On Tue, Sep 19, 2017 at 2:12 PM, Tom Lane <tgl@sss.pgh.pa.us> wrote:
"David G. Johnston" <david.g.johnston@gmail.com> writes: > Actually, this does work, just not the way one would immediately expect.
On closer inspection, what's actually happening in your example is that you're getting the SELECT's ct1 result crammed into the first column of c1, and then a default NULL is stuck into its second column: > ct1: (text, text)
> DO $$ > SELECT ('1', '2')::ct1 INTO c1; > RAISE NOTICE '%', c1; > END; > $$;
> Notice: ("(1,2)",)
Notice the parens/comma positioning. It's only because text is such a lax datatype that you didn't notice the problem.
I saw exactly what you described - that it didn't error out and that the text representation of the single output composite was being stored in the first field of the target composite. i.e., that it "worked". Does that fact that it only works if the first field of the composite type is text change your opinion that the behavior is OK to break?
