When using struct to unpack floats, I'm getting
inconsistent results when using the '<d' specification
instead of '@d' on a little-endian machine (Intel).
Here's a short python snippet that demonstrates the
problem.

import sys, struct

s = '\x00\x00\x00\x00\x00\x00\xf8\x7f'

>>> sys.byteorder
'little'

# This is correct...
>>> struct.unpack('@d', s)
(nan,)

# These should be equivalent for unpacking a single 
# double on little-endian arch... but they're not.
>>> struct.unpack('<d', s)
(inf,)
>>> struct.unpack('=d', s)
(inf,)

Logged In: YES 
user_id=6656

Can you try Python from svn HEAD?  Or did you?

Logged In: YES 
user_id=980271

I had tried with 2.3.5 and 2.4.1 and the bug was present in
both versions. I just tried with svn HEAD, and the bug is
fixed there (at least in the revision that I tried). 

It'd be nice if this bug fix could be included in the next
2.4 point release, assuming the fix is not too complicated.
Is there a process for nominating bugfixes for the main-2.4
branch?

Logged In: YES 
user_id=6656

I'm glad my fix worked.  I'm not personally inclined to port the fixes to the 2.4 
branch, as they are indeed fairly involved.