Dynamically generate columns for crosstab in PostgreSQL

I realise this is an older post but struggled for a little while on the same issue.

My Problem Statement: I had a table with muliple values in a field and wanted to create a crosstab query with 40+ column headings per row.

My Solution was to create a function which looped through the table column to grab values that I wanted to use as column headings within the crosstab query.

Within this function I could then Create the crosstab query. In my use case I added this crosstab result into a separate table.

E.g.

CREATE OR REPLACE FUNCTION field_values_ct ()
 RETURNS VOID AS $$
DECLARE rec RECORD;
DECLARE str text;
BEGIN
str := '"Issue ID" text,';
   -- looping to get column heading string
   FOR rec IN SELECT DISTINCT field_name
        FROM issue_fields
        ORDER BY field_name
    LOOP
    str :=  str || '"' || rec.field_name || '" text' ||',';
    END LOOP;
    str:= substring(str, 0, length(str));

    EXECUTE 'CREATE EXTENSION IF NOT EXISTS tablefunc;
    DROP TABLE IF EXISTS temp_issue_fields;
    CREATE TABLE temp_issue_fields AS
    SELECT *
    FROM crosstab(''select issue_id, field_name, field_value from issue_fields order by 1'',
                 ''SELECT DISTINCT field_name FROM issue_fields ORDER BY 1'')
         AS final_result ('|| str ||')';
END;
$$ LANGUAGE plpgsql;

@erwin-brandstetter: The return type of the function isn't an issue if you're always returning a JSON type with the converted results.

Here is the function I came up with:

CREATE OR REPLACE FUNCTION report.test(
    i_start_date TIMESTAMPTZ,
    i_end_date TIMESTAMPTZ,
    i_interval INT
    ) RETURNS TABLE (
    tab JSON
    ) AS $ab$
DECLARE
    _key_id TEXT;
    _text_op TEXT = '';
    _ret JSON;
BEGIN
    -- SELECT DISTINCT for query results
    FOR _key_id IN
    SELECT DISTINCT at_name
      FROM report.company_data_date cd 
      JOIN report.company_data_amount cda ON cd.id = cda.company_data_date_id 
      JOIN report.amount_types at ON cda.amount_type_id  = at.id 
     WHERE date_start BETWEEN i_start_date AND i_end_date
       AND interval_type_id = i_interval
    LOOP
    -- build function_call with datatype of column
        IF char_length(_text_op) > 1 THEN
            _text_op := _text_op || ', ' || _key_id || ' NUMERIC(20,2)';
        ELSE
            _text_op := _text_op || _key_id || ' NUMERIC(20,2)';
        END IF;
    END LOOP;
    -- build query with parameter filters
    RETURN QUERY
    EXECUTE '
        SELECT array_to_json(array_agg(row_to_json(t)))
          FROM (
        SELECT * FROM crosstab(''SELECT date_start, at.at_name,  cda.amount ct 
          FROM report.company_data_date cd 
          JOIN report.company_data_amount cda ON cd.id = cda.company_data_date_id 
          JOIN report.amount_types at ON cda.amount_type_id  = at.id 
         WHERE date_start between $$' || i_start_date::TEXT || '$$ AND $$' || i_end_date::TEXT || '$$ 
           AND interval_type_id = ' || i_interval::TEXT || ' ORDER BY date_start'') 
            AS ct (date_start timestamptz, ' || _text_op || ')
             ) t;';
END;
$ab$ LANGUAGE 'plpgsql';

So, when you run it, you get the dynamic results in JSON, and you don't need to know how many values were pivoted:

select * from report.test(now()- '1 week'::interval, now(), 1);
                                                                                                                     tab                                                                                                                      
----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
 [{"date_start":"2015-07-27T08:40:01.277556-04:00","burn_rate":0.00,"monthly_revenue":5800.00,"cash_balance":0.00},{"date_start":"2015-07-27T08:50:02.458868-04:00","burn_rate":34000.00,"monthly_revenue":15800.00,"cash_balance":24000.00}]
(1 row)

Edit: If you have mixed datatypes in your crosstab, you can add logic to look it up for each column with something like this:

  SELECT a.attname as column_name, format_type(a.atttypid, a.atttypmod) AS data_type 
    FROM pg_attribute a 
    JOIN pg_class b ON (a.attrelid = b.relfilenode) 
    JOIN pg_catalog.pg_namespace n ON n.oid = b.relnamespace 
   WHERE n.nspname = $$schema_name$$ AND b.relname = $$table_name$$ and a.attstattarget = -1;"

You can use the provided C function crosstab_hash for this.

The manual is not very clear in this respect. It's mentioned at the end of the chapter on crosstab() with two parameters:

You can create predefined functions to avoid having to write out the result column names and types in each query. See the examples in the previous section. The underlying C function for this form of crosstab is named crosstab_hash.

For your example:

CREATE OR REPLACE FUNCTION f_cross_test_db(text, text)
  RETURNS TABLE (kernel_id int, key1 int, key2 int, key3 int)
  AS '$libdir/tablefunc','crosstab_hash' LANGUAGE C STABLE STRICT;

Call:

SELECT * FROM f_cross_test_db(
      'SELECT kernel_id, key, value FROM test_db ORDER BY 1,2'
     ,'SELECT DISTINCT key FROM test_db ORDER BY 1');

Note that you need to create a distinct crosstab_hash function for every crosstab function with a different return type.

Related:

  • PostgreSQL row to columns

Your function to generate the column list is rather convoluted, the result is incorrect (int missing after kernel_id), it can be replaced with this SQL query:

SELECT 'kernel_id int, '
       || string_agg(DISTINCT key::text, ' int, '  ORDER BY key::text)
       || ' int, DUMMY text'
FROM   test_db;

And it cannot be used dynamically anyway.