On Wednesday, June 15, 2022, Bryn Llewellyn <
bryn@yugabyte.com> wrote:
I’ve copied a self-contained testcase below. Is the error that the "as intended" test causes due to a known limitation—or even a semantic dilemma that I'm failing to spot? Or might it be due to a bug?
I read the note in create domain as basically “don’t do this” (the not null part) but the issue you are pointing out seems unrelated to that.
/*
This one cases the error, thus:
ERROR: failed to find conversion function from key_vals_nn to record[]
CONTEXT: SQL expression "(kv1_nn = any(kvs_nn))"
*/;
select f('as intended');
The fact that a domain over an array isn’t being seen as an array here seems like a bug. POLA violation at least, and I don’t recall any notes regarding this dynamic in the docs.
However, a more trivial case does work, at least in HEAD:
postgres=# create domain mytext as text[] not null;
CREATE DOMAIN
postgres=# select '1' = any(array['1','2']::mytext);
?column?
----------
t
(1 row)
postgres=# create type kv AS ( key text, val text );
CREATE TYPE
postgres=# create domain kvarr as kv[];
CREATE DOMAIN
postgres=# select ('1','one')::kv = any (array[('1','one')::kv]);
?column?
----------
t
(1 row)
postgres=# select ('1','one')::kv = any ((array[('1','one')::kv])::kvarr);
ERROR: failed to find conversion function from kvarr to record[]
So the interaction of a composite type and the domain over array seems to be the scope of the issue - which makes me thing bug even more.
David J.