Yes, see the followup.  My comments before were all misinterpreting size_t.

Same result on x86_64 linux. On a 64-bit platform the 24 byte structure now occupies 24 bytes instead of being padded to 32.  Nice.  On a 32-bit platform it should remain 16 bytes.

The PyGC_Head union structure is NOT part of the ABI laid out in http://www.python.org/dev/peps/pep-0384/ and is accurately excluded from the .h file when Py_LIMITED_API is defined so changing this in 3.4 should not be a problem.  This structure occupies the space gc tracked PyObject* pointers.

-gps


On Fri, Dec 14, 2012 at 12:42 AM, Mark Dickinson <dickinsm@gmail.com> wrote:
On Fri, Dec 14, 2012 at 7:27 AM, Gregory P. Smith <greg@krypto.org> wrote:
> So changing the definition of the dummy side of the union makes zero
> difference to already compiled code as it (a) doesn't change the structure's
> size and (b) all existing implementations already align these on an 8 byte
> boundary.

It looks to me as though the struct size *is* changed, at least on
some platforms.  Before this commit, I get (OS X 10.6, 64-bit
non-debug build):

Python 3.4.0a0 (default:b4c383f31881+, Dec 14 2012, 08:30:39)
[GCC 4.2.1 (Apple Inc. build 5664)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> class A(object): pass
...
>>> a = A()
>>> import sys
>>> sys.getsizeof(a)
64


After it:

Python 3.4.0a0 (default:76bc92fb90c1+, Dec 14 2012, 08:33:48)
[GCC 4.2.1 (Apple Inc. build 5664)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> class A(object): pass
...
>>> a = A()
>>> import sys
>>> sys.getsizeof(a)
56

--
Mark