On Mon, Jun 17, 2024 at 9:05 PM Chapman Flack <jcflack@acm.org> wrote:
>
> Hi,
>
> On 06/17/24 02:43, Amit Langote wrote:
> > <replaceable>context_item</replaceable> expression can be a value of
> > any type that can be cast to <type>jsonb</type>. This includes types
> > such as <type>char</type>, <type>text</type>, <type>bpchar</type>,
> > <type>character varying</type>, and <type>bytea</type> (with
> > <code>ENCODING UTF8</code>), as well as any domains over these types.
>
> Reading this message in conjunction with [0] makes me think that we are
> really talking about a function that takes a first parameter of type jsonb,
> and behaves exactly that way (so any cast required is applied by the system
> ahead of the call). Under those conditions, this seems like an unusual
> sentence to add in the docs, at least until we have also documented that
> tan's argument can be of any type that can be cast to double precision.
>
I guess it would be fine to add an unusual sentence to the docs.
imagine a function: array_avg(anyarray) returns anyelement.
array_avg calculate an array's elements's avg. like
array('{1,2,3}'::int[]) returns 2.
but array_avg won't make sense if the input argument is a date array.
so mentioning in the doc: array_avg can accept anyarray, but anyarray
cannot date array.
seems ok.
> On the other hand, if the behavior of the functions were to be changed
> (perhaps using prosupport rewriting as suggested in [1]?) so that it was
> not purely describable as a function accepting exactly jsonb with a
> possible system-applied cast in front, then in that case such an added
> explanation in the docs might be very fitting.
>
prosupport won't work, I think.
because json_exists, json_value, json_query, json_table don't have
pg_proc entries.
These are more like expressions.