[Numpy-tickets] [NumPy] #582: Crash when embedded using Borland Builder 6

[NumPy]

#582: Crash when embedded using Borland Builder 6
------------------------+---------------------------------------------------
 Reporter:  thomas      |       Owner:  somebody
     Type:  defect      |      Status:  new     
 Priority:  normal      |   Milestone:          
Component:  numpy.core  |     Version:  none    
 Severity:  major       |    Keywords:          
------------------------+---------------------------------------------------
 Hi,

 when used in an embedded scenario with some non-g++ compilers, every call
 of "import numpy" leads to a crash (Floating Point Overflow) because of
 the following function in umathmodule.c:

 static double
 pinf_init(void)
 {
  double mul = 1e10;
  double tmp = 0.0;
  double pinf;
  pinf = mul;
  for (;;) {
  pinf *= mul;
  if (pinf == tmp) break;
  tmp = pinf;
  }
  return pinf;
 }

 When pinf supercedes 1E300, g++ will turn it to infinity, whereas other
 compilers (e.g. Borland C++ Builder) will throw an overflow exception.

 My recommendation is to directly generate the double value for infinity by
 IEEE 754 specification:

   char inf_string[9] = "\x00\x00\x00\x00\x00\x00\xF0\x7F";
   double pinf = ((double*)inf_string)[0];

 on little endian architectures, and

   char inf_string[9] = "\x7F\xF0\x00\x00\x00\x00\x00\x00";
   double pinf = ((double*)inf_string)[0];

 on big endian architectures.

 Thank you!

-- 
Ticket URL: <http://scipy.org/scipy/numpy/ticket/582>
NumPy <http://projects.scipy.org/scipy/numpy>
The fundamental package needed for scientific computing with Python.