Re: Range Types - typo + NULL string constructor

On Sun, 2011-09-18 at 18:08 +0200, Erik Rijkers wrote:
> Below are 2 changes. The first change is an elog saying 'lower' instead of 'upper'.

Done, thank you. New patch attached.

Changes:
* documentation fixes
* added document for pg_range catalog
* cleaned up errors, increased error checking
* improved pg_dump

TODO:
* Support casts and typmod.
- This requires adding a RangeCoerceExpr, or possibly
overloading ArrayCoerceExpr somehow. This is likely to
require a lot of boilerplate code and a fairly large diff.
* Cache lookups better to avoid unnecessary SearchSysCache calls.
* I need to find a clean way to get the operator class name in pg_dump.

> Rangetypes as it stands uses NULL to indicate INF or -INF:
>
> select int4range(2, NULL);
>
> int4range
> ------------
> [ 2, INF )
> (1 row)
>
>
> but refuses to accept it in the string-form:
>
> select '[ 2 , NULL )'::int4range;
> ERROR: NULL range boundaries are not supported
> LINE 1: select '[ 2 , NULL )'::int4range;

I think this might require more opinions. There is a trade-off here
between convenience and confusion: accepting NULL is convenient in the
constructors, because it avoids the need to have extra constructors just
for unbounded ranges; but could lead to confusion between NULL and INF
(which are not the same).

In the string form, it doesn't add any convenience to accept NULL; but
as you point out, it seems inconsistent without it.

Thoughts?

Regards,
Jeff Davis