Stephan Szabo <sszabo@megazone.bigpanda.com> writes:
> Hmm, I just got my machine to give a similar failure mode with
> a slightly wacky input.
Perhaps more to the point:
regression=# select timestamptz '1901/12/13 0:0:0';
timestamptz
---------------------
1901-12-13 00:00:00
(1 row)
regression=# select timestamptz '1901/12/14 0:0:0';
timestamptz
------------------------
1901-12-14 00:00:00-05
(1 row)
Note the lack of timezone in the first output.
It looks like 1901/12/14 is the oldest date for which the system will
return timezone information; IIRC, this is the oldest date representable
as a 32-bit time_t. PG implicitly assumes that timestamps before that
are always GMT.
This still doesn't explain why Arnold sees a failure with to_date and
we don't, though.
regards, tom lane