TimeTest zone offset weirdness

I'm currently looking at TimeTest.testGetTimeZone() which has:

 final Time midnight = new Time(0, 0, 0);
 Calendar cal = Calendar.getInstance();
 cal.setTimeZone(TimeZone.getTimeZone("GMT"));
 long localOffset = Calendar.getInstance().get(Calendar.ZONE_OFFSET);
 [...]
 time = rs.getTime(1, cal);
 assertEquals(midnight.getTime() , time.getTime() + localOffset);

The value in the database that's being retrieved is '00:00:00' -- it is
a TIME WITHOUT TIME ZONE.

As I interpret things:
- 'midnight' should represent 1970-01-01 00:00:00 in the local timezone,
i.e. local midnight.
- 'time' should represent 1970-01-01 00:00:00 GMT, i.e. the database
value interpreted in the provided (GMT) calendar.

However as localOffset is the number of milliseconds that the local
timezone is *ahead* of GMT, shouldn't that test be for 'time.getTime() -
localOffset'?

e.g. in the NZ timezone, localOffset is 43200000 (milliseconds) as NZ is
12 hours ahead of GMT in January 1970. Then midnight.getTime() is
-43200000 as NZ reaches local midnight before GMT.

Given that this doesn't fail in the current driver, presumably this
means that getTime() is currently returning a Time with milliseconds =
-86400000 ... which is surely wrong.

With the timestamp-related changes I'm doing I also overhauled the Time
and Date parsers; now getTime() is returning a Time with milliseconds =
0 and the assertion fails.

Is this actually an existing test/driver error or am I misinterpreting
what the behaviour should be?

-O