Copy records with dynamic column names - arrays

Question

I have two tables with different columns in PostgreSQL 9.3:
CREATE TABLE person1(
NAME TEXT NOT NULL,
AGE INT NOT NULL
);
CREATE TABLE person2(
NAME TEXT NOT NULL,
AGE INT NOT NULL,
ADDRESS CHAR(50),
SALARY REAL
);
INSERT INTO person2 (Name, Age, ADDRESS, SALARY)
VALUES ('Piotr', 20, 'London', 80);
I would like to copy records from person2 to person1, but column names can change in program, so I would like to select joint column names in program. So I create an array containing the intersection of column names. Next I use a function: insert into .... select, but I get an error, when I pass the array variable to the function by name. Like this:
select column_name into name1 from information_schema.columns where table_name = 'person1';
select column_name into name2 from information_schema.columns where table_name = 'person2';
select * into cols from ( select * from name1 intersect select * from name2) as tmp;
-- Create array with name of columns
select array (select column_name::text from cols) into cols2;
CREATE OR REPLACE FUNCTION f_insert_these_columns(VARIADIC _cols text[])
RETURNS void AS
$func$
BEGIN
EXECUTE (
SELECT 'INSERT INTO person1 SELECT '
|| string_agg(quote_ident(col), ', ')
|| ' FROM person2'
FROM unnest(_cols) col
);
END
$func$ LANGUAGE plpgsql;
select * from cols2;
array
------------
{name,age}
(1 row)
SELECT f_insert_these_columns(VARIADIC cols2);
ERROR: column "cols2" does not exist
What's wrong here?

Answer 1

You seem to assume that SELECT INTO in SQL would assign a variable. But that is not so.
It creates a new table and its use is discouraged in Postgres. Use the superior CREATE TABLE AS instead. Not least, because the meaning of SELECT INTO inside plpgsql is different:
Combine two tables into a new one so that select rows from the other one are ignored
Concerning SQL variables:
User defined variables in PostgreSQL
Hence you cannot call the function like this:
SELECT f_insert_these_columns(VARIADIC cols2);
This would work:
SELECT f_insert_these_columns(VARIADIC (TABLE cols2 LIMIT 1));
Or cleaner:
SELECT f_insert_these_columns(VARIADIC array) -- "array" being the unfortunate column name
FROM cols2
LIMIT 1;
About the short TABLE syntax:
Is there a shortcut for SELECT * FROM?
Better solution
To copy all rows with columns sharing the same name between two tables:
CREATE OR REPLACE FUNCTION f_copy_rows_with_shared_cols(
IN _tbl1 regclass
, IN _tbl2 regclass
, OUT rows int
, OUT columns text)
LANGUAGE plpgsql AS
$func$
BEGIN
SELECT INTO columns -- proper use of SELECT INTO!
string_agg(quote_ident(attname), ', ')
FROM (
SELECT attname
FROM pg_attribute
WHERE attrelid IN (_tbl1, _tbl2)
AND NOT attisdropped -- no dropped (dead) columns
AND attnum > 0 -- no system columns
GROUP BY 1
HAVING count(*) = 2
) sub;
EXECUTE format('INSERT INTO %1$s(%2$s) SELECT %2$s FROM %3$s'
, _tbl1, columns, _tbl2);
GET DIAGNOSTICS rows = ROW_COUNT; -- return number of rows copied
END
$func$;
Call:
SELECT * FROM f_copy_rows_with_shared_cols('public.person2', 'public.person1');
Result:
rows | columns
-----+---------
3 | name, age
Major points
Note the proper use of SELECT INTO for assignment inside plpgsql.
Note the use of the data type regclass. This allows to use schema-qualified table names (optionally) and defends against SQL injection attempts:
Table name as a PostgreSQL function parameter
About GET DIAGNOSTICS:
Count rows affected by DELETE
About OUT parameters:
Returning from a function with OUT parameter
The manual about format().
Information schema vs. system catalogs.