I am using Postgres 8.4.1 on Windows XP Professional Service Pack 3.
I have a PL/pgSQL function which is defined as "returns record". The record
contains three values. In one execution path, the values are read from a table,
the selected columns being of types int, smallint and char(1). In another
execution path, the second and third values are the literals 1 and 'R'. In the
original version of the function the assignment in the second case was as
follows:
v_rv = (v_id, 1, 'R') ;
where v_rv is a variable of type record, and v_id is of type int. The client
application calls the function as follows:
select col1, col2, col3 from func(?, ?, ?) as (col1 int, col2 smallint, col3
char(1))
As far as I remember, when I was using Postgres 8.1.4 that worked, but under
Postgres 8.4.1 it results in the errors "Returned type integer does not match
expected type smallint" and "Returned type unknown does not match expected type
character". I can avoid the error by altering the assignment thus:
v_rv = (v_id, 1::smallint, 'R'::char(1)) ;
but I am puzzled as to why these explicit casts should be necessary. Is this
covered anywhere in the documentation?
