I haven't thought about casting null before, it's interesting. It'll make the implementation a little easier, but won't answer my question completely. I still need to create a type in one query and use it in another, don't I? It just seems strange to me that I can't define a function/trigger/view/anything to show a matrix in one easy request.
A self-contained example probably would have been more effective here...I can barely follow along but understand the two core issues:
1. A query must define all of its columns at execution time.
2. An anyelement output function must have at least one anyelement input.
Focusing on that alone one convention is to define a type and then pass in a "null" of that type as the anyelement input. You then expect to get a non-null value of that type upon function completion.
<not tested>
CREATE TYPE pair (x int, y int);
CREATE FUNCTION random_item(anyelement) RETURNS anyelement;
This is one more tool for the belt. I don't understand the details of your requirements to suggest the best way to combine pseudo type functions and dynamic SQL.
At some point you have to detect the column count from an initial query and dynamically incorporate it into a second query. Anything beyond that is because you are trying to make things better (faster, more understandable).
David J.
Please keep replies on-list and do not top-post.
As I ended with - the columns of a query must be known at compile time. Since your column count is determined by a query you cannot solve your problem with a single query.
This is fundamental limitation of SQL. You could make something that LOOKS like a matrix but only contains a single column (either textual or array) depending on your ultimate presentation goal. Look at the "format(...)" function if you want to play with this approach.
David J.
I'm sorry for this, I wasn't able to find any guide on using mailing lists, so I just pressed "reply" in sincere hope that it would work.
Thank you for the reply. Something that looks like a matrix is completely enough for me.