Pad arrays with NULL to maximum length for custom aggregate function

Question

From the answer of the question How to use array_agg() for varchar[],

We can create a custom aggregate function to aggregate n-dimensional arrays in Postgres like:

Answer 1

Using the custom aggregate function array_agg_mult() like defined in this related answer:

• Selecting data into a Postgres array

Your expected result is impossible:

{{1},NULL,{abc}}

Would have to be:

{{1},{NULL},{abc}}

Simple case with 0 or 1 array elements

For the simple case to just replace the empty array: You can achieve that with:

WITH t(arr) AS (
    VALUES
      ('{1}'::text[])
     ,('{}')
     ,('{abc}')
   )
SELECT array_agg_mult(ARRAY[CASE WHEN arr = '{}' THEN '{NULL}' ELSE arr END])
FROM   t;

Dynamic padding for n elements

Using array_fill() to pad arrays with NULL elements up to the maximum length:

SELECT array_agg_mult(ARRAY[
         arr || array_fill(NULL::text
                         , ARRAY[max_elem - COALESCE(array_length(arr, 1), 0)])
       ]) AS result
FROM   t, (SELECT max(array_length(arr, 1)) AS max_elem FROM t) t1;

Still only works for 1-dimensional basic arrays.

Explain

• Subquery t1 computes the maximum length of the basic 1-dimensional array.
• COALESCE(array_length(arr, 1), 0) computes the length of the array in this row.
  COALESCE defaults to 0 for NULL.
• Generate padding array for the difference in length with array_fill().
• Append that to arr with ||
• Aggregate like above with array_agg_mult().

SQL Fiddle. demonstrating all.
_{Output in SQL Fiddle is misleading, so I cast result to text there.}