6

Assuming I have a struct like the following:

struct Struct {
    char Char;
    int Int;
};

and sizeof( int ) is greater than one and the compiler adds padding for the Char member variable - is the compiler-generated code allowed to change the values of the padding bytes?

I mean if I use pointer arithmetic and write some data into the padding bytes surrounding the Char member variable and later do variable.Char = assignment is it possible that the code generated by the compiler will also overwrite some of the padding bytes?

sharptooth
  • 167,383
  • 100
  • 513
  • 979
  • 1
    @nobugz, I think it is a valid question. Image you have memset a structure to 0 before setting the values, than you can use memcmp later to compare two structures, but only if the compiler does not touch the padding bytes. – quinmars Jan 06 '10 at 14:29
  • Don't mess with padding, `memset` or `memcpy` with classes and structures. The **safer** and **more robust** approach is to implement methods for reading from and writing to a packed buffer (e.g. `unsigned char[]`). The insurance is worth the extra executable space and time. – Thomas Matthews Jan 06 '10 at 17:44
  • @Thomas Matthews, I think you misunderstood me. I was talking about comparing two structures, see http://stackoverflow.com/questions/141720/how-do-you-compare-structs-for-equality-in-c/141791#141791 – quinmars Jan 06 '10 at 19:08
  • If you are concerned about byte alignment, most compilers let you declare it explicitly. For example in VS, you can use #pragma pack(1) to tell the compiler to pack your structure exactly as declared with no additional padding. In this case, sizeof(Struct) would be exactly sizeof(char)+sizeof(int), and nothing more. The only valid use I have ever seen for using this mechanism is socket programming, tho I'm sure there are other reasons. – John Dibling Jan 06 '10 at 22:32

4 Answers4

13

What if the compiler were smart enough to use a word write to save the char? Your carefully saved data would be lost. ;-)

Richard Pennington
  • 19,673
  • 4
  • 43
  • 72
10

The following sentence is wrong: No, it would not overwrite the padding bytes. But it probably is not a good practice to use that. If you need it, add member variables there.

I researched based on comments indicating (correctly) that I am stupid:

The C Standard has an "Annex J" with section J.1 Unspecified behavior. It says, "The value of padding bytes when storing values in structures or unions". The implication is that the compiler can generate whatever instructions it wants to write the data into the structure, which may allow it to overwrite padding after a member.

Mark Wilkins
  • 40,729
  • 5
  • 57
  • 110
4

You sure can write something there, and memset-ing an instance of such structure does that. However, it's not safe and never a good idea to do so. Some other day another developer puts a #pragma somewhere or adds a member to struct declaration and your code will explode in many weird and fancy ways, which could take quite a while to debug.

Dmitry
  • 6,590
  • 2
  • 26
  • 19
1

The only reason to do this would be something like a plugin evilly tricking a host application into storing extra data.

Don't do it though, because at some point in the future it will break, and it will be a severe headache for all concerned.

James
  • 24,676
  • 13
  • 84
  • 130