MacLochlainns Weblog

Picking up where I left off on yesterday’s post on PostgreSQL arrays, you can also write multidimensional arrays provided all the nested arrays are equal in size. You can’t use the CARDINALITY function to determine the length of nested arrays, you must use the ARRAY_LENGTH to determine the length of subordinate arrays.

Here’s an example file with a multidimensional array of integers:

DO
$$
DECLARE
  /* Declare an array of integers with a subordinate array of integers. */
  list  int[][] = array[array[1,2,3,4]
                       ,array[1,2,3,4]
                       ,array[1,2,3,4]
                       ,array[1,2,3,4]
                       ,array[1,2,3,4]];
  row   varchar(20) = '';
BEGIN
  /* Loop through the first dimension of integers. */
  <<Outer>>
  FOR i IN 1..ARRAY_LENGTH(list,1) LOOP
    row = '';
    /* Loop through the second dimension of integers. */
    <<Inner>>
    FOR j IN 1..ARRAY_LENGTH(list,2) LOOP
      IF LENGTH(row) = 0 THEN
        row = row || list[i][j];
      ELSE
        row = row || ',' || list[i][j];
      END IF;
    END LOOP;
    /* Exit outer loop. */
    RAISE NOTICE 'Row [%][%]', i, row;
  END LOOP;
END;
$$;

It prints:

NOTICE:  Row [1][1,2,3,4]
NOTICE:  Row [2][1,2,3,4]
NOTICE:  Row [3][1,2,3,4]
NOTICE:  Row [4][1,2,3,4]
NOTICE:  Row [5][1,2,3,4]
DO

Multidimensional arrays are unique to PostgreSQL but you can have nested lists of tables or varrays inside an Oracle database. Oracle also supports nested lists that are asynchronous.

As always, I hope this helps those trying sort out the syntax.

Somebody asked me how to navigate a collection in PostgreSQL’s PL/pgSQL and whether they supported table and varray data types, like Oracle’s PL/SQL. The most important thing to correct was that PostgreSQL supports only array types.

The only example that I found with a google search used a FOREACH-loop, like this:

1
2
3
4
5
6
7
8
9
10
11
12
13
14

DO
$$
DECLARE
  /* An array of integers. */
  list  int[] = array[1,2,3,4,5];
  /* Define a local variable for array members. */
  i     int;
BEGIN
  /* Loop through the integers. */
  FOREACH i IN ARRAY list LOOP
    RAISE NOTICE '[%]', i;
  END LOOP;
END;
$$;

It prints:

NOTICE:  [1]
NOTICE:  [2]
NOTICE:  [3]
NOTICE:  [4]
NOTICE:  [5]

As I suspected the student didn’t want to use a FOREACH-loop. The student wanted to use a for-loop, which was much closer to the Oracle PL/SQL syntax with which they were most familiar. That example is:

1
2
3
4
5
6
7
8
9
10
11
12

DO
$$
DECLARE
  /* An array of integers. */
  list  int[] = array[1,2,3,4,5];
BEGIN
  /* Loop through the integers. */
  FOR i IN 1..5 LOOP
    RAISE NOTICE '[%]', list[i];
  END LOOP;
END;
$$;

However, it’s bad form to use a literal for the upper number in a range for-loop, and you should use the CARDINALITY function in PostgreSQL because there is no collection API, like Oracle’s COUNT method. There is an ARRAY_LENGTH function but it’s really only necessary when you use a multidimensional array.

The modified code is:

1
2
3
4
5
6
7
8
9
10
11
12

DO
$$
DECLARE
  -- An array of integers.
  list  int[] = array[1,2,3,4,5];
BEGIN
  /* Loop through the integers. */
  FOR i IN 1..CARDINALITY(list) LOOP
    RAISE NOTICE '[%]', list[i];
  END LOOP;
END;
$$;

If you use the ARRAY_LENGTH function, line #8 would look like:

7
8

  /* Loop through the integers, and determines the length of the first dimension. */
  FOR i IN 1..ARRAY_LENGTH(list,1) LOOP

As always, I hope this helps those looking for a clear solution to basic activities.

Transaction Management

Learning Outcomes

• Learn how to use Multiversion Concurrency Control (MVCC).
• Learn how to manage ACID-compliant transactions.
• Learn how to use:
  • SAVEPOINT Statement
  • COMMIT Statement
  • ROLLBACK Statement

Lesson Material

Transaction Management involves two key components. One is Multiversion Concurrency Control (MVCC) so one user doesn’t interfere with another user. The other is data transactions. Data transactions packag SQL statements in the scope of an imperative language that uses Transaction Control Language (TCL) to extend ACID-compliance from single SQL statements to groups of SQL statements.

Multiversion Concurrency Control (MVCC)

Multiversion Concurrency Control (MVCC) uses database snapshots to provide transactions with memory-persistent copies of the database. This means that users, via their SQL statements, interact with the in-memory copies of data rather than directly with physical data. MVCC systems isolate user transactions from each other and guarantee transaction integrity by preventing dirty transactions, writes to the data that shouldn’t happen and that make the data inconsistent. Oracle Database 12c prevents dirty writes by its MVCC and transaction model.

Transaction models depend on transactions, which are ACID-compliant blocks of code. Oracle Database 12c provides an MVCC architecture that guarantees that all changes to data are ACID-compliant, which ensures the integrity of concurrent operations on data—transactions.

ACID-compliant transactions meet four conditions:

Atomic

They complete or fail while undoing any partial changes.

Consistent

They change from one state to another the same way regardless of whether
the change is made through parallel actions or serial actions.

Isolated

Partial changes are never seen by other users or processes in the concurrent system.

Durable

They are written to disk and made permanent when completed.

Oracle Database 12c manages ACID-compliant transactions by writing them to disk first, as redo log files only or as both redo log files and archive log files. Then it writes them to the database. This multiple-step process with logs ensures that Oracle database’s buffer cache (part of the instance memory) isn’t lost from any completed transaction. Log writes occur before the acknowledgement-of-transactions process occurs.

The smallest transaction in a database is a single SQL statement that inserts, updates, or deletes rows. SQL statements can also change values in one or more columns of a row in a table. Each SQL statement is by itself an ACID-compliant and MVCC-enabled transaction when managed by a transaction-capable database engine. The Oracle database is always a transaction-capable system. Transactions are typically a collection of SQL statements that work in close cooperation to accomplish a business objective. They’re often grouped into stored programs, which are functions, procedures, or triggers. Triggers are specialized programs that audit or protect data. They enforce business rules that prevent unauthorized changes to the data.

SQL statements and stored programs are foundational elements for development of business applications. They contain the interaction points between customers and the data and are collectively called the application programming interface (API) to the database. User forms (typically web forms today) access the API to interact with the data. In well-architected business application software, the API is the only interface that the form developer interacts with.

Database developers, such as you and I, create these code components to enforce business rules while providing options to form developers. In doing so, database developers must guard a few things at all cost. For example, some critical business logic and controls must prevent changes to the data in specific tables, even changes in API programs. That type of critical control is often written in database triggers. SQL statements are events that add, modify, or delete data. Triggers guarantee that API code cannot make certain additions, modifications, or deletions to critical resources, such as tables. Triggers can run before or after SQL statements. Their actions, like the SQL statements themselves, are temporary until the calling scope sends an instruction to commit the work performed.

A database trigger can intercept values before they’re placed in a column, and it can ensure that only certain values can be inserted into or updated in a column. A trigger overrides an INSERT or UPDATE statement value that violates a business rule and then it either raises an error and aborts the transaction or changes the value before it can be inserted or updated into the table. Chapter 12 offers examples of both types of triggers in Oracle Database 12c.
MVCC determines how to manage transactions. MVCC guarantees how multiple users’ SQL statements interact in an ACID compliant manner. The next two sections qualify how data transactions work and how MVCC locks and isolates partial results from data transactions.

Data Transaction

Data Manipulation Language (DML) commands are the SQL statements that transact against the data. They are principally the INSERT, UPDATE, and DELETE statements. The INSERT statement adds new rows in a table, the UPDATE statement modifies columns in existing rows, and the DELETE statement removes a row from a table.

The Oracle MERGE statement transacts against data by providing a conditional insert or update feature. The MERGE statement lets you add new rows when they don’t exist or change column values in rows that do exist.

Inserting data seldom encounters a conflict with other SQL statements because the values become a new row or rows in a table. Updates and deletes, on the other hand, can and do encounter conflicts with other UPDATE and DELETE statements. INSERT statements that encounter conflicts occur when columns in a new row match a preexisting row’s uniquely constrained columns. The insertion is disallowed because only one row can contain the unique column set.

These individual transactions have two phases in transactional databases such as Oracle. The first phase involves making a change that is visible only to the user in the current session. The user then has the option of committing the change, which makes it permanent, or rolling back the change, which undoes the transaction. Developers use Transaction Control Language (TCL) commands to confirm or cancel transactions. The COMMIT statement confirms or makes permanent any change, and the ROLLBACK statement cancels or undoes any change.

A generic transaction lifecycle for a two-table insert process implements a business rule that specifies that neither INSERT statement works unless they both work. Moreover, if the first INSERT statement fails, the second INSERT statement never runs; and if the second INSERT statement fails, the first INSERT statement is undone by a ROLLBACK statement to a SAVEPOINT.

After a failed transaction is unwritten, good development practice requires that you write the failed event(s) to an error log table. The write succeeds because it occurs after the ROLLBACK statement but before the COMMIT statement.

A SQL statement followed by a COMMIT statement is called a transaction process, or a two-phase commit (2PC) protocol. ACID-compliant transactions use a 2PC protocol to manage one SQL statement or collections of SQL statements. In a 2PC protocol model, the INSERT, UPDATE, MERGE, or DELETE DML statement starts the process and submits changes. These DML statements can also act as events that fire database triggers assigned to the table being changed.

Transactions become more complex when they include database triggers because triggers can inject an entire layer of logic within the transaction scope of a DML statement. For example, database triggers can do the following:

• Run code that verifies, changes, or repudiates submitted changes
• Record additional information after validation in other tables (they can’t write to the table being changed—or, in database lexicon, “mutated”
• Throw exceptions to terminate a transaction when the values don’t meet business rules

As a general rule, triggers can’t contain a COMMIT or ROLLBACK statement because they run inside the transaction scope of a DML statement. Oracle databases give developers an alternative to this general rule because they support autonomous transactions. Autonomous transactions run outside the transaction scope of the triggering DML statement. They can contain a COMMIT statement and act independently of the calling scope statement. This means an autonomous trigger can commit a transaction when the calling transaction fails.

As independent statements or collections of statements add, modify, and remove rows, one statement transacts against data only by locking rows: the SELECT statement. A SELECT statement typically doesn’t lock rows when it acts as a cursor in the scope of a stored program. A cursor is a data structure that contains rows of one-to-many columns in a stored program. This is also known as a list of record structures.

Cursors act like ordinary SQL queries, except they’re managed by procedure programs row by row. There are many examples of procedural programming languages. PL/SQL and SQL/PSM programming languages are procedural languages designed to run inside the database. C, C++, C#, Java, Perl, and PHP are procedural languages that interface with the database through well-defined interfaces, such as Java Database Connectivity (JDBC) and Open Database Connectivity (ODBC).

Cursors can query data two ways. One way locks the rows so that they can’t be changed until the cursor is closed; closing the cursor releases the lock. The other way doesn’t lock the rows, which allows them to be changed while the program is working with the data set from the cursor. The safest practice is to lock the rows when you open the cursor, and that should always be the case when you’re inserting, updating, or deleting rows that depend on the values in the cursor not changing until the transaction lifecycle of the program unit completes.

Loops use cursors to process data sets. That means the cursors are generally opened at or near the beginning of program units. Inside the loop the values from the cursor support one to many SQL statements for one to many tables.

Stored and external programs create their operational scope inside a database connection when they’re called by another program. External programs connect to a database and enjoy their own operational scope, known as a session scope. The session defines the programs’ operational scope. The operational scope of a stored program or external program defines the transaction scope. Inside the transaction scope, the programs interact with data in tables by inserting, updating, or deleting data until the operations complete successfully or encounter a critical failure. These stored program units commit changes when everything completes successfully, or they roll back changes when any critical instruction fails. Sometimes, the programs are written to roll back changes when any instruction fails.

In the Oracle Database, the most common clause to lock rows is the FOR UPDATE clause, which is appended to a SELECT statement. An Oracle database also supports a WAIT n seconds or NOWAIT option. The WAIT option is a blessing when you want to reply to an end user form’s request and can’t make the change quickly. Without this option, a change could hang around for a long time, which means virtually indefinitely to a user trying to run your application. The default value in an Oracle database is NOWAIT, WAIT without a timeout, or wait indefinitely.

You should avoid this default behavior when developing program units that interact with customers. The Oracle Database also supports a full table lock with the SQL LOCK TABLE command, but you would need to embed the command inside a stored or external program’s instruction set.

A unit test script may contain SQL or PL/SQL statements or it may call another script file that contains SQL or PL/SQL statements. Moreover, a script file is a way to bundle several activities into a single file because most unit test programs typically run two or more instructions as unit tests.

Unconditional Script File

You can write a simple unit test like the example program provided in the Lab 1 Help Section, which includes conditional logic. However, you can write a simpler script that is unconditional and raises exceptions when preconditions do not exist.

The following script file creates a one table and one_s sequence. The DROP TABLE and DROP SEQUENCE statements have the same precondition, which is that the table or sequence must previously exist.

-- Drop table one.
DROP TABLE one;
 
-- Crete table one.
CREATE TABLE one
( one_id    NUMBER
, one_text  VARCHAR2(10));
 
-- Drop sequence one_s.
DROP SEQUENCE one_s;
 
-- Create sequence one_s.
CREATE SEQUENCE one_s;

After writing the script file, you can save it in the lab2 subdirectory as the unconditional.sql file. After you login to the SQL*Plus environment from the lab2 subdirectory. You call the unconditional.sql script file from inside the SQL*Plus environment with the following syntax:

@unconditional.sql

It will display the following output, which raises an exception when the one table or one_s sequence does not already exist in the schema or database:

DROP TABLE one
           *
ERROR at line 1:
ORA-00942: table or view does not exist
 
Table created.
 
DROP SEQUENCE one_s
              *
ERROR at line 1:
ORA-02289: sequence does not exist
 
Sequence created.

An unconditional script raises exceptions when a precondition of the statement does not exist. The precondition is not limited to objects, like the table or sequence; and the precondition may be specific data in one or several rows of one or several tables. You can avoid raising conditional errors by writing conditional scripts.

Conditional Script File

A conditional script file contains statements that check for a precondition before running a statement, which effectively promotes their embedded statements to a lambda function. The following logic recreates the logic of the unconditional.sql script file as a conditional script file:

-- Conditionally drop a table and sequence.
BEGIN
  FOR i IN (SELECT   object_name
            ,        object_type
            FROM     user_objects
            WHERE    object_name IN ('ONE','ONE_S')
            ORDER BY object_type ) LOOP
    IF i.object_type = 'TABLE' THEN
      EXECUTE IMMEDIATE 'DROP TABLE '||i.object_name||' CASCADE CONSTRAINTS';
    ELSE
      EXECUTE IMMEDIATE 'DROP SEQUENCE '||i.object_name;
    END IF;
  END LOOP;
END;
/
 
-- Crete table one.
CREATE TABLE one
( one_id    NUMBER
, one_text  VARCHAR2(10));
 
-- Create sequence one_s.
CREATE SEQUENCE one_s;

You can save this script in the lab2 subdirectory as conditional.sql and then unit test it in SQL*Plus. You must manually drop the one table and one_s sequence before running the conditional.sql script to test the preconditions.

You will see that the conditional.sql script does not raise an exception because the one table or one_s sequence is missing. It should generate output to the console, like this:

PL/SQL procedure successfully completed.
 
Table created.
 
Sequence created.

As a rule, you should always write conditional script files. Unconditional script files throw meaningless errors, which may cause your good code to fail a deployment test that requires error free code.

Selective Aggregation

Learning Outcomes

• Learn how to combine CASE operators and aggregation functions.
• Learn how to selective aggregate values.
• Learn how to use SQL to format report output.

Selective aggregation is the combination of the CASE operator and aggregation functions. Any aggregation function adds, sums, or averages the numbers that it finds; and when you embed the results of a CASE operator inside an aggregation function you get a selective result. The selectivity is determined by the WHEN clause of a CASE operator, which is more or less like an IF statement in an imperative programming language.

The prototype for selective aggregation is illustrated with a SUM function below:

SELECT   SUM(CASE
               WHEN left_operand = right_operand THEN result
               WHEN left_operand > right_operand THEN result
               WHEN left_operand IN (SET OF comma-delimited VALUES) THEN result
               WHEN left_operand IN (query OF results) THEN result
               ELSE alt_result
             END) AS selective_aggregate
FROM     some_table;

A small example let’s you see how selective aggregation works. You create a PAYMENT table and PAYMENT_S sequence for this example, as follows:

-- Create a PAYMENT table.
CREATE TABLE payment
( payment_id     NUMBER
, payment_date   DATE	      CONSTRAINT nn_payment_1 NOT NULL
, payment_amount NUMBER(20,2) CONSTRAINT nn_payment_2 NOT NULL
, CONSTRAINT pk_payment PRIMARY KEY (payment_id));
 
-- Create a PAYMENT_S sequence.
CREATE SEQUENCE payment_s;

After you create the table and sequence, you should insert some data. You can match the values below or choose your own values. You should just insert values for a bunch of rows.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56

DECLARE
  -- Create local collection data types.
  TYPE pmtval IS TABLE OF NUMBER(20,2);
  TYPE smonth IS TABLE OF VARCHAR2(3);
 
  -- Create variable to hold the list of payments.
  payments PMTVAL := pmtval();
 
  -- Declare month arrays.
  short_month SMONTH := smonth('JAN','FEB','MAR','APR','MAY','JUN'
                              ,'JUL','AUG','SEP','OCT','NOV','DEC');
 
  -- Declare variable values.
  month      VARCHAR2(3);
  year       NUMBER := '2019';
  pmt_date   DATE; 
  tpmt_date  VARCHAR2(11);
 
  -- Declare default number of random payments.
  payment_number  NUMBER := 10000;
BEGIN
  -- Populate payment list. 
  FOR i IN 1..payment_number LOOP
    payments.EXTEND;
    SELECT   ROUND(dbms_random.value() * 1000,0)
    ||       '.'
    ||       ROUND(dbms_random.value() * 100,0)
    INTO     payments(payments.COUNT)
    FROM     dual;
  END LOOP;
 
  -- Create and populate payment date and amount.
  FOR i IN 1..payment_number LOOP
    -- Assign random month value.
    month := short_month(dbms_random.value(1,short_month.COUNT));
 
    -- Assign random day of the month value and assemble random date.
    IF month IN ('JAN','MAR','MAY','JUL','AUG','OCT','DEC') THEN
      pmt_date := ROUND(dbms_random.value(1,31),0) || '-' || month || '-' || year;
    ELSIF month IN ('APR','JUN','SEP','NOV') THEN
      pmt_date := ROUND(dbms_random.value(1,30),0) || '-' || month || '-' || year;
    ELSE
      pmt_date := ROUND(dbms_random.value(1,28),0) || '-' || month || '-' || year;
    END IF;
 
    -- Insert values into the PAYMENT table.
    INSERT INTO payment
    ( payment_id, payment_date, payment_amount )
    VALUES
    ( payment_s.NEXTVAL, pmt_date, payments(i));
  END LOOP;
 
  -- Commit the writes.
  COMMIT;
END;
/

After inserting 10,000 rows, you can get an unformatted total with the following query:

-- Query total amount.
SELECT   SUM(payment_amount) AS payment_total
FROM     payment;

It outputs the following:

PAYMENT_TOTAL
-------------
   5011091.75

You can nest the result inside the TO_CHAR function to format the output, like

-- Query total formatted amount.
SELECT   TO_CHAR(SUM(payment_amount),'999,999,999.00') AS payment_total
FROM     payment;

It outputs the following:

PAYMENT_TOTAL
---------------
   5,011,091.75

Somebody may suggest that you use a PIVOT function to rotate the data into a summary by month but the PIVOT function has limits. The pivoting key must be numeric and the column values will use only those numeric values.

-- Pivoted summaries by numeric monthly value.
SELECT   *
FROM    (SELECT EXTRACT(MONTH FROM payment_date) payment_month
         ,      payment_amount
         FROM   payment)
         PIVOT (SUM(payment_amount) FOR payment_month IN
                 (1,2,3,4,5,6,7,8,9,10,11,12));

It outputs the following:

	 1	    2	       3	  4	     5		6	   7	      8 	 9	   10	      11	 12
---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ---------- ----------
 245896.55  430552.36  443742.63  457860.27  470467.18	466370.71  415158.28  439898.72  458998.09  461378.56  474499.22  246269.18

You can use selective aggregation to get the results by a character label, like

SELECT   SUM(
           CASE
             WHEN EXTRACT(MONTH FROM payment_date) = 1
             AND  EXTRACT(YEAR FROM payment_date) = 2019  THEN payment_amount
           END) AS "JAN"
,        SUM(
           CASE
             WHEN EXTRACT(MONTH FROM payment_date) = 2
             AND  EXTRACT(YEAR FROM payment_date) = 2019  THEN payment_amount
           END) AS "FEB"
,        SUM(
           CASE
             WHEN EXTRACT(MONTH FROM payment_date) = 3
             AND  EXTRACT(YEAR FROM payment_date) = 2019  THEN payment_amount
           END) AS "MAR"
,        SUM(
           CASE
             WHEN EXTRACT(MONTH FROM payment_date) IN (1,2,3)
             AND  EXTRACT(YEAR FROM payment_date) = 2019 THEN payment_amount
           END) AS "1FQ"
,        SUM(
           CASE
             WHEN EXTRACT(MONTH FROM payment_date) = 4
             AND  EXTRACT(YEAR FROM payment_date) = 2019  THEN payment_amount
           END) AS "APR"
FROM     payment;

It outputs the following:

       JAN	  FEB	     MAR	1FQ	   APR
---------- ---------- ---------- ---------- ----------
 245896.55  430552.36  443742.63 1120191.54  457860.27

You can format the output with a combination of the TO_CHAR and LPAD functions. The TO_CHAR allows you to add a formatting mask, complete with commas and two mandatory digits to the right of the decimal point. The reformatted query looks like

COL JAN FORMAT A13 HEADING "Jan"
COL FEB FORMAT A13 HEADING "Feb"
COL MAR FORMAT A13 HEADING "Mar"
COL 1FQ FORMAT A13 HEADING "1FQ"
COL APR FORMAT A13 HEADING "Apr"
SELECT   LPAD(TO_CHAR(SUM(
           CASE
             WHEN EXTRACT(MONTH FROM payment_date) = 1
             AND  EXTRACT(YEAR FROM payment_date) = 2019  THEN payment_amount
           END),'9,999,999.00'),13,' ') AS "JAN"
,        LPAD(TO_CHAR(SUM(
           CASE
             WHEN EXTRACT(MONTH FROM payment_date) = 2
             AND  EXTRACT(YEAR FROM payment_date) = 2019  THEN payment_amount
           END),'9,999,999.00'),13,' ') AS "FEB"
,        LPAD(TO_CHAR(SUM(
           CASE
             WHEN EXTRACT(MONTH FROM payment_date) = 3
             AND  EXTRACT(YEAR FROM payment_date) = 2019  THEN payment_amount
           END),'9,999,999.00'),13,' ') AS "MAR"
,        LPAD(TO_CHAR(SUM(
           CASE
             WHEN EXTRACT(MONTH FROM payment_date) IN (1,2,3)
             AND  EXTRACT(YEAR FROM payment_date) = 2019 THEN payment_amount
           END),'9,999,999.00'),13,' ') AS "1FQ"
,        LPAD(TO_CHAR(SUM(
           CASE
             WHEN EXTRACT(MONTH FROM payment_date) = 4
             AND  EXTRACT(YEAR FROM payment_date) = 2019  THEN payment_amount
           END),'9,999,999.00'),13,' ') AS "APR"
FROM     payment;

It displays the formatted output:

Jan	      Feb	    Mar 	  1FQ		Apr
------------- ------------- ------------- ------------- -------------
   245,896.55	 430,552.36    443,742.63  1,120,191.54    457,860.27

INSERT Statement

Learning Outcomes

• Learn how to use positional- and named-notation in INSERT statements.
• Learn how to use the VALUES clause in INSERT statements.
• Learn how to use subqueries in INSERT statements.

The INSERT statement lets you enter data into tables and views in two ways: via an INSERT statement with a VALUES clause and via an INSERT statement with a query. The VALUES clause takes a list of literal values (strings, numbers, and dates represented as strings), expression values (return values from functions), or variable values.

Query values are results from SELECT statements that are subqueries (covered earlier in this appendix). INSERT statements work with scalar, single-row, and multiple-row subqueries. The list of columns in the VALUES clause or SELECT clause of a query (a SELECT list) must map to the positional list of columns that defines the table. That list is found in the data dictionary or catalog. Alternatively to the list of columns from the data catalog, you can provide a named list of those columns. The named list overrides the positional (or default) order from the data catalog and must provide at least all mandatory columns in the table definition. Mandatory columns are those that are not null constrained.

Oracle databases differ from other databases in how they implement the INSERT statement. Oracle doesn’t support multiple-row inserts with a VALUES clause. Oracle does support default and override signatures as qualified in the ANSI SQL standards. Oracle also provides a multiple- table INSERT statement. This section covers how you enter data with an INSERT statement that is based on a VALUES clause or a subquery result statement. It also covers multiple-table INSERT statements.

The INSERT statement has one significant limitation: its default signature. The default signature is the list of columns that defines the table in the data catalog. The list is defined by the position and data type of columns. The CREATE statement defines the initial default signature, and the ALTER statement can change the number, data types, or ordering of columns in the default signature.

The default prototype for an INSERT statement allows for an optional column list that overrides the default list of columns. When you provide the column list you choose to implement named-notation, which is the right way to do it. Relying on the insertion order of the columns is a bad idea. An INSERT statement without a list of column names is a position-notation statement. Position-notation is bad because somebody can alter that order and previously written INSERT statements will break or put data in the wrong columns.

Like methods in OOPLs, an INSERT statement without the optional column list constructs an instance (or row) of the table using the default constructor. The override constructor for a row is defined by any INSERT statement when you provide an optional column list. That’s because it overrides the default constructor.

The generic prototype for an INSERT statement is confusing when it tries to capture both the VALUES clause and the result set from a query. Therefore, I’ve opted to provide two generic prototypes.

Insert by value

The first uses the VALUES clause:

INSERT
INTO table_name
[( column1, column2, column3, ...)] VALUES
( value1, value2, value3, ...);

Notice that the prototype for an INSERT statement with the result set from a query doesn’t use the VALUES clause at all. A parsing error occurs when the VALUES clause and query both occur in an INSERT statement.

The second prototype uses a query and excludes the VALUES clause. The subquery may return one to many rows of data. The operative rule is that all columns in the query return the same number of rows of data, because query results should be rectangles—rectangles made up of one to many rows of columns.

Insert by subquery

Here’s the prototype for an INSERT statement that uses a subquery:

INSERT
INTO table_name
[( column1, column2, column3, ...)]
( SELECT value1, value2, value3, ... FROM table_name WHERE ...);

A query, or SELECT statement, returns a SELECT list. The SELECT list is the list of columns, and it’s evaluated by position and data type. The SELECT list must match the definition of the table or the override signature provided.

Default signatures present a risk of data corruption through insertion anomalies, which occur when you enter bad data in tables. Mistakes transposing or misplacing values can occur more frequently with a default signature, because the underlying table structure can change. As a best practice, always use named notation by providing the optional list of values; this should help you avoid putting the right data in the wrong place.

The following subsections provide examples that use the default and override syntax for INSERT statements in Oracle databases. The subsections also cover multiple-table INSERT statements and a RETURNING INTO clause, which is an extension of the ANSI SQL standard. Oracle uses the RETURNING INTO clause to manage large objects, to return autogenerated identity column values, and to support some of the features of Oracle’s dynamic SQL. Note that Oracle also supports a bulk INSERT statement, which requires knowledge of PL/SQL.

DESCRIBE table_name

Name                                 Null?    Type
------------------------------------ -------- --------
RENTAL_ID                            NOT NULL NUMBER
CUSTOMER_ID                          NOT NULL NUMBER
CHECK_OUT_DATE                       NOT NULL DATE
RETURN_DATE                                   DATE
CREATED_BY                           NOT NULL NUMBER
CREATION_DATE                        NOT NULL DATE
LAST_UPDATED_BY                      NOT NULL NUMBER
LAST_UPDATE_DATE                     NOT NULL DATE

Name                                      Null?    Type
------------------------------------ -------- --------
RENTAL_ID                            NOT NULL NUMBER
CUSTOMER_ID                          NOT NULL NUMBER
CHECK_OUT_DATE                       NOT NULL DATE
RETURN_DATE                                   DATE
CREATED_BY                           NOT NULL NUMBER
CREATION_DATE                        NOT NULL DATE
LAST_UPDATED_BY                      NOT NULL NUMBER
LAST_UPDATE_DATE                     NOT NULL DATE

SQL> INSERT INTO rental
  2  VALUES
  3  ( 1086
  4  , 1009
  5  , SYSDATE
  6  , TRUNC(SYSDATE + 5) 7 ,1
  8  , SYSDATE
  9  , 1
 10  , SYSDATE);

  5  , TO_DATE('15-APR-2011 12:53:01','DD-MON-YYYY HH24:MI:SS')

  5  , CAST('15-APR-2011 12:53:02' AS DATE)

  5  ,  CAST('15-APR-2011 12:53:02' AS DATE) FROM dual
        *
ERROR AT line 1:
ORA-01830: DATE format picture ends before converting entire input string

  5  ,  CAST(CAST('15-APR-2011 12:53:02' AS TIMESTAMP) AS DATE)

SQL> SELECT   rental_s1.NEXTVAL FROM dual;

  3  ( rental_s1.NEXTVAL

SQL> SELECT   contact_id
  2  FROM     contact
  3  WHERE    last_name = 'Potter'
  4  AND      first_name = 'Harry';

SQL> INSERT INTO rental
  2  ( rental_id
  3  , customer_id
  4  , check_out_date
  5  , return_date
  6  , created_by
  7  , creation_date
  8  , last_updated_by
  9  , last_update_date )
10 VALUES
 11  ( rental_s1.NEXTVAL
 12  ,(SELECT   contact_id
 13    FROM     contact
 14    WHERE    last_name = 'Potter'
 15    AND      first_name = 'Harry')
 16  , SYSDATE
 17  , TRUNC(SYSDATE + 5)
 18  , 1
 19  , SYSDATE
 20  , 3
 21  , SYSDATE);

,(SELECT   contact_id
  *
ERROR AT line 3:
ORA-01427: single-ROW subquery returns more than one ROW

Name                             Null?    Type
 ------------------------------- -------- -----------------------
ITEM_DESC                        NOT NULL CLOB
ITEM_ICON                        NOT NULL BLOB
ITEM_PHOTO                                BINARY FILE LOB

EMPTY_BLOB()
EMPTY_CLOB()

 10  , BFILENAME('VIRTUAL_DIRECTORY_NAME', 'file_name.png')

INSERT INTO some_table
[( column1, column2, column3, ...)]
VALUES
( value1, value2, value3, ...)
RETURNING column1 INTO local_variable;

Name                                 Null?    Type
------------------------------------ -------- --------------
PARENT_ID                            NOT NULL NUMBER
PARENT_NAME                                   VARCHAR2(10)

Name                                 Null?    Type
------------------------------------ -------- --------------
CHILD_ID                             NOT NULL NUMBER
PARENT_ID                                     NUMBER
PARENT_NAME                                   VARCHAR2(10)

SQL> INSERT INTO parent
  2  VALUES
  3  ( parent_s1.NEXTVAL 4 ,'One Parent');
 
SQL> INSERT INTO child
  2  VALUES
  3  ( child_s1.NEXTVAL 4 , parent_s1.CURRVAL 5 ,'One Child');

SQL> INSERT INTO child
  2  VALUES
  3  ( child_s1.NEXTVAL -- CHILD_ID
  4  , parent_s1.CURRVAL -- PARENT_ID
  5  ,'One Child') -- CHILD_NAME
  6  /

INSERT
INTO table_name
[( column1, column2, column3, ...)]
( SELECT value1, value2, value3, ... FROM table_name WHERE ...);

A student wanted a better example of writing a MySQL Persistent Stored Module (PSM) that maintains transactional scope across a couple tables. Here’s the one I wrote about ten years ago to build the MySQL Video Store model. It looks I neglected to put it out there before, so here it is for reference.

-- Conditionally drop procedure if it exists.
DROP PROCEDURE IF EXISTS contact_insert;
 
-- Reset the delimiter so that a semicolon can be used as a statement and block terminator.
DELIMITER $$
 
SELECT 'CREATE PROCEDURE contact_insert' AS "Statement";
CREATE PROCEDURE contact_insert
( pv_member_type         CHAR(12)
, pv_account_number      CHAR(19)
, pv_credit_card_number  CHAR(19)
, pv_credit_card_type    CHAR(12)
, pv_first_name          CHAR(20)
, pv_middle_name         CHAR(20)
, pv_last_name           CHAR(20)
, pv_contact_type        CHAR(12)
, pv_address_type        CHAR(12)
, pv_city                CHAR(30)
, pv_state_province      CHAR(30)
, pv_postal_code         CHAR(20)
, pv_street_address      CHAR(30)
, pv_telephone_type      CHAR(12)
, pv_country_code        CHAR(3)
, pv_area_code           CHAR(6)
, pv_telephone_number    CHAR(10)) MODIFIES SQL DATA
 
BEGIN
 
  /* Declare variables to manipulate auto generated sequence values. */
  DECLARE member_id            int unsigned;
  DECLARE contact_id           int unsigned;
  DECLARE address_id           int unsigned;
  DECLARE street_address_id    int unsigned;
  DECLARE telephone_id         int unsigned;
 
  /* Declare local constants for who-audit columns. */
  DECLARE lv_created_by        int unsigned DEFAULT 1001;
  DECLARE lv_creation_date     DATE         DEFAULT UTC_DATE();
  DECLARE lv_last_updated_by   int unsigned DEFAULT 1001;
  DECLARE lv_last_update_date  DATE         DEFAULT UTC_DATE();
 
  /* Declare a locally scoped variable. */
  DECLARE duplicate_key INT DEFAULT 0;
 
  /* Declare a duplicate key handler */
  DECLARE CONTINUE HANDLER FOR 1062 SET duplicate_key = 1;
 
  /* Start the transaction context. */
  START TRANSACTION;
 
  /* Create a SAVEPOINT as a recovery point. */
  SAVEPOINT all_or_none;
 
  /* Insert into the first table in sequence based on inheritance of primary keys by foreign keys. */
  INSERT INTO member
  ( member_type
  , account_number
  , credit_card_number
  , credit_card_type
  , created_by
  , creation_date
  , last_updated_by
  , last_update_date )
  VALUES
  ((SELECT   common_lookup_id
    FROM     common_lookup
    WHERE    common_lookup_context = 'MEMBER'
    AND      common_lookup_type = pv_member_type)
  , pv_account_number
  , pv_credit_card_number
  ,(SELECT   common_lookup_id
    FROM     common_lookup
    WHERE    common_lookup_context = 'MEMBER'
    AND      common_lookup_type = pv_credit_card_type)
  , lv_created_by
  , lv_creation_date
  , lv_last_updated_by
  , lv_last_update_date );
 
  /* Preserve the sequence by a table related variable name. */
  SET member_id = last_insert_id();
 
  /* Insert into the first table in sequence based on inheritance of primary keys by foreign keys. */
  INSERT INTO contact
  VALUES
  ( null
  , member_id
  ,(SELECT   common_lookup_id
    FROM     common_lookup
    WHERE    common_lookup_context = 'CONTACT'
    AND      common_lookup_type = pv_contact_type)
  , pv_first_name
  , pv_middle_name
  , pv_last_name
  , lv_created_by
  , lv_creation_date
  , lv_last_updated_by
  , lv_last_update_date );  
 
  /* Preserve the sequence by a table related variable name. */
  SET contact_id = last_insert_id();
 
  /* Insert into the first table in sequence based on inheritance of primary keys by foreign keys. */
  INSERT INTO address
  VALUES
  ( null
  , last_insert_id()
  ,(SELECT   common_lookup_id
    FROM     common_lookup
    WHERE    common_lookup_context = 'MULTIPLE'
    AND      common_lookup_type = pv_address_type)
  , pv_city
  , pv_state_province
  , pv_postal_code
  , lv_created_by
  , lv_creation_date
  , lv_last_updated_by
  , lv_last_update_date );  
 
  /* Preserve the sequence by a table related variable name. */
  SET address_id = last_insert_id();
 
  /* Insert into the first table in sequence based on inheritance of primary keys by foreign keys. */
  INSERT INTO street_address
  VALUES
  ( null
  , last_insert_id()
  , pv_street_address
  , lv_created_by
  , lv_creation_date
  , lv_last_updated_by
  , lv_last_update_date );  
 
  /* Insert into the first table in sequence based on inheritance of primary keys by foreign keys. */
  INSERT INTO telephone
  VALUES
  ( null
  , contact_id
  , address_id
  ,(SELECT   common_lookup_id
    FROM     common_lookup
    WHERE    common_lookup_context = 'MULTIPLE'
    AND      common_lookup_type = pv_telephone_type)
  , pv_country_code
  , pv_area_code
  , pv_telephone_number
  , lv_created_by
  , lv_creation_date
  , lv_last_updated_by
  , lv_last_update_date);
 
  /* This acts as an exception handling block. */  
  IF duplicate_key = 1 THEN
 
    /* This undoes all DML statements to this point in the procedure. */
    ROLLBACK TO SAVEPOINT all_or_none;
 
  END IF;
 
  /* This commits the write when successful and is harmless otherwise. */
  COMMIT;
 
END;
$$
 
-- Reset the standard delimiter to let the semicolon work as an execution command.
DELIMITER ;

You can then call the procedure, like:

SELECT 'CALL contact_insert() PROCEDURE 5 times' AS "Statement";
CALL contact_insert('INDIVIDUAL','R11-514-34','1111-1111-1111-1111','VISA_CARD','Goeffrey','Ward','Clinton','CUSTOMER','HOME','Provo','Utah','84606','118 South 9th East','HOME','011','801','423\-1234');
CALL contact_insert('INDIVIDUAL','R11-514-35','1111-2222-1111-1111','VISA_CARD','Wendy',null,'Moss','CUSTOMER','HOME','Provo','Utah','84606','1218 South 10th East','HOME','011','801','423-1234');
CALL contact_insert('INDIVIDUAL','R11-514-36','1111-1111-2222-1111','VISA_CARD','Simon','Jonah','Gretelz','CUSTOMER','HOME','Provo','Utah','84606','2118 South 7th East','HOME','011','801','423-1234');
CALL contact_insert('INDIVIDUAL','R11-514-37','1111-1111-1111-2222','MASTER_CARD','Elizabeth','Jane','Royal','CUSTOMER','HOME','Provo','Utah','84606','2228 South 14th East','HOME','011','801','423-1234');
CALL contact_insert('INDIVIDUAL','R11-514-38','1111-1111-3333-1111','VISA_CARD','Brian','Nathan','Smith','CUSTOMER','HOME','Spanish Fork','Utah','84606','333 North 2nd East','HOME','011','801','423-1234');

I hope this code complete approach helps those looking to learn how to write MySQL PSMs.

If you’re wondering about this post, it shows the basic array of a set of integers and strings before showing you how to create nested tables of data in PostgreSQL. By the way, they’re not called nested tables in PostgreSQL, like they are in Oracle but perform like their Oracle cousins.

Let’s create a table with an auto-incrementing column and two arrays, one array of integers and another of strings:

-- Conditionally drop the demo table.
DROP TABLE IF EXISTS demo;
 
-- Create the test table.
CREATE TABLE demo
( demo_id     serial
, demo_number integer[5]
, demo_string varchar(5)[7]);

You can insert test values like this:

INSERT INTO demo
(demo_number, demo_string)
VALUES
( array[1,2,3,4,5]
, array['One','Two','Three','Four','Five','Six','Seven']);

Then, you can query them with this unnest function, like:

SELECT  unnest(demo_number) AS numbers
,       unnest(demo_string) AS strings
FROM    demo;

It returns:

 numbers | strings
---------+---------
 1       | One
 2       | Two
 3       | Three
 4       | Four
 5       | Five
         | Six
         | Seven
(7 rows)

You may note that the two arrays are asymmetrical. It only becomes an issue when you navigate the result in a PL/pgSQL cursor or imperative programming language, like Python.

Now, let’s do something more interesting like work with a composite user-defined type, like the player structure. You would create the composite user-defined type with this syntax:

-- Conditionally drop the player type.
DROP TYPE IF EXISTS player;
 
-- Create the player type.
CREATE TYPE player AS
( player_no        integer
, player_name      varchar(24)
, player_position  varchar(14)
, ab               integer
, r                integer
, h                integer
, bb               integer
, rbi              integer );

You can create a world_series table that include a players column that uses an array of player type, like

-- Conditionally drop the world_series table.
DROP TABLE IF EXISTS world_series;
 
-- Create the player type.
CREATE TABLE world_series
( world_series_id  serial
, team             varchar(24)
, players          player[30]
, game_no          integer
, year             integer );

If you’re familiar with the Oracle Database, you’d have to specify a tested table in the syntax. Fortunately, PostgreSQL doesn’t require that.

Insert two rows with the following statement:

INSERT INTO world_series
( team
, players
, game_no
, year )
VALUES
('San Francisco Giants'
, array[(24,'Willie Mayes','Center Fielder',5,0,1,0,0)::player
       ,(5,'Tom Haller','Catcher',4,1,2,0,2)::player]
, 4
, 1962 );

You can append to the array with the following syntax. A former student and I have a disagreement on whether this is shown in the PostgreSQL 8.15.4 Modifying Array documentation. I believe it can be inferred from the document and he doesn’t believe so. Anyway, here’s how you add an element to an existing array in a table with the UPDATE statement:

UPDATE world_series
SET    players = (SELECT array_append(players,(7,'Henry Kuenn','Right Fielder',3,0,0,1,0)::player) FROM world_series)
WHERE  team = 'San Francisco Giants'
AND    year = 1962
AND    game_no = 4;

Like Oracle’s nested tables, PostgreSQL’s arrays of composite user-defined types requires writing a PL/pgSQL function. I’ll try to add one of those shortly in another blog entry to show you how to edit and replace entries in stored arrays of composite user-defined types.

You can query the unnested rows and get a return set like a Python tuple with the following query:

SELECT unnest(players) AS player_list
FROM   world_series
WHERE  team = 'San Francisco Giants'
AND    year = 1962
AND    game_no = 4;

It returns the three rows from the players array:

                 player_list
----------------------------------------------
 (24,"Willie Mayes","Center Field",5,0,1,0,0)
 (5,"Tom Haller",Catcher,4,1,2,0,2)
 (7,"Henry Kuenn","Right Fielde",3,0,0,1,0)
(3 rows)

It returns the data set in entry-order. If we step outside of the standard 8.15 Arrays PostgreSQL Documentation, you can do much more with arrays (or nested tables). The balance of this example demonstrates some new syntax that helps you achieve constructive outcomes in PostgreSQL.

You can use a Common Table Expression (CTE) to get the columnar display of the player composite user-defined type. This type of solution is beyond the standard , like:

WITH list AS
 (SELECT unnest(players) AS row_result
  FROM   world_series
  WHERE  team = 'San Francisco Giants'
  AND    year = 1962
  AND    game_no = 4)
SELECT  (row_result).player_name
,       (row_result).player_no
,       (row_result).player_position
FROM     list;

If you’re unfamiliar with accessing composite user-defined types, I wrote a post on that 7 years ago. You can find the older blog entry PostgreSQL Composites on my blog.

It returns only the three requested columns of the player composite user-defined type:

 player_name  | player_no | player_position
--------------+-----------+-----------------
 Willie Mayes |        24 | Center Fielder
 Tom Haller   |         5 | Catcher
 Henry Kuenn  |         7 | Right Fielder
(3 rows)

You should note that the data is presented in an entry-ordered manner when unnested alone in the SELECT-list. That behavior changes when the SELECT-list includes non-array data.

The easiest way to display data from the non-array and array columns is to list them inside the SELECT-list of the CTE, like:

WITH list AS
 (SELECT game_no AS game
  ,      year
  ,      unnest(players) AS row_result
  FROM   world_series
  WHERE  team = 'San Francisco Giants'
  AND    year = 1962
  AND    game_no = 4)
SELECT   game
,        year 
,       (row_result).player_name
,       (row_result).player_no
,       (row_result).player_position
FROM     list;

It returns an ordered set of unnested rows when you include non-array columns, like:

 game | year | player_name  | player_no | player_position
------+------+--------------+-----------+-----------------
    4 | 1962 | Henry Kuenn  |         7 | Right Fielder
    4 | 1962 | Tom Haller   |         5 | Catcher
    4 | 1962 | Willie Mayes |        24 | Center Fielder
(3 rows)

While you can join the world_series table to the unnested array rows (returned as a derived table, its a bad idea. The mechanics to do it require you to return the primary key column in the same SELECT-list of the CTE. Then, you join the CTE list to the world_series table by using the world_series_id primary key.

However, there is no advantage to an inner join approach and it imposes unnecessary processing on the database server. The odd rationale that I heard when I noticed somebody was using a CTE to base-table join was: “That’s necessary so they could use column aliases for the non-array columns.” That’s not true because you can use the aliases inside the CTE, as shown above when game is an alias to the game_no column.

As always, I hope this helps those looking to solve a problem in PostgreSQL.

In MySQL, the @@sql_mode parameter should generally use ONLY_FULL_GROUP_BY. If it doesn’t include it and you don’t have the ability to change the database parameters, you can use a MySQL PSM (Persistent Stored Module), like:

Create the set_full_group_by procedure:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27

-- Drop procedure conditionally on whether it exists already.
DROP PROCEDURE IF EXISTS set_full_group_by;
 
-- Reset delimter to allow semicolons to terminate statements.
DELIMITER $$
 
-- Create a procedure to verify and set connection parameter.
CREATE PROCEDURE set_full_group_by()
  LANGUAGE SQL
  NOT DETERMINISTIC
  SQL SECURITY DEFINER
  COMMENT 'Set connection parameter when not set.'
BEGIN
 
  /* Check whether full group by is set in the connection and
     if unset, set it in the scope of the connection. */
  IF NOT EXISTS
    (SELECT NULL
     WHERE  REGEXP_LIKE(@@SQL_MODE,'ONLY_FULL_GROUP_BY'))
  THEN
    SET SQL_MODE=(SELECT CONCAT(@@sql_mode,',ONLY_FULL_GROUP_BY'));
  END IF;
END;
$$
 
-- Reset the default delimiter.
DELIMITER ;

Run the following SQL command before you attempt the exercises in the same session scope:

CALL set_full_group_by();

As always, I hope this helps those looking for a solution. Naturally, you can simply use the SET command on line #21 above.

In October 2019, I wrote a post with anonymous block programs to drop tables, sequences, routines, and triggers. Two weeks later, I wrote another post to drop all overloaded routines. However, I recognized the other day that I should have written a function that let you target which function or procedure you want to drop.

The older code only let you drop all of your functions or procedures. That was overkill when you’re working on new functions or procedures.

This post provides a utility for those writing functions and procedures in a public schema of any database in a PostgreSQL installation. It is designed to drop functions or procedures from the public schema.

The code follows below:

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104

CREATE OR REPLACE
  FUNCTION drop_routine( IN pv_routine_name    VARCHAR(64)
                       , IN pv_routine_type    VARCHAR(64))
  RETURNS INTEGER AS
  $$
  DECLARE
    /* Declare the current catalog. */
    lv_local_catalog  VARCHAR(64) := current_database();
 
    /* Declare return type variable. */
    lv_retval  INTEGER := 1;
 
    /* Manage data dictionary case mechanics:
       ======================================
	     routine_name is always in lowercase.
	     routine_type is always in uppercase. */
    lv_routine_name  VARCHAR(64) := LOWER(pv_routine_name);
    lv_routine_type  VARCHAR(64) := UPPER(pv_routine_type);
 
    /* Declare an indefinite length string for SQL statement. */
    sql  VARCHAR;
 
    /* Declare variables to manage cursor return values. */
    row  RECORD;
    arg  VARCHAR;
 
    /* Declare parameter list. */
    list VARCHAR;
 
    /* Declare a routine cursor. */
    routine_cursor CURSOR( cv_routine_name  VARCHAR
                         , cv_routine_type  VARCHAR ) FOR
      SELECT r.routine_name
      ,      r.specific_name
      ,      r.routine_type
      FROM   information_schema.routines r
      WHERE  r.specific_catalog = current_database()
      AND    r.routine_schema = 'public'
      AND    r.routine_type = cv_routine_type
      AND    r.routine_name = cv_routine_name;
 
    /* Declare a parameter cursor. */
    parameter_cursor CURSOR( cv_specific_name  VARCHAR ) FOR
      SELECT args.data_type
      FROM   information_schema.parameters args
      WHERE  args.specific_catalog = current_database()
      AND    args.specific_schema = 'public'
      AND    args.specific_name = cv_specific_name;
 
  BEGIN
    /* Open the cursor. */
    OPEN routine_cursor(lv_routine_name, lv_routine_type);
    <<row_loop>>
    LOOP
      /* Fetch table names. */
      FETCH routine_cursor INTO row;
 
      /* Exit when no more records are found. */
      EXIT row_loop WHEN NOT FOUND;
 
      /* Initialize parameter list. */
      list := '(';
 
      /* Open the parameter cursor. */
      OPEN parameter_cursor(row.specific_name::varchar);
      <<parameter_loop>>
      LOOP
        FETCH parameter_cursor INTO arg;
 
        /* Exit the parameter loop. */
        EXIT parameter_loop WHEN NOT FOUND;
 
        /* Add parameter and delimit more than one parameter with a comma. */
        IF LENGTH(list) > 1 THEN
          list := CONCAT(list,',',arg);
        ELSE
          list := CONCAT(list,arg);
        END IF;
      END LOOP;
 
      /* Close the parameter list. */
      list := CONCAT(list,')');
 
      /* Close the parameter cursor. */
      CLOSE parameter_cursor;
 
      /* Concatenate together a DDL to drop the table with prejudice. */
      sql := 'DROP '||row.routine_type||' IF EXISTS '||row.routine_name||list;
 
      /* Execute the DDL statement. */
      EXECUTE sql;
 
      /* Assign success flag of 0. */
      lv_retval := 0;
 
    END LOOP;
 
    /* Close the routine_cursor. */
    CLOSE routine_cursor;
 
    /* Return the output text variable. */
    RETURN lv_retval;
  END
  $$ LANGUAGE plpgsql;

If you now create a series of hello overloaded functions, like:

CREATE OR REPLACE
  FUNCTION hello()
  RETURNS text AS
  $$
  DECLARE
    output  VARCHAR;
  BEGIN
    SELECT 'Hello World!' INTO output;
    RETURN output;
  END
  $$ LANGUAGE plpgsql;
 
CREATE OR REPLACE
  FUNCTION hello(whom text)
  RETURNS text AS
  $$
  DECLARE
    output  VARCHAR;
  BEGIN
    SELECT CONCAT('Hello ',whom,'!') INTO output;
    RETURN output;
  END
  $$ LANGUAGE plpgsql;
 
CREATE OR REPLACE
  FUNCTION hello(id int, whom text)
  RETURNS text AS
  $$
  DECLARE
    output  VARCHAR;
  BEGIN
    SELECT CONCAT('[',id,'] Hello ',whom,'!') INTO output;
    RETURN output;
  END
  $$ LANGUAGE plpgsql;

After you create the overloaded functions, you can query their status from the information_schema.routines table in the data dictionary:

SELECT routine_name
,      specific_name
,      routine_type
FROM   information_schema.routines
WHERE  specific_catalog = current_setting('videodb.catalog_name')
AND    routine_schema = 'public'
AND    routine_name = 'hello';

Which shows you the three versions of the hello function:

 routine_name | specific_name | routine_type
--------------+---------------+--------------
 hello        | hello_18100   | FUNCTION
 hello        | hello_18101   | FUNCTION
 hello        | hello_18102   | FUNCTION
(3 rows)

You can drop all versions of the hello functions by calling the drop_routine function:

SELECT CASE
         WHEN drop_routine('hello','function') = 0
         THEN 'Success'
         ELSE 'Failure'
       END AS drop_routine;

It returns the following:

 drop_routine
--------------
 Success
(1 row)

As always, I hope this helps those looking for how to routinely test new functions and procedures.