Обсуждение: Re: [BUGS] BUG #1993: Adding/subtracting negative time intervals

[ bugs list removed, hackers added.]

Tom Lane wrote:
> Bruce Momjian <pgman@candle.pha.pa.us> writes:
> > I saw a lot of disussion because I forgot to specify that my tests were
> > for EST5EDT, but what about the use of interval_justify_hours() in
> > timestamp_mi().  Is this something we want to change?
> 
> It's too late to mess with it for 8.1, but see my previous message
> proposing a set of TODO items for future work.

Yes, it is late, but I am worried about adding an interface change that
we will later revert in 8.2.  In 8.0.X I see the query returning the '25
hour' answer:
SELECT('2005-10-29 13:22:00-04'::timestamptz +('2005-10-30 13:22:00-05'::timestamptz - '2005-10-29
13:22:00-04'::timestamptz))at time zone 'EST';      timezone--------------------- 2005-10-30 13:22:00(1 row)
 

In current CVS the top query returns '14:22:00'.  Do we change this for
8.1, then change it back in 8.2?  That seems bad to me.

Actually, 8.0.X returns '1 day, 1 hour' for the subtraction, which we
treat in 8.0.X as '25 hours':SELECT        ('2005-10-30 13:22:00-05'::timestamptz -         '2005-10-29
13:22:00-04'::timestamptz);   ?column?---------------- 1 day 01:00:00(1 row)
 

In 8.0.X, because we didn't have a 'days' field, we could treat '1 day 1
hour' as always '25 hours', and could display the results as days/hours.
If we remove interval_justify_hours(), then we are always going to
display timestamp subtraction in hours (not days), e.g. '6422 hours'
(yea, ugly) unless they manually call interval_justify_hours().

Keep in mind that the addition of the interval_justify_hours() did
generate some regression test changes, so removing
interval_justify_hours() might just take the results back to what we had
in 8.0.  My point is that regression changes caused by its removal might
not be a good guide to determining compatibility with 8.0.X.

I guess my point is that we are changing 8.0.X behavior so we better be
sure it is now the way we want it to remain.

--  Bruce Momjian                        |  http://candle.pha.pa.us pgman@candle.pha.pa.us               |  (610)
359-1001+  If your life is a hard drive,     |  13 Roberts Road +  Christ can be your backup.        |  Newtown Square,
Pennsylvania19073
 

Bruce Momjian <pgman@candle.pha.pa.us> writes:
> Keep in mind that the addition of the interval_justify_hours() did
> generate some regression test changes, so removing
> interval_justify_hours() might just take the results back to what we had
> in 8.0.

Not hardly.  I tried already.  The existing timestamp_mi behavior is
probably as close to 8.0 as we can get given the change in underlying
representation.

> I guess my point is that we are changing 8.0.X behavior so we better be
> sure it is now the way we want it to remain.

[ shrug... ]  We've changed datetime behavior in every past release,
we're changing it for 8.1, we'll probably change it some more for 8.2,
and again after that.  All the datetime code is a work in progress.
Get used to it.
        regards, tom lane

Tom Lane wrote:
> Bruce Momjian <pgman@candle.pha.pa.us> writes:
> > Keep in mind that the addition of the interval_justify_hours() did
> > generate some regression test changes, so removing
> > interval_justify_hours() might just take the results back to what we had
> > in 8.0.
> 
> Not hardly.  I tried already.  The existing timestamp_mi behavior is
> probably as close to 8.0 as we can get given the change in underlying
> representation.

You mean the '6432 hours' is a worse change, OK.

> > I guess my point is that we are changing 8.0.X behavior so we better be
> > sure it is now the way we want it to remain.
> 
> [ shrug... ]  We've changed datetime behavior in every past release,
> we're changing it for 8.1, we'll probably change it some more for 8.2,
> and again after that.  All the datetime code is a work in progress.
> Get used to it.

OK, as long as we are sure we are not going to change it back to 8.0
behavior.

--  Bruce Momjian                        |  http://candle.pha.pa.us pgman@candle.pha.pa.us               |  (610)
359-1001+  If your life is a hard drive,     |  13 Roberts Road +  Christ can be your backup.        |  Newtown Square,
Pennsylvania19073
 

Bruce Momjian <pgman@candle.pha.pa.us> writes:
> Tom Lane wrote:
>> Not hardly.  I tried already.  The existing timestamp_mi behavior is
>> probably as close to 8.0 as we can get given the change in underlying
>> representation.

> You mean the '6432 hours' is a worse change, OK.

Well, it's sure not a small change, and we're still undecided whether
that's what we want in the long run.

Also, we'd have to deal with some of the other TODO items I mentioned
before we could make it work at all.  There's at least one regression
test that computes an interval larger than 2^31 hours (how do you think
I found out about that problem ;-))
        regards, tom lane