Python extensions using MinGW and CXX

[Reinhard Nadrchal]

Gerhard Häring wrote:

>Reinhard Nadrchal <reinhard at proceryon.at> wrote:
>  
>
>>To create the python import library I did the following (with my 
>>specific paths provided):
>>
>>pexports /C/WINDOWS/system32/python20.dll > python20.def
>>dlltool  --dllname /C/WINDOWS/system32/python20.dll --def python20.def 
>>--output-lib libpython20.a
>>and copied libpython20.a to the Python20/libs directory.
>>    
>>
>
>Try not to use any paths for python20.dll. I. e.:
>
>$ pexports python20.dll >python20.def
>$ dlltool --dllname python20.dll --def python20.def --output-lib libpython20.a
>
>You'll need to have copied python20.dll in your working directory first.
>
>I remember another user had weird problems when he used absolute paths and
>the above method worked for him.
>  
>
This doesn't change anything, here some excerpts from what the linker says:
<snip>
Info: resolving _PyFloat_Type by linking to __imp__PyFloat_Type 
(auto-import)
.
fu000002.o(.idata$3+0xc): undefined reference to `libpython20_a_iname'
.
nmth000000.o(.idata$4+0x0): undefined reference to `_nm__PyFloat_Type'
<snap>

>  
>
>>If I, on the other hand, link in python20.dll I get no errors and ld 
>>seems to be happy. The trouble starts when I execute 'import foo' from 
>>the python interpreter -  it simply crashes without  any comment.
>>
>>g++ -shared -lpython20 foo.o -ofoo.pyd
>>
>>That's the way I've tried it so far. I've also tried to link without the 
>>-shared option using -mdll,
>>    
>>
>
>Python's distutils uses -mdll, but I really don't know the meaning of
>either one in the win32 context :-(
>  
>
The only think I've seen so far is that you cannot mix -mdll and -shared.

>  
>
>>but in this case Python doesn't recognize 
>>the 'initfoo' entry point - maybe this is the key ?! Ho do I have to 
>>declare/define the init function ???
>>    
>>
>
>With DL_EXPORT:
>
>
>extern "C" {
>DL_EXPORT(void) initfoo() {
>    // ...
>}
>}
>#v-
>
This doesn't help; if I try c++ -mdll instead of c++ -shared the 
entrypoint (initfoo) is not recognized by python. By the way: the 
init... function is declared in a header (foo.h) and defined in the 
source file. Would it be ok to skip the declaration in the header (I 
think so) ???

>
>-- Gerhard
>  
>
Oh god, I'm afraid that we will have to use VC++ ;-) with some 60,000 
lines of c++ sourcecode written for gcc - or do you have any other ideas ???

reinhard