Two problems with float.fromhex:

(1) it incorrectly rejects trailing whitespace on infs and nans:

>>> float.fromhex('a.0 ')
10.0
>>> float.fromhex('nan ')
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
ValueError: invalid hexadecimal floating-point string

(2) The implementation uses PyOS_strnicmp for detecting nans and infs,
which is locale-aware.  A locale-unaware case-insensitive comparison 
should be used instead.  I don't know of any locale where this actually 
makes a difference, but I think it should be fixed anyway.

Here's a patch.

Fixed in r72564 (trunk), r72565 (py3k).  The trailing whitespace bugfix 
was backported in r72566 (2.6) and r72567 (3.0), but those versions still 
use locale-aware isspace and islower.

As a semi-accidental by-product, the fix also changes the behaviour of 
float.fromhex('-nan'): previously, float.fromhex('-nan') and 
float.fromhex('nan') had the same sign;  now they have opposite signs. Sane code that's affected by the sign bit of a nan should be extremely 
rare, so I don't think this change is going to bother anyone.