Re: Structure alignment changes when a constructor or destructor is added

[Marc Glisse <marc dot glisse at inria dot fr>]

On Wed, 7 Dec 2011, Marc Glisse wrote:

> On Tue, 6 Dec 2011, Jonathan Wakely wrote:
>
> > On 6 December 2011 20:59, Bruce Fraser wrote:
> > > The following code was compiled with g++ 4.1.2 and again with g++
> > > 4.6.2 on 64-bit Linux (more details at the end). ?Note that S1 and S3
> > > have identical data members. ?S1 has a constructor defined, S3 does
> > > not. ?S2 inherits from S1 and S4 inherits from S2. ?The size and
> > > layout SHOULD be the same, but it is not as you can see from the
> > > output.
> >
> > For C++98 / C++03 there are no guarantees about class layout except
> > for POD types, and the presence of base classes and user-defined
> > constructors prevents your types being PODs.
> >
> > C++11 provides the more fine-grained concept of a standard-layout
> > struct, but even then I don't think there's any guarantee your S2 and
> > S4 will have the same layout.
>
> But g++ doesn't use a random layout, it (mostly) follows:
> http://sourcery.mentor.com/public/cxx-abi/abi.html
> so it makes sense to look at whether the layouts should be the same.
>
> It seems to define the layout differently depending on whether a type is POD 
> (in the C++03 sense).
>
> In particular, I note: "the C++ standard requires that compilers not overlay 
> the tail padding in a POD", but the ABI doesn't seem to impose any such 
> restriction on non-PODs.
>
> It does look peculiar that this incites you to artificially make your types 
> non-POD for a more compact layout...

And for std::tuple, this implies that:
struct A { long a; char b; };
sizeof(std::tuple<A,char>)==24
sizeof(std::tuple<char,A>)==24
because std::tuple has A as a member. If it had A as a base class, one of 
those would be 16.

The ABI is fun :-)

--
Marc Glisse