Re: Unexpected casts while using date_trunc()

Chris Bandy <bandy(dot)chris(at)gmail(dot)com> writes:
> On 5/24/18 2:31 PM, Tom Lane wrote:
>> Andrew Gierth <andrew(at)tao11(dot)riddles(dot)org(dot)uk> writes:
>>> There's also the option of adding an explicit function
>>> date_trunc(text,date) returns date, which is a workaround that I (and
>>> probably quite a few other people) have used.

> Are we in agreement that the return type should be date?

That is a good question, actually. That would be a larger behavior change
than just avoiding the undesired conversion to TZ. I had imagined this
as just being equivalent to date_trunc(text, date::timestamp). Casting
the result back down to date seems safe, though.

Another thing to consider is that the effective range of date is wider
than timestamp's, meaning coerce-to-timestamp can fail. Is it worth
providing a whole additional code path so we never coerce the date to
timestamp at all? I'd tend to think not, but perhaps somebody wants to
argue differently.

regards, tom lane