Thread: class layout guarantees

Originally Posted by matsp

I'm pretty sure that the compiler will always put data in the order you define it. There is, however, no guarantee that the data won't have gaps in it - for example:

Code:

struct s {
  int a;
  char b;
  int c;
};

The gap between b and c is probably going to be 3 bytes.

I would have thought just having one type (ie. u_char) would make padding a non-issue, but I'm not sure.

Your constructor has no say in the matter of which order the data comes.

No, but my question was more along the lines of: does having anything but a default-constructor make the class a non-POD type? I've remember read somewhere that it does. If it becomes a non-POD type, does that mean the layout isn't guaranteed?

Of coruse, if you have virtual function(s) in your class, then there will be a vtable entry somewhere along with your data -usually the first 4 or 8 bytes.

--
Mats

Oh yeah, I suppose where that comes is just random too...

This is starting to look like a dead end.

Originally Posted by matsp

What problem are you actually trying to solve?

To exand what I said in the first post:

Code:

Header hdr;
read( reinterpret_cast<u_char*>(hdr) );
// Everything in it's right place now
// use hdr here

That means that the reinterpret_cast<u_char*>(hdr) would have to yield an array of unsigned chars to the read function, where the layout of the struct matches that of the expected array. Eg.

Code:

struct Header { u_char a; u_char b; } hdr;
// read, as above, expecting an array like: u_char arr[] = { 'a', 'b' };
hdr.a == 'a';
hdr.b == 'b';

From what everyone says, this is UB, so I'll have to do this another way. Gah.

However, just FYI, this part seems to work fine using MSVC8 and gcc 4.* on linux. On the other hand, when I try and create a structure like the one I decoded, MSVC seems to not like it and alters the struct in some way (I'm only 60% sure that's what's happening, since I haven't bothered looking at the assembly).