Are structs sequential in memory

[Supadude]

If I define a struct as such:

struct mystruct
{
    char one;
    char two;
    char three;
    char four;
};

Is it guarenteed to be sequential in memory such that I can do an assignment like this:

mystruct s;
char *p = (char *) &s;

And I can access and treat the struct as sequential bytes? Will this work, will it always work, and is it even a good idea to do? The reason is for sending large packets of serial data, I would rather format the packet as a struct and then change the struct to an array of bytes to be sent out.

[zacs7]

Most operating systems add padding to structs (to align the struct and a power of 2 (faster)) in memory, some also might order them differently.

[Supadude]

So this shouldnt be done then. Is there any easy way to format large serial packets so that each byte doesnt have to be masked and packed individually?

[robatino]

It's true that in general, there may be padding between struct members. However, in the special case where all the members are the same type, I'm not sure whether the Standard allows it to happen.

> some also might order them differently.
Are you sure? I thought there was a guarantee this couldn't happen, but am not absolutely sure.

[zacs7]

Depends on how the OS decides to allocate memory. Although I think your right, they won't reorder the struct. But they can add padding to the end.

So this shouldnt be done then. Is there any easy way to format large serial packets so that each byte doesnt have to be masked and packed individually?

Yes, use a payload sort of scheme.

[Salem]

The order is always as written.

The gaps between them for padding and alignment however can vary.

Also, there is never any padding between the start of the struct and the first member.

[robatino]

Are you sure there's no guarantee that there's no padding if all the struct members are the same type? The reason I ask is that such a struct could be treated like an array of the given type (for example the OP's example could be treated like an array of 4 chars) and array elements are guaranteed to be contiguous in memory. It seems surprising that the Standard wouldn't bother to give a guarantee if it's so easy to do.

[kryptkat]

how do you find out how much padding there is?

[robatino]

how do you find out how much padding there is?

You can print the member addresses:

#include <stdio.h>

struct A {
  char x;
  int y;
};

int main(void)
{
  struct A a;
  printf("address of a.x is %p\n", &a.x);
  printf("address of a.y is %p\n", &a.y);
  return 0;
}

Output on my box:

address of a.x is 0xbf9b716c
address of a.y is 0xbf9b7170

The difference is 4 so there are 3 bytes of padding between a.x and a.y.

[brewbuck]

And I can access and treat the struct as sequential bytes? Will this work, will it always work, and is it even a good idea to do? The reason is for sending large packets of serial data, I would rather format the packet as a struct and then change the struct to an array of bytes to be sent out.

If you are willing to restrict yourself to a specific compiler or set of compilers, there are usually compiler-specific (read "non-portable") ways to tell the compiler to pack the struct as tight as possible -- no padding between members or at the end of the struct.

Portably, the only thing you can rely on is that the order of the struct members in memory is generally the same as the order they are defined in the struct, although padding may be added between members. So in general you can't assume a struct is precisely equivalent to an unpadded sequence of bytes.

Rather than invoke compiler-specific directives, I would recommend just biting (byting?) the bullet and writing some basic routines to do PORTABLE input/output of various data types. Write a function which handles chars (trivial), a function for 2 byte quantities (usually shorts, but not guaranteed), a function for 4 byte quantities, etc. It is possible to make these macros instead of functions, but with caveats.

As always, the three main issues are packing/padding, endianness, and signed integer representation. The general "science" of how to portably input and output binary data structures is called "marshaling."

[vart]

how do you find out how much padding there is?

use sizeof struct - sum sizeof members

[Supadude]

Yes, use a payload sort of scheme.

Could you elaborate on this?

[Kennedy]

Not sure how portable this is (I know that it works for 16-bit DOS, Linux, and Windows XX):

__attribute__((aligned (1),packed))

Will align your struct at 1 byte offsets.

[kawk]

Depends on how the OS decides to allocate memory. Although I think your right, they won't reorder the struct. But they can add padding to the end.

If the OS supports the Berkley Network Package (sockets) is has do to it in the same order, always, because of one of the structs (can't remember which, i'll look at beej).

EDIT: It's struct sockaddr, heres the link: http://beej.us/guide/bgnet/output/ht...t.html#structs

[Perspective]

"I have a bunch of elements of the same type that I want to store in contiguous memory..."

I don't understand why this converstation is even about structs, use an array.