problem withzic 7.81 (1997a)

[Dave Sparks]

I recently downloaded tz{code,data}1997a.tar.gz in order to update our
timezone database.  Because we will not be using any dates earlier than
1970 I used zic -s to save space.

All the timezone information was compiled, but when I tried to install
the data into our database four timezones failed the sanity checks.  I
found that the compiled files for these timezones were each 44 bytes long
and contained only zeros.  The zones were:

    Australia/Darwin
    Africa/Accra
    Africa/Freetown
    Africa/Johannesburg

(none of which have transitions after 1970).

I looked at zic.c and made the following changes:

  207% diff zic.c zic.OLD.c
    3c3
    < static char   elsieid[] = "@(#)zic.c  7.81  +DaveS";
    ---
    > static char   elsieid[] = "@(#)zic.c  7.81";
    1572a1573
    >               usestart = i > 0 && (zp - 1)->z_untiltime > min_time;
    1576d1576
    <               usestart = (timecnt == 0 && !useuntil) || (i > 0 && ((zp - 1)->z_untiltime > min_time));

which was partially successful (the Australian timezone was handled, but
the African timezones needed modifications which wouldn't have been
necessary if the fix had been correct:

  208% diff africa africa.OLD
    325,326c325
    <                        0:00   Ghana   %s      1942
    <                        0:00   -       WAT
    ---
    >                        0:00   Ghana   %s
    561,562c560
    <                        0:00   SL      WA%sT   1962
    <                        0:00   -       WAT
    ---
    >                        0:00   SL      WA%sT
    581,582c579
    <                       2:00    SA      SA%sT   1944
    <                       2:00    -       SAT
    ---
    >                       2:00    SA      SA%sT

).  After looking more carefully at the timezone data, I realised that
what the troublesome timezones had in common wasn't just that there were
no transitions after 1970, but also that the last transition was by Rule.
Since by then I had compiled data which we could use, I didn't look at
zic.c again to try to work out a better fix.

Incidentally, the assumption that the first Standard Time offset should
be used for times before the first transition doesn't work well for
Southern Hemisphere timezones: in many cases Summer Time will be in
effect at the earliest time handled.

(In case you were wondering why we need to import the timezone data into
our own database: the application in question has to handle multiple
timezones and is nor written in C or C++.)

-- 
Dave Sparks, Staffordshire, England

(posted from my private mailbox to avoid my employer being legally
liable, since in English law a disclaimer would be ineffective.)