America/Sao_Paulo DST problem

[Olson, Arthur David (NIH/NCI)]

Here are a few lines of "zdump" output for America/Sao_Paulo:

	America/Sao_Paulo  Sun Oct 19 03:00:00 2003 UTC = Sun Oct 19
01:00:00 2003 BRST isdst=1
	America/Sao_Paulo  Sun Feb 15 01:59:59 2004 UTC = Sat Feb 14
23:59:59 2004 BRST isdst=1
	America/Sao_Paulo  Sun Feb 15 02:00:00 2004 UTC = Sat Feb 14
23:00:00 2004 BRT isdst=0
	America/Sao_Paulo  Sun Oct 17 02:59:59 2004 UTC = Sat Oct 16
23:59:59 2004 BRT isdst=0

The important thing to note is at midnight on February 14 clocks are set
back an hour.
This means that there are two instants when a well-maintained Sao Paulo wall
clock would read 11:00:00 PM on February 14
(as well as two instants when it would 11:00:01, 11:00:02, and so on through
11:59:59).

The "my_date.c" program feeds year-mon-mday-hour-min-sec into mktime; the
values being fed in lie in the "fall back" hour above.
Mktime is supposed to return an instant when a wall clock shows the time
that's been input to it; here, since there are two such instants (and since
"tm_isdst" has been set to -1), mktime is allowed to pick either instant to
return. It's luck of the draw that determines which instant mktime returns.

				--ado