The current docs unnecessarily refer readers to the C99 standard. "integer constant representing the rounding mode used for arithmetic operations. This reflects the value of the system FLT_ROUNDS macro at interpreter startup time. See section 5.2.4.2.2 of the C99 standard for an explanation of the possible values and their meanings."

The docs should quote the standard, "The rounding mode for floating-point addition is characterized by the implementation defined value of FLT_ROUNDS: -1 indeterminable, 0 toward zero, 1 to nearest, 2 toward positive infinity, 3 toward negative infinity.  All other values for FLT_ROUNDS characterize implementation-defined rounding."
behavior.

Also, we should document the meaning of the other fields.  Several of them are not self explanatory:

min_10_exp = -307

   does not mean that 2.3e-308 isn't normalized

   does mean that 10**-307 is normalized and 10**-308 isn't


min_exp = -1021

   does not mean that 2**-1021 is the lowest normalized value

   does mean ldexp(0.5, -1021) == float_info.min


decimal_dig = 15

   does not mean digits required to specify every unique float

   does mean, "number of decimal digits, q, such that any 
   floating-point number with q decimal digits can be rounded
   into a floating-point number with p radix b digits and back
   again without change to the q decimal digits"

Python does not set the meaning and the value of these fields itself. It exposes C constants. Should we just copy definitions from corresponding C or floating-point standards?

Oh, seems it was what Raymond proposed initially.

Here's a link:  http://c0x.coding-guidelines.com/5.2.4.2.2.html