[LEAPSECS] Java: ThreeTen/JSR-310

[Michael Deckers]

   On 2011-01-28 18:34, Gerard Ashton asked:


>  Windows and Unix have general reputations of not doing it right; does

>  anyone know of a hardware/operating system combination that handles leap

>  seconds correctly? If so, does it have a defined approach to providing a

>  quasi-UTC that hides leap seconds to applications that wish to live in

>  blissful ignorance?


    No, I do not know where it is done "correctly" -- I just know some
    programming languages where datetimes and time scales are modelled
    in other ways than in the JSR-310 proposal. Let me just mention two.

    Take ADA95, for example. The language accepts the existence of
    various independent time scales, that is, physical quantities that
    all take their values in the same space of datetimes. That space
    provides the operations of an affine space over the one-dimensional
    vector space of times, with units s, min, h, d, etc.
    Datetimes values can be denoted using calendars or with affine
    units such as Julian dates or Besselina epochs.

    Whether values of one time scale can be converted (in whatever
    sense) to values of another time scale is not imposed by the
    interface. Some time scales can be converted exactly (eg, TCG to
    TT), some exactly only for past epochs (eg, UTC and TAI), some
    only approximately, even for the past (eg, TT to TAI).

    ADA95 also recognizes the fact that it is sometimes needed
    to add additional information to a datetime value: viz, the
    information that the datetime value occurs twice in a non-monotone
    time scale. This is needed to order such amended datetime values
    in the case of leap seconds, and in the case of civil time scales
    while they switch from summer time to winter time. Other additional
    operations on these values comprise better conversions to other
    time scales, eg, unambiguous conversion from UTC to TAI, or from
    civil time to UTC with the help of a zic file.

    SQL provides datetime values with an additional offset so that
    they comprise both a local time scale and a UTC value. This is
    another form of amended datetime values that is important in
    applications. It can also model a UTC time stamp with known
    offset to TAI. This amendment is independent of the previous
    one, so that I would expect a complete support to provide datetime
    types with both amendments, too.

    I have left out all the gory details and I hope this comment is
    still not too abstract.

    Michael Deckers.