On Fri, May 19, 2023 at 2:58 AM David G. Johnston < david.g.johns...@gmail.com> wrote:
> On Thursday, May 18, 2023, Raymond Brinzer <ray.brin...@gmail.com> wrote: > >> scratch=# select row(2,3)::test_type; >> > > Unknown typed value, immediately converted to a known concrete instance of > test_type. It is never actually resolved as record. > > All of the others must concretely be resolved to record to escape their > query level, and if you then try to cast the concrete record to some other > concrete type a cast needs to exist. > I see. That suggests, oddly, that pg_typeof() is changing the thing it's observing, because row(2,3) was not a record but an 'unknown typed value' before pg_typeof() was called on it. Good to know. CREATE CAST seems not to like working with pseudo-types. Neither the source nor the target can be 'anyarray' or 'record'. So that seems out. On the other hand, the pseudo-type doc says, 'Functions coded in C (whether built-in or dynamically loaded) can be declared to accept or return any of these pseudo data types.' I'm assuming that an 'unknown typed value" corresponds to the 'unknown' pseudo-type. So it seems like a C function which took an anyarray as its parameter, and returned a value of type 'unknown' would bypass the need for a specifically defined cast. (Emphasis on "seems".) -- Ray Brinzer
