Strange result with math.atan2()

[Tim Peters]

[Vedran Furač]
> I think that this results must be the same:
>
> In [3]: math.atan2(-0.0,-1)
> Out[3]: -3.1415926535897931

Whether -0.0 and 0.0 are different floats internally depends on your
hardware floating-point; on most machines today, they are different
floats, but _compare_ equal to each other.

> In [4]: math.atan2(-0,-1)
> Out[4]: 3.1415926535897931

"0" and "1" are integer literals, and there is no current hardware on
which -0 produces a different bit pattern internally than 0.  Of
course you can _want_ them to be different, but you'll hope in vain
;-)

> In [5]: -0 == -0.0
> Out[5]: True

You'll also find that -0.0 == +0.0 returns True, despite that atan2
may treat them differently.  The IEEE 754 standard for floating-point
arithmetic requires that they compare equal, BTW.

> This is python 2.4.4c0 on Debian GNU/Linux.

Doesn't much matter.  The hardware may matter, though, and so may the
C library implementing atan2, and so may the set of option flags
passed to the platform C compiler when Python is compiled.  For
example, DEC Alpha hardware has little native support for signed
zeroes, and compilers for that box accept the HW limitations by
default.

Bottom line:  if you want atan2 to treat signed zeroes differently,
never use integer arguments (or results computed from integers), and
even then don't expect it to be portable.