[Python-Dev] [Python-checkins] cpython: Issue #14779: Do not use get_config_var('SIZEOF_VOID_P') on OS X 64-/32-bit

[Stefan Krah]

"Martin v. L?wis" <martin at v.loewis.de> wrote:
[http://bugs.python.org/issue14779]
>> -        self.sizeof_void_p = get_config_var('SIZEOF_VOID_P')
>> +        self.sizeof_void_p = get_config_var('SIZEOF_VOID_P') \
>> +                                if sys.platform != 'darwin' else None
>>           if not self.sizeof_void_p:
>> -            self.sizeof_void_p = 8 if architecture()[0] == '64bit' else 4
>> +            self.sizeof_void_p = 8 if sys.maxsize>  2**32 else 4
>>
>
> Why not unconditionally use sys.maxsize?

Because the tests need sizeof(void *). In an array with suboffsets void
pointers are embedded at the start of the array.

The C standard doesn't guarantee sizeof(void *) == sizeof(size_t). In
fact, there are machines where sizeof(void *) > sizeof(size_t):

http://comments.gmane.org/gmane.comp.programming.garbage-collection.boehmgc/651
http://www-01.ibm.com/support/docview.wss?uid=swg27019425


If you change pyconfig.h to 128 bit pointers while leaving sizeof(size_t)
and sizeof(ssize_t) at 8, pyport.h by itself doesn't catch the mismatch.

/* The size of `uintptr_t', as computed by sizeof. */
#define SIZEOF_UINTPTR_T 16
/* The size of `void *', as computed by sizeof. */
#define SIZEOF_VOID_P 16


However, now that I tried to compile Python with that pyconfig.h,
longobject.c *does* catch it:

Objects/longobject.c:943:5: error: #error "PyLong_FromVoidPtr: sizeof(PY_LONG_LONG) < sizeof(void*)"
Objects/longobject.c:970:5: error: #error "PyLong_AsVoidPtr: sizeof(PY_LONG_LONG) < sizeof(void*)"



If sizeof(void *) == sizeof(size_t) is the general assumption for compiling
Python, I think the test should happen prominently in either pyport.h or
Python.h.


> I'd also hard-code that sys.maxsize ought to be either 2**31-1 or 2**63-1.

I would have done exactly that, but the example in the docs that was quoted
to me in the issue uses > 2**32:

http://docs.python.org/dev/library/platform.html


Stefan Krah