Inserting DEFAULT value into a column when a param

Question

I would like to write a stored procedure like this:

CREATE OR REPLACE FUNCTION my_function(param_1 text, param_2 text DEFAULT NULL::text) RETURNS bigint AS
$$
DECLARE ret bigint;
BEGIN
    INSERT INTO my_table(val_1, val_2) VALUES (param_1, param_2);

    -- do some more stuff

    RETURN ret;
END;
$$
LANGUAGE plpgsql;

However, I would like to use val_2 column's DEFAULT value instead of NULL - if NULL is provided as the param_2 value.

Something like this:

INSERT INTO my_table(val_1, val_2) VALUES (param_1, COALESCE(param_2, DEFAULT));

is obviously wrong, since the INSERT statement specification explicitly states an expression OR DEFAULT can be used, DEFAULT itself is not available in expressions.

I found two solutions myself but I'm not satisfied with them.

1. Select the DEFAULT value from the information schema and use it in the COALESCE expression.

I'm no expert but it seems like there should be a simpler and more elegant way to do it.

2. Use INSERT and then UPDATE

Like this:

-- ...
INSERT INTO my_table(val_1) VALUES (param_1)
RETURNING id INTO id_var;

IF (param_2) IS NOT NULL THEN
    UPDATE my_table SET val_2 = param_2 WHERE id = id_var;
END IF;
-- ...

There is however a catch in this solution. The actual table of the production system has some intricate triggers which run on UPDATE statements on this table so I would generally like to avoid using updates if possible.

Generally, I'll possibly stick to the second solution but that would possibly require adding some hacks to the aforementioned triggers. But if there is a way to avoid this - I will be very grateful for pointing it out.

Answer 1

Using dynamic-SQL:

CREATE OR REPLACE FUNCTION my_function(param_1 text, param_2 text DEFAULT NULL::text)
  RETURNS bigint AS
$$
DECLARE ret bigint;
BEGIN

 EXECUTE 'INSERT INTO my_table(val_1, val_2) VALUES (' ||
          quote_literal(param_1) || ',' ||
          CASE WHEN param_2 IS NULL THEN 'DEFAULT' ELSE quote_literal(param_2) END ||
          ')';

 RETURN ret;
END;
$$
LANGUAGE plpgsql;

SqlFiddleDemo

Answer 2

This is a bit long for a comment.

One method would be to declare the column NOT NULL. Inserting the NULL value would generate a constraint violation, which you can catch in the insert using on constraint.

This seems like a correct approach. If the column has a default value, then you probably do not want it to be NULL.

Answer 3

If insert is as simple as in the question, I'd use two different INSERT statements, since the resulting code is small and readable and fast:

IF param_2 IS NULL THEN
  INSERT INTO my_table(val_1) VALUES (param_1);
ELSE
  INSERT INTO my_table(val_1, val_2) VALUES (param_1, param_2);
END IF;

For inserts with a greater level of complexity, for example with more than one optional parameter is used, I may opt for the dynamic-SQL solution by @lad2025 above.

Answer 4

As param_2 can only be one of null or not null only one of the selects will return a row to be inserted:

with i as (
    insert into my_table (val_1)
    select param_1
    where param_2 is null
)
insert into my_table (val_1, val_2)
select param_1, param_2
where param_2 is not null

If it is necessary to return the inserted values:

with i_null as (
    insert into my_table (val_1)
    select param_1
    where param_2 is null
    returning *
), i_notnull as (
    insert into my_table (val_1, val_2)
    select param_1, param_2
    where param_2 is not null
    returning *
)
select * from i_null
union all
select * from i_notnull