homepage

When I use Py_UNICODE_ISSPACE() in my C++ extension, I got following error.
test.obj : error LNK2019: unresolved external symbol
"__declspec(dllimport) unsigned char const * const _Py_ascii_whitespace"
(__imp_?_Py_ascii_whitespace@@3QBEB) referenced in function "struct
_object * __cdecl fff(struct _object *,struct _object *)"
(?fff@@YAPAU_object@@PAU1@0@Z)
Py_ascii_whitespace defined in unicodeobject.h should be enclosed by
'extern "C" {' block for C++ support.
Tested with Python 2.6.1/VS2008 express.

On 2009年01月05日 17:09, Atsuo Ishimoto wrote:
> New submission from Atsuo Ishimoto <ishimoto@gembook.org>:
> 
> When I use Py_UNICODE_ISSPACE() in my C++ extension, I got following error.
> 
> test.obj : error LNK2019: unresolved external symbol
> "__declspec(dllimport) unsigned char const * const _Py_ascii_whitespace"
> (__imp_?_Py_ascii_whitespace@@3QBEB) referenced in function "struct
> _object * __cdecl fff(struct _object *,struct _object *)"
> (?fff@@YAPAU_object@@PAU1@0@Z)
> 
> Py_ascii_whitespace defined in unicodeobject.h should be enclosed by
> 'extern "C" {' block for C++ support.
Agreed. There already is such a block in unicodeobject.h, so perhaps
a little rearranging could do the trick.

I believe attached issue4846.diff should fix the problem, but I don't 
have access to the affected platform (Windows?), so I did not test it.
I also wonder whether it would be more appropriate to redefine PyAPI_* 
macros to declare API symbols with extern "C" when compiled under C++? 
This seems to be a better approach than cluttering all header files with 
#ifdef __cplusplus and worrying about the scope of extern "C" {} 
wrappers.
Note that I considered whether _Py_ascii_whitespace declaration should 
be done inside if !defined(HAVE_USABLE_WCHAR_T) .. directive and 
concluded that it should not. My reasoning is as follows: _Py_ascii_whitespace must be defined unconditionally in the .c file 
because when python core is compiled, it is not known how an extension 
module would define HAVE_USABLE_WCHAR_T. While it is not strictly 
necessary to bring _Py_ascii_whitespace definition in an extension 
module that has "USABLE_WCHAR_T", there is little to be gained from 
hiding it because it is visible to the linker. Thus for the common uses 
there is little practical difference between the two options but 
unconditional declaration requires less code. On the other hand in the 
context of issue4805, having unconditional declaration would allow 
removing extern "C" wrappers from unicodeobject.c without loosing C++ 
compilability of python core.

You beat me by 10 minutes, Alexander, otherwise I'd have had a similar
patch, except moving the declaration. I agree with your rationale
though, so yours is even better.
Concerning including the extern "C" in the PyAPI_* macros, I had the
same idea and think that that it would unclutter code a bit.

Not that it would matter on a two-line patch, but I find it useful to
add myself to the "nosy" list without any comments as so as I find the
issue interesting even if I have nothing to contribute at the moment.
This way I am more likely to notice when someone posts a patch before
duplicating the effort.
> Concerning including the extern "C" in the PyAPI_* macros, I had the
> same idea and think that that it would unclutter code a bit.
This would be a bigger change and getting it right would require
expertise with non-standard declarations (__declspec(..) etc) that I
don't have. Ulrich, you seem to be a Windows expert, can you suggest
a patch?

Fixed in r68344 on trunk.
Backport and forward-port candidate.

On 2009年01月05日 18:54, Alexander Belopolsky wrote:
> I also wonder whether it would be more appropriate to redefine PyAPI_* 
> macros to declare API symbols with extern "C" when compiled under C++? 
> This seems to be a better approach than cluttering all header files with 
> #ifdef __cplusplus and worrying about the scope of extern "C" {} 
> wrappers.
This is standard practice, but of course has to be done with some
care.
In general, only macro and type definitions should live outside
the extern "C" sections, #includes must live outside those sections.
Changing the PyAPI_* macros would help with this, but I'm not sure whether
that's a catch-all solution. If it's not, then it's better to stick
with the extern "C" sections.

Actually, providing the patch for PyAPI_Data/PyAPI_Func wouldn't be so 
hard, but there are lots of headers that need to be changed, i.e. all 
their existing extern "C" clauses removed. What I would do is build 
the Python shared library as it is and then use e.g. depends.exe 
(www.dependencywalker.com) to make sure the symbols before and after 
are the same.
That's a different issue though and it's not pressing. Is it possible 
to clone this issue and make it a feature request (Include extern"C" 
in PyAPI_* macros)?