Logged In: YES 
user_id=31435

Sorry, all behavior wrt infinities (and NaNs, and signed zeroes) 
is a platform- and release-dependent accident.  I've added a 
link to this report in PEP 42, under the "Non-accidental IEEE-
754 support" feature request heading.

On most platforms you can worm around your specific problem 
by doing

INFINITY = 1e200 * 1e200

instead.

The business about how infinities display, and whether they 
can be read back in, is another batch of accidents inherited 
from the platform C I/O routines.