit] It's kind of funny how many times that buggy code was copy-and-pasted without anyone actually reading it . . . . [edit]

It prints 2 on msvc.

For some reason, it prints 2 even when I enable structure padding (with GCC). I guess structure padding doesn't apply to structures consisting entirely of chars, on this system at least.

So here's another one to test if you like . . . on my 64-bit system, it prints 16 with structure padding enabled and 9 with it disabled.

Code:

---

#ifdef _MSC_VER
#pragma pack(push, r1, 1)
#endif
struct data
{
    char one;
    double two;
}
#ifdef __GNUC__
    __attribute__((__packed__))
#elif defined(_MSC_VER)
#pragma pack(pop, r1)
#else
    #error "Cannot eliminate structure padding"
#endif
;

#include <stdio.h>

int main() {
    printf("&#37;d\n", (int)sizeof(struct data));
    return 0;
}

---

There's another change, too: the semicolon was removed from the GCC declaration, to remove this warning:

Code:

---

structpad.c:16: warning: ISO C does not allow extra ‘;’ outside of a function

---

Quote:

---

Originally Posted by dwks

So here's another one to test if you like . . . on my 64-bit system, it prints 16 with structure padding enabled and 9 with it disabled.

---

Yup, same here, even on a 32-bit system.

Right, it would depend on the size of words and of a double. Just thought I'd mention it.

So there you have it: revision 4 or whatever of our GCC+MSVC non-padded structure. ;)