On Jan 6, 2012, at 07:08, John Gilmore wrote:
> The scheme for including leap seconds that I posted yesterday has been
> widely---by at least two readers---misunderstood. It or a functional
> equivalent must/should be used when the underlying time measurement is
> a TOD-clock [STCKE] value, an atomic-clock reading, or the like. It
> should NOT be used when this measurement is a TIME-macro based one.
> Such values already include leap-second corrections. To make them a
> second time would be otiose, and the result so obtained would be
> wrong.
>
Where is the inclusion of leap-second corrections documented?
I suppose it might be considered an intuitive consequence of
the specification of TZ.
Is the perhaps surprising combination:
TIME STCK,TZ=LT
permissible and meaningful? Of course it only complicates
calculation of intervals when a Daylight Saving or leap-second
correction may be included, absent the putative availability
of correction history.
> Let me also exploit this post to note that glb-seeking and lub-seeking
> binary search are much more important than the traditional
> match-seeking binary-search schemes that appear---almost always badly
> implemented because phobic about ternary comparisons---in programming
> textbooks.
>
The only language I've used that implemented ternary comparisons
was Mainsail. But (too) many languages with weak type checking
tolerate such as "A <= B < C" with the perhaps startling to the
novice presumption of left-association: "( A <= B ) < C". Oops!
Were you thinking instead of comparisons with binary domain and
ternary range?
gil
