Re: timestamp arithmetic (a possible bug?)

Stephan Szabo <sszabo(at)megazone(dot)bigpanda(dot)com> writes:
>>> Should we be providing an interval + timestamp operator as well since it
>>> looks like the spec implies both orderings should work?
>>
>> If you see spec support for it, then yes ... where do you read that
>> exactly?

> SQL92 (draft) 4.5.3 Operators involving datetimes and intervals (the table
> appears to be the same in SQL99 4.7.3)

Yeah. It looks like we have most of these, but would need to add
interval + date
interval + timetz
interval + timestamp
interval + timestamptz
and for consistency
integer + date
Curiously, we do have interval + time without time zone ... I guess Tom
Lockhart overlooked these when he was working in the area.

I notice also that date - date yields an integer (ie, number of days)
where I think that strict spec compliance would mandate yielding an
interval instead. I'm uneager to change this though.

regards, tom lane