There has been disagreement in the P1788 group for some time about the proper status of interval formats such as mid-rad, that are not based on storing the lower and upper bounds. There has also been criticism of the current draft text as too prescriptive: why go beyond a minimal specification that ensures the Fundamental Theorem of Interval Arithmetic (FTIA) holds? In response there have been discussions over several weeks, mainly between Corliss, Hayes, Kearfott, Keil, Pryce and Zuras. These have produced a revised approach to the standard. I present it here and aim to propose it as a motion shortly.
The basic change is to bring in the concept of a level 2 “interval datatype”, which can be either “explicit” or “implicit”. I believe this gives a structure where
All interval datatypes are equal,
in a way that satisfies the proponents of mid-rad representation. It may be that “some are more equal than others”, but this is due to intrinsic differences between implicit and explicit types.
If this, or something like it, passes, the draft standard text will need substantial revision. However most of this consists in rearranging existing text, and replacing some terms by new ones to make concepts more precise.
I am hastening to get something out before I go on holiday 13–27 July, so I submit this as work in progress. I believe the basic structure is sound but various issues need further discussion before there is a firm motion.