Dave Cramer <pg@fastcrypt.com> writes:
> Here are the comments in the code.
> // We must use UNSPECIFIED here, or inserting a Date-with-timezone into a
> // timestamptz field does an unexpected rotation by the server's TimeZone:
Ugh. But why not send it as a timestamptz?
(I suspect that the law of conservation of astonishment applies, such that
some other corner cases would act oddly if you do that. But it might be
a net improvement anyway.)
> Apparently while dates don't store timezone information we do accept
> timezone info in date literals ?
Sure. Times, too. The same input parser is used for date, timestamp,
timestamptz, time, and timetz, and then we just throw away irrelevant
fields.
regression=# select '2015-08-11 10:55:04.509393-04'::date;
date
------------
2015-08-11
(1 row)
regards, tom lane