Steve Rogerson <steve.pg@yewtc.demon.co.uk> writes:
> # select (now());
> now
> -------------------------------
> 2017-04-03 11:57:09.891043+01
> (1 row)
> sjr_local1db=# select (now() AT TIME ZONE 'UTC');
> timezone
> ----------------------------
> 2017-04-03 10:57:11.714571
> (1 row)
> sjr_local1db=# select (now() AT TIME ZONE 'UTC') AT TIME ZONE 'UTC';
> timezone
> -------------------------------
> 2017-04-03 11:57:14.088515+01
> (1 row)
> This makes no sense to me.
Looks perfectly fine from here. You're rotating a timestamp with time
zone (displayed in your local zone, evidently GMT+1) to a timestamp
without time zone expressed in UTC, and then back to a timestamp with time
zone. That round trip should be a no-op, barring weird corner cases.
I'd be the first to agree that the notation is pretty opaque --- why
use the same "operator" for both transformation directions? --- but
don't blame us, blame the SQL spec.
regards, tom lane