Why does the range type's upper function behave inconsistently?

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hello,

On 05/07/2015 19:13, Dane Foster wrote:

I don't understand the inconsistent behavior of the range types'
upper function in regard to inclusive ranges.

For example(s): 1. SELECT upper(int4range(1, 4, '[]')) = 4; --
FALSE 2. SELECT upper(int8range(1, 4, '[]')) = 4; -- FALSE 3.
SELECT upper(numrange(1, 4, '[]')) = 4; -- TRUE 4. SELECT
upper(tstzrange('2015-07-01: 00:00:00', now(), '[]')) = now(); --
TRUE 5. SELECT upper(daterange('2015-01-01', current_date, '[]'))
= current_date; -- FALSE

#1 & #2 are FALSE because upper returns 5 instead of 4; and #5 is
FALSE because upper returns: current_date + interval '1 day'. I
don't understand the logic behind why it would return the inclusive
upper bound value for some ranges and not others. If anyone can
shed some light on this behavior it would be greatly appreciated.

One of things I originally tried to use upper for was CHECK
constraints. That was until I wrote some unit tests and realized
that upper doesn't consistently work the way I expected. Of course
my assumptions are probably wrong so that's why I'm asking for
clarification.

Because for discrete range types, the canonical form is used, which is
[). Check
http://www.postgresql.org/docs/current/static/rangetypes.html and the
discrete range types paragraph.

Regards.

Regards,

Dane

- --
Julien Rouhaud
http://dalibo.com - http://dalibo.org
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2.0.17 (GNU/Linux)

iQEcBAEBAgAGBQJVmWhvAAoJELGaJ8vfEpOqRa0H/1+QaaZm3JrGOks2FN/24j3/
US4+Zc8AJWarOtd9Nxe4FGkUeVN1kEitJVOXLn1f6tyWRTJZ1A6v8ZaJzykqj3Bj
6cifqmq+c+NNXFyOS9vou7gzIiDxrIYmDTLBc7LqT8eWUmkQKGQT4no4Cre3uD4F
kAp/CvFBpyVLCGMsBP4fW7ShnyVlwk2r1KEDn8rgpVW5rPBV7KPrneoEPJ9EBHt0
jlnYpsxgnsu6OkbmTE3gA0a9Mx/pfJlN9r2TaVjH0oOVvgFDWYX6uLVJDtFJYQrf
zOEjVBaGJQ1CT+2M2GEWQj7X4Px/o6tXbEx9sZikp/xD//+rH5LAuKf3NhPGE1w=
=caV1
-----END PGP SIGNATURE-----

--
Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general

Dane Foster <studdugie@gmail.com> writes:

I don't understand the inconsistent behavior of the range types' upper
function in regard to inclusive ranges.

The behavior is different for discrete vs. continuous ranges. For
example,

regression=# select int4range(1, 4, '[]');
int4range
-----------
[1,5)
(1 row)

regression=# select numrange(1, 4, '[]');
numrange
----------
[1,4]
(1 row)

In the discrete case we normalize the bounds to '[)' style so that ranges
that contain the same sets of values will compare as equal even when they
were written differently. But there's no practical way to do that for
continuous types. See
http://www.postgresql.org/docs/9.4/static/rangetypes.html#RANGETYPES-DISCRETE

regards, tom lane

--
Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general

On 07/05/2015 10:13 AM, Dane Foster wrote:

I don't understand the inconsistent behavior of the range types' upper
function in regard to inclusive ranges.

For example(s):
1. SELECT upper(int4range(1, 4, '[]')) = 4; -- FALSE
2. SELECT upper(int8range(1, 4, '[]')) = 4; -- FALSE
3. SELECT upper(numrange(1, 4, '[]')) = 4; -- TRUE
4. SELECT upper(tstzrange('2015-07-01: 00:00:00', now(), '[]')) = now();
-- TRUE
5. SELECT upper(daterange('2015-01-01', current_date, '[]')) =
current_date; -- FALSE

To follow up on Julien Rouhaud post, if you do:

production=# select daterange('2015-01-01', current_date, '[]');
daterange
-------------------------
[2015-01-01,2015-07-06)
(1 row)

see that the '[]] has been changed to '[)' with tomorrows date as the
upper bound.

#1 & #2 are FALSE because upper returns 5 instead of 4; and #5 is FALSE
because upper returns: current_date + interval '1 day'. I don't
understand the logic behind why it would return the inclusive upper
bound value for some ranges and not others. If anyone can shed some
light on this behavior it would be greatly appreciated.

One of things I originally tried to use upper for was CHECK constraints.
That was until I wrote some unit tests and realized that upper doesn't
consistently work the way I expected. Of course my assumptions are
probably wrong so that's why I'm asking for clarification.

Regards,

Dane

--
Adrian Klaver
adrian.klaver@aklaver.com

--
Sent via pgsql-general mailing list (pgsql-general@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-general

Thanks everyone. I understand now. The funny thing is I read the
documentation many weeks before actually using range types for the first
time but it didn't click that the documentation was describing the behavior
I was observing, until now.

Thanks again,

Dane

On Sun, Jul 5, 2015 at 1:33 PM, Adrian Klaver <adrian.klaver@aklaver.com>
wrote: