[Python-bugs-list] [ python-Bugs-525705 ] [2.2] underflow raise OverflowException

[noreply@sourceforge.net]

Bugs item #525705, was opened at 2002-03-04 23:56
You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=105470&aid=525705&group_id=5470

Category: Python Interpreter Core
Group: Platform-specific
Status: Open
Resolution: None
Priority: 7
Submitted By: Huaiyu Zhu (hzhu)
Assigned to: Tim Peters (tim_one)
Summary: [2.2] underflow raise OverflowException

Initial Comment:
Python 2.2 (#1, Mar  4 2002, 15:25:20) 
[GCC 2.95.2 19991024 (release)] on linux2
>>> 1e-200**2
OverflowError: (34, 'Numerical result out of range')

The machine runs SuSE 7.1. $ rpm -q gcc glibc
gcc-2.95.2-149
glibc-2.2-7

On this machine, the following test C program produces
1e+200   inf 34 Numerical result out of range 34
1e-200     0 34 Numerical result out of range 34
without -lieee, and 

1e+200   inf 0 Success 34
1e-200     0 0 Success 34
with -lieee.

However, editing pyconfig.h.in to add line
#define HAVE_LIBIEEE 1
does not change Python's behavior.

This problem does not occur on another machine, running
Red Hat 6.1 upgraded to 7.1.  It exhibits exactly the
same behavior regarding C, but Python underflows to 0.0
properly. 


The test C program is

#include <math.h>
#include <stdlib.h>
#include <errno.h>
extern int errno;

main() {
  double x;
  x = 1e200;
  printf("%g %5g %d %s %d\n", x, pow(x, 2), errno,
strerror(errno), ERANGE);
  x = 1e-200;
  printf("%g %5g %d %s %d\n", x, pow(x, 2), errno,
strerror(errno), ERANGE);
}


----------------------------------------------------------------------

>Comment By: Jeremy Hylton (jhylton)
Date: 2002-03-09 05:45

Message:
Logged In: YES 
user_id=31392

Tests work as expected on my Linux box (originally a RedHat 
6.2 system, with a bunch of upgrades, including glibc 
2.1).  The tests show the same results regardless of 
whether I link with -lieee.


----------------------------------------------------------------------

Comment By: Tim Peters (tim_one)
Date: 2002-03-09 05:00

Message:
Logged In: YES 
user_id=31435

I'm hoping this is fixed in CVS now.  Since we're never 
going to sort out the Linux config mess, it tries to worm 
around it via more macros.

Include/pyport.h; new revision: 2.45
Modules/cmathmodule.c; new revision: 2.28
Objects/floatobject.c; new revision: 2.111

----------------------------------------------------------------------

Comment By: Tim Peters (tim_one)
Date: 2002-03-09 03:43

Message:
Logged In: YES 
user_id=31435

Python used to link with -lieee.  Rev 1.139 of configure.in 
changed that.  The checkin comment is:

"""
Gregor Hoffleit: Don't link with the libieee library if 
it's not necessary
"""

Thanks, Gregor <wink/sigh>.

----------------------------------------------------------------------

Comment By: Tim Peters (tim_one)
Date: 2002-03-09 03:40

Message:
Logged In: YES 
user_id=31435

Assigned to me, because the Linux weenies are apparently 
too frightened of radix points to get near this <wink>.

----------------------------------------------------------------------

Comment By: Tim Peters (tim_one)
Date: 2002-03-06 00:01

Message:
Logged In: YES 
user_id=31435

Changed Group to "Platform specific" and added "[2.2]" to 
the Summary line instead.

----------------------------------------------------------------------

Comment By: Tim Peters (tim_one)
Date: 2002-03-05 00:48

Message:
Logged In: YES 
user_id=31435

Boosted priority.  I'd assign to me, except I don't know 
anything about Linux config.

Paul Rubin reported success (no more bogus OverflowError) 
after

"""
I manually put "-lieee" into the LIBS=... line in the 
makefile and rebuilt.
"""

----------------------------------------------------------------------

You can respond by visiting: 
http://sourceforge.net/tracker/?func=detail&atid=105470&aid=525705&group_id=5470