Re: Do we want a hashset type?

On Mon, Jun 26, 2023 at 4:55 PM Joel Jacobson <joel(at)compiler(dot)org> wrote:

> On Mon, Jun 26, 2023, at 13:06, jian he wrote:
> > Can you try to glue the attached to the hashset data type input
> > function.
> > the attached will parse cstring with double quote and not. so '{1,2,3}'
> > == '{"1","2","3"}'. obviously quote will preserve the inner string as
> > is.
> > currently int4hashset input is delimited by comma, if you want deal
> > with range then you need escape the comma.
>
> Not sure what you're trying to do here; what's the problem with
> the current int4hashset_in()?
>
> I think it might be best to focus on null semantics / three-valued logic
> before moving on and trying to implement support for more types,
> otherwise we would need to rewrite more code if we find general
> thinkos that are problems in all types.
>
> Help wanted to reason about what the following queries should return:
>
> SELECT hashset_union(NULL::int4hashset, '{}'::int4hashset);
>
> SELECT hashset_intersection(NULL::int4hashset, '{}'::int4hashset);
>
> SELECT hashset_difference(NULL::int4hashset, '{}'::int4hashset);
>
> SELECT hashset_symmetric_difference(NULL::int4hashset, '{}'::int4hashset);
>
> Should they return NULL, the empty set or something else?
>
> I've renamed hashset_merge() -> hashset_union() to better match
> SQL's MULTISET feature which has a MULTISET UNION.
>

Shouldn't they return the same thing that left(NULL::text,1) returns?
(NULL)...
Typically any operation on NULL is NULL.

Kirk...