To: Bryce Nesbitt <bryce1@obviously.com>
cc: pgsql-sql@postgresql.org
Subject: Re: extract or date_part on an interval? How many e 
In-reply-to: <479D3E63.2080601@obviously.com> 
References: <479D3E63.2080601@obviously.com>
Comments: In-reply-to Bryce Nesbitt <bryce1@obviously.com>
	message dated "Sun, 27 Jan 2008 18:30:59 -0800"
Date: Sun, 27 Jan 2008 22:10:40 -0500
Message-ID: <16843.1201489840@sss.pgh.pa.us>
From: Tom Lane <tgl@sss.pgh.pa.us>

Bryce Nesbitt <bryce1@obviously.com> writes:
> Hmm.  Seemed so simple.  But how do I get the number of years an
> interval represents?  extract is clearly the wrong way:

There is nothing simple about datetime calculations, ever :-(

Let me exhibit why this particular case is not as simple as you
could wish:

regression=# select '2007-02-01'::timestamp + interval '1 year';
      ?column?       
---------------------
 2008-02-01 00:00:00
(1 row)

regression=# select '2007-02-01'::timestamp + interval '365 days';
      ?column?       
---------------------
 2008-02-01 00:00:00
(1 row)

regression=# select '2008-02-01'::timestamp + interval '1 year';
      ?column?       
---------------------
 2009-02-01 00:00:00
(1 row)

regression=# select '2008-02-01'::timestamp + interval '365 days';
      ?column?       
---------------------
 2009-01-31 00:00:00
(1 row)

That is, there isn't any fixed conversion factor between N days
and N years, so the interval datatype treats them as incommensurate.

If you're willing to settle for an approximate answer, you can
do extract(epoch from interval) and then divide by however many
seconds you want to believe are in a year.  This will give various
wrong answers in various corner cases, but I'm not sure there is
a right answer.

			regards, tom lane