BUG #14854: daterange[] is an anyarray or anyrange?

On Fri, Oct 13, 2017 at 12:45 PM, <balazs@obiserver.hu> wrote:

The following bug has been logged on the website:

Bug reference: 14854
Logged by: Balazs Szilfai
Email address: balazs@obiserver.hu
PostgreSQL version: 9.6.5
Operating system: Debian Linux
Description:

I can't create function with param to accept a pseudo-type to daterange and
daterange[] (array of daterange).

I tried:

CREATE FUNCTION range_overlap_array_any(anyrange, anyarray) RETURNS
boolean
AS $$SELECT false;$$ LANGUAGE sql IMMUTABLE STRICT;

​​
CREATE FUNCTION range_overlap_array_any(anyrange, anyrange) RETURNS
boolean
AS $$SELECT false;$$ LANGUAGE sql IMMUTABLE STRICT;

My queries and the error messages:

SELECT range_overlap_array_any(daterange('2017-01-01', '2017-04-01'),
array[daterange('2016-12-10', '2016-12-11')]);

ERROR: function range_overlap_array_any(daterange, daterange[]) does not
exist

SELECT range_overlap_array_any(daterange('2017-01-01', '2017-04-01'),
array['x'::text]);

ERROR: function range_overlap_array_any(daterange, text[]) does not exist

What's the mistake? Or did I break something?

​When a pseudo-type is used in a function parameter specification the
system enforces the constraint that the same "base" type is used for all
arguments during function invocation.

https://www.postgresql.org/docs/10/static/extend-type-system.html#extend-types-polymorphic

"Each position (either argument or return value) declared as anyelement is
allowed to have any specific actual data type, but in any given call they
must all be the same actual type. [...]" - the rest of that paragraph
basically explains with many words that which I summarize above but without
the concept of "base type" to ease comprehension.

In this case "date" is your base type so the valid combination of arguments
is
(daterange, date[])​

Your first invocation (daterange, daterange[]) works if you define your
function as "(anyelement, anyarray)"; thus making "daterange" your base
type when invoked that way.

The invocation (daterange, text[]) is not a valid combination for any pure
pseudo-argument function.

David J.

Thanks a lot!

Ok, the (daterange, date[]) invocation is working now with (anyelement,
anyarray) param list.
But now is working with eg. (date, date[]) invocation. I can't forbid it
anyway?

Balazs

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

On Fri, Oct 13, 2017 at 1:42 PM, Szilfai Balázs <balazs@obiserver.hu> wrote:

Thanks a lot!

Ok, the (daterange, date[]) invocation is working now with (anyelement,
anyarray) param list.

I ​assume you meant (daterate, daterange[])​

But now is working with eg. (date, date[]) invocation. I can't forbid it

anyway?

​No. It also matches (text, text[]) and "(integer, integer[])". The best
you can do is detect non-range values in the first argument and raise an
exception. Or define explicitly typed functions for the variants you care
about and drop the polymorphism.

David J.

"David G. Johnston" <david.g.johnston@gmail.com> writes:

On Fri, Oct 13, 2017 at 12:45 PM, <balazs@obiserver.hu> wrote:

CREATE FUNCTION range_overlap_array_any(anyrange, anyarray) RETURNS
boolean
AS $$SELECT false;$$ LANGUAGE sql IMMUTABLE STRICT;

SELECT range_overlap_array_any(daterange('2017-01-01', '2017-04-01'),
array['x'::text]);

ERROR: function range_overlap_array_any(daterange, text[]) does not exist

The invocation (daterange, text[]) is not a valid combination for any pure
pseudo-argument function.

Yeah. The expectation with these pseudotypes is that you're trying to
declare a function that takes some set of arguments of related types.

Back when we first invented the idea of polymorphic pseudotypes,
which was a good long time ago now, there was discussion of having
a second set of pseudotypes that are tied to a second underlying
"base type" type variable, so that while

myfunc(anyrange, anyarray)

means "a range over some type, and an array over that same type"
then you could write, say

myfunc(anyrange, anyelement2, anyarray2)

to mean "a range over some type A, and a value of some possibly-different
type B, and an array over type B". We didn't do it because nobody had a
particularly compelling use-case at the time, and also because if we
needed 2 sets of pseudotypes then maybe we needed 3, etc; it wasn't clear
where to stop.

Nearly fifteen years later, we still haven't seen a compelling example
for inventing a second set of pseudotypes. If you've got one it would
definitely be interesting.

regards, tom lane

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