Re: timestamps and dates

Antti Haapala <antti(dot)haapala(at)iki(dot)fi> writes:
>> So zones in 'right' folder have leap second support on. The difference is
>> correct - 22 (i had it wrong before), the number of leap seconds inserted
>> since UTC Epoch on 1 Jan 1972.

Yeah. That's the second report we've had of systems running in a
leap-second zone by default. I think it would be a good idea for
Postgres to check for this situation and complain. But how strongly
should it complain? Refuse to start up? Adopt GMT instead? What if
asking for GMT gets a leap-second zone?

> ilmo=# select '1998-31-12 23:59:60 UTC'::timestamp with time zone;
> ERROR: Bad timestamp external representation '1998-31-12 23:59:60 UTC'

> My timestamp surely is legal according to ISO-8601.

That's a good point. We got complaints about this all the time back
when we had roundoff problems in that code, but no one ever stopped to
point out that such a timestamp actually is legal per spec. (Strictly
speaking I think :60 should only be accepted at points where there
actually was a leap second, but we're not gonna check for that...)

regards, tom lane