[Numpy-discussion] Distutils: using different linker options for c++ and c code

[David Cournapeau]

Robert Kern wrote:
>
> Quite possibly. Can you run the segfaulting code in a debugger so we
> can try to isolate the actual cause? It is possible that we can patch
> it up to work with msvcr71.

I finally managed to do something, for reference:
    - I hacked distutils to put -g everywhere (using the --debug in 
distutils lead to nowhere for me: the debugging library is not ABI 
compatible so just copying python25 to python25_d does not work, and I 
can't rebuild python in debug mode since I don't have VS 2003)
    - I rebuilt everything (scipy only)
    - I used drmingw as the so called JIT debugger (registering it with 
drmingw -i): now, when scipy.sparse crashes, I get into the code.


python.exe caused an Access Violation at location 66d571c1 in module 
_bsr.pyd Reading from location 015d6000.

Registers:
eax=015d5fe8 ebx=00000006 ecx=00000002 edx=00000012 esi=00000006 
edi=0000000a
eip=66d571c1 esp=0021c488 ebp=0021c4bc iopl=0         nv up ei ng nz ac 
po cy
cs=001b  ss=0023  ds=0023  es=0023  fs=003b  gs=0000             
efl=00000297

Call stack:
66D571C1  _bsr.pyd:66D571C1  void csr_binop_csr<int, short, 
std::plus<short> >(int, int, int const*, int const*, short const*, int 
const*, int const*, short const*, int*, int*, short*, std::plus<short> 
const&)  csr.h:663
    ...
                        nnz++;
                    }
 >                    B_j = Bj[++B_pos];
                }
            }
    ...

66D472B0  _bsr.pyd:66D472B0  void bsr_binop_bsr<int, short, 
std::plus<short> >(int, int, int, int, int const*, int const*, short 
const*, int const*, int const*, short const*, int*, int*, short*, 
std::plus<short> const&)  bsr.h:550
    ...
            }
   
 >            Cp[i+1] = nnz;
        }
    }
    ...

66D3934C  _bsr.pyd:66D3934C  void bsr_plus_bsr<int, short>(int, int, 
int, int, int const*, int const*, short const*, int const*, int const*, 
short const*, int*, int*, short*)  bsr.h:580
    ...
                            I Cp[],       I Cj[],       T Cx[])
    {
 >        
bsr_binop_bsr(n_row,n_col,R,C,Ap,Aj,Ax,Bp,Bj,Bx,Cp,Cj,Cx,std::plus<T>());
    }
   
    ...

66D01087  _bsr.pyd:66D01087  _wrap_bsr_plus_bsr  bsr_wrap.cxx:1226
    ...
    SWIG_Py_Void(void)
    {
 >      PyObject *none = Py_None;
      Py_INCREF(none);
      return none;
    ...

1E08E137  python25.dll:1E08E137  PyCFunction_Call
1E1E7A10  python25.dll:1E1E7A10  PyTuple_Type

So it *may* be unrelated to msvc runtime; maybe it just caused a problem 
which has always been there. I will try to see if valgrind says anything 
useful for sparsetools under linux.

David