[C++-sig] Boost.Python appears to be catching Fortran READ "exceptions".

[Mike]

Hi,

I am a new Boost.Python user.


Summary
---------
I have created a Boost.Python module (named 'hello') that calls a FORTRAN
subroutine (named FortDll) via a C++ function (named call_fortran ).  The
FORTRAN code, which reads a data file, has the line:
    READ(....,END=999)    ! goto line labled 999 when eof reached
to handle the end of file condition.  In non-Python environment it works as
expected, but under Python the eof condition is being caught by Boost.Python
as a "RuntimeError: unidentifiable C++ exception".  If I create my module
without Boost.Python then I just get a crash.

Details
-------
I have the following Boost.Python module:

----Start of C++ code ----
// Using MS.NET 7.1.3088, Python 2.4, boost 1.31
// creates hello.dll
#include <boost/python.hpp>
extern "C" void __stdcall FortDll();
void call_fortran()
{
  FortDll();
}
using namespace boost::python;
BOOST_PYTHON_MODULE(hello)
{
    def("call_fortran", call_fortran);
}
----End of C++ code ----

The code for FortDll is as follows:

----Start of FORTRAN (F90) code ----
!
!Built using Intel(R) Fortran Compiler for MS.NET 2003, Version
8.0.1877.2003
!
SUBROUTINE FortDll
! Export subroutine into Dll
!DEC$ATTRIBUTES  DLLEXPORT, DEFAULT, STDCALL, DECORATE, ALIAS:'FortDll'
::FortDll
  INTEGER*4 I
  OPEN(UNIT=33, FILE='INTS.DAT', ACCESS='SEQUENTIAL', STATUS='OLD')
  DO WHILE(.TRUE.)
777 READ(33,*,END=999) I
   ! When the eof is reached the above line causes
   ! an exception in Python,so that .....
    PRINT *,I
  ENDDO
999 CONTINUE  ! <<<<<< ...this line isn't reached!!!
  PRINT *,'999 reached'
  CLOSE(33)
END SUBROUTINE

----End of FORTRAN (F90) code ----

The test data file ints.dat is
----Start of ints.dat----
      1
      2
      3
      4
---EOF----

The results from testing are as follows:

----Start of Python session in debugger----
Python 2.4 (#60, Nov 30 2004, 11:49:19) [MSC v.1310 32 bit (Intel)] on win32
Type "help", "copyright", "credits" or "license" for more information.
>>> import hello
>>> hello.call_fortran()
           1
           2
           3
           4
Traceback (most recent call last):
  File "<stdin>", line 1, in ?
RuntimeError: unidentifiable C++ exception
>>>
----End of Python session----

The problem appears at line labled 777 in the Fortran code.   As shown above
in the Python test case, when the FORTRAN hits a read error it appears
that the normal execution of code, as specified by
   READ(..., END= 999) I ! goto line 999 when eof is reached
is not getting a chance to be executed.
When I catch the exception in the debugger, I get
    First-chance exception at 0x00b58ca9 (FortDll.dll) in python.exe:
0xC0000005: Access violation reading location 0x00000000

My initial thought that this was to do with Boost.Python, however I get a
crash/core dump
using a Python module built around:
static PyObject * call_fortran(PyObject *self, PyObject *args)
{
  FortDll(); // causes Python to crash when eof is reached.
  return Py_BuildValue("i", 0);
}


Note that in a 'non-Python' environment, similar to this:
int main(int ,char** ) {
    FortDll( );
}
the output is :

           1
           2
           3
           4
   999 reached

ie. FortDll() works as expected.

My app I want to prototype in Python has several C++ classes that rely on
some routines written in FORTRAN.  Can I stop Python handling the errors
that would have been handled in any case by the FORTRAN run time system?

Regards
Mike