I found a really weird comportment with NANs:

>>> from math import nan, isnan, inf
>>> nan==nan
False
>>> nan!=nan
True
>>> a=nan  # maybe get another instance would fix it (or so I thought)
>>> a==nan
False
>>> a==a
False
>>> a is a  
True
>>> # because `is` works, it could be a
>>> # deliberate comportment. is it?
>>> a is nan
True
>>> isnan(a) and isnan(nan)
True
>>> nan == -nan
False
>>> nan is -nan
False
>>> a < 1
False
>>> a > 0
False
>>> a < inf
False
>>> b=external_pyd_call_that_returns_nan()
>>> isnan(b)
True
>>> b == nan
False

that sums what I tried up.

thanks for reading this.

That's how IEEE 754's NaN is defined. Seem to be behaving according to the standard.

Yes, it's both intended and annoying ;-)  The standard specifies that, by default, comparisons involving NANs are "unordered":  a NAN is _none_ of "less than", "equal to", or "greater than" any other float, including any other NaN, and including itself.