Python 2.4 does not marshal infinity floating point properly under Win32

[Pierre Rouleau]

Hi all,

When using Python 2.4.x on a Win32 box,
marshal.loads(marshal.dumps(1e66666)) returns 1.0 instead of infinity
as it should and does under Python 2.5 (also running on Win32 ).

This problem was reported in another thread here by Peter Hansen
http://groups-beta.google.com/group/...16739705c9304f

Is this considered an important enough bug to fix it in Python 2.4?

The following script exercise the problem when executed::

#---------------------------test_marshal.py--------------------------------------------------
import marshal

INFINITY = 1e66666
inf = 1e300 * 1e300
nan = inf - inf
def show(title, msg):
print '%-10s : ' % title,
msg2 = ''
count = 0
for chr in msg:
count += 1
val = ord(chr)
print hex(val),
if 31 < val < 128:
msg2 += chr
else:
msg2 += '.'
print ' ' * (9-count), ' : ', msg2

show('Nan', marshal.dumps(nan))
show('Infinity', marshal.dumps(INFINITY))
show('Infinity', marshal.dumps(inf))
show('Infinity', marshal.dumps(1e666))

val = marshal.loads(marshal.dumps(1e666))
print val
val = marshal.loads(marshal.dumps(INFINITY))
print val
assert val==INFINITY
#-------------------------------------------------------------------------------------------------------

When running the script on Win32 box with Python 2.4.3 or 2.4.4 I get::

D:\dev\python\test>t_marshal
Nan : 0x66 0x7 0x2d 0x31 0x2e 0x23 0x49 0x4e 0x44 :
f.-1.#IND
Infinity : 0x66 0x6 0x31 0x2e 0x23 0x49 0x4e 0x46 : f.1.#INF
Infinity : 0x66 0x6 0x31 0x2e 0x23 0x49 0x4e 0x46 : f.1.#INF
Infinity : 0x66 0x6 0x31 0x2e 0x23 0x49 0x4e 0x46 : f.1.#INF
1.0
1.0
Traceback (most recent call last):
File "D:\dev\python\test\t_marshal.py", line 33, in ?
assert val==INFINITY
AssertionError

But when running the same script on a Win32 box, using Python 2.5, I
get::

C:\dev\python\test>t_marshal
Nan : 0x67 0x0 0x0 0x0 0x0 0x0 0x0 0xf8 0xff : g........
Infinity : 0x67 0x0 0x0 0x0 0x0 0x0 0x0 0xf0 0x7f : g.......¦
Infinity : 0x67 0x0 0x0 0x0 0x0 0x0 0x0 0xf0 0x7f : g.......¦
Infinity : 0x67 0x0 0x0 0x0 0x0 0x0 0x0 0xf0 0x7f : g.......¦
1.#INF
1.#INF
#-------------------------------------------------

--

Pierre Rouleau

[Grant Edwards]

On 2006-11-30, Pierre Rouleau <pr*********@gmail.comwrote:

When using Python 2.4.x on a Win32 box,
marshal.loads(marshal.dumps(1e66666)) returns 1.0 instead of infinity
as it should and does under Python 2.5 (also running on Win32 ).

This problem was reported in another thread here by Peter Hansen
http://groups-beta.google.com/group/...16739705c9304f

Is this considered an important enough bug to fix it in Python 2.4?

This must be at least the third thread on this topic so far
this week. And I didn't start _any_ of them (this week).

:)

--
Grant Edwards grante Yow! Imagine--a WORLD
at without POODLES...
visi.com

[Martin v. Löwis]

Pierre Rouleau schrieb:

Is this considered an important enough bug to fix it in Python 2.4?

To the contrary - it's not considered a bug at all. Python didn't
make any promises about "unregular" floating point values, so things
like that just may happen.

The fix that made Python offer stronger guarantees in 2.5 cannot
be ported back to 2.4, since it would break the marshal format of
that release.

Finally, it's not very likely that *any* additional 2.4.x releases
are made at all; only 2.5.x is still actively maintained.

Regards,
Martin