Motion M0035.01 -- Overflow vs Unbounded: NO
I vote NO on Nate's motion on overflow families.
There is no specific aspect of this motion that could be changed to
make it acceptable to me: it proposes a wholesale replacement of the
underlying machinery of the standard with new logical entities, namely
sets of intervals to denote what is typically viewed as plain intervals
of three types: empty, bounded non-empty, and unbounded.
The conceptual complication is simply not warranted. The specific
problems it addresses can more easily (and understandably) be dealt
with by other means. I have no problem with functions being undefined
at one level, but defined at the next level where additional conceptual
or contextual information is available.
Whether any particular application depends on the presence or absence of
unbounded intervals can in my opinion be dealt with by the application
itself, perhaps checking that its inputs satisfy its constraints, and
reporting "invalid" if unexpected unbounded inputs are given.
More serious is the issue of operations that would yield different
results for the same inputs, depending on the availability or not of
unbounded intervals. Adding "flavours" along the lines John Pryce
and I have suggested offers another way to deal with this difficulty.
In any case, Nate's motion does not help here.
Michel.
---Sent: 2012-05-29 21:21:08 UTC