[tz] [PROPOSED 3/3] Vanguard form now uses subsecond precision

[Michael H Deckers]

    On 2022-07-28 05:02, Paul Eggert via tz proposed changes in files:


>   africa       |  3 ++-
>   asia         | 19 ++++++++++++-------
>   europe       | 13 +++++++------
>   northamerica | 18 +++++++++++-------
>   southamerica | 15 ++++++++++++++-


    In tzdb, the difference local mean time LMT - UT only plays a role 
for the
    switch from LMT to a grid time zone time (and sometimes back to LMT, as
    in Amsterdam). So the relevant quantity is LMT - UT as estimated at the
    time of this switch.

    In the best documented cases, the switch is precisely specified by law
    (as for Dublin); and one can assume that the legal authorities
    have specified the switch to (at least) a precision that they thought
    to be relevant for the local time observatories and the most demanding
    applications at the time (such as navigation).

    In less well documented cases, we have an estimate of LMT - UT by local
    authorities (eg, from Milne 1899) at an epoch close to that of the 
switch;
    if these local authorities used subsecond precision then we may assume
    that such precision was considered relevant.

    But where we do not have such estimates, subsecond precision seems to be
    unwarranted as it would imply the wrong uncertainty for the size of 
the jump.
    As a case in point, the estimate used in tzdb for LMT - UT at 
Jakarta comes
    from 1876, but the switch to the grid time zone time UT + 07:20 only 
happened
    in 1932. Clearly, the old estimate of LMT - UT cannot be taken as 
evidence
    at subsecond precision for the jump made 56 years later.

    Regardless of whether subsecond precision is justified in each case,
    the following points in the proposed changes also need attention:

    • Unsurprisingly, no change is proposed for backzone -- but this is
      bound to produce new inconsistencies:
      backzone/America/Cayman uses Kingston Mean Time KMT and should
      therefore be made consistent with America/Jamaica.

    • backzone/America/Rosario has not been maintained since 2002-04-04 (eg,
      it has wrong offset for Córdoba Mean Time CMT, has no DST 2007..2009).
      It should be dropped from tzdb since it is unclear how it is supposed
      to differ from America/Argentina/Buenos_Aires; otherwise it must be
      properly maintained, including the new offset for CMT.

    • Similarly, backzone/America/Argentina/ComodRivadavia does not agree
      with America/Argentina/Catamarca since 1970 because the former has
      not been maintained since 2005-08-22, while Catamarca has since been
      changed. (Having the sources in the same file certainly would have
      helped!) So ComodRivadavia does not even satisfy the purported
      criterion for members of backzone. Remove or maintain properly.

    • Zone Asia/Jakarta needs an UNTIL AT column
         #STDOFF   7:07:12.5                   23:47:12.5

    Michael Deckers.