Talk:REAL 64 (interface)

There are problems with some of the comparison features in the presence of NaNs.

For instance, the postcondition of <=.

We need to solve the equality issue first, before the best contracts can be arrived at (i.e. we may be able to do better than just make all the postconditions apply only if neither value is NaN - an alternative approach might be to exclude the possibility of NaNs in the precondition, but this might make life too awkward for the clients. But NaNs make life awkward, so maybe that is the best way).