MacLochlainns Weblog

While preparing for next week’s classes, I ran across a new future feature of MySQL – the virtual column. It appears, according to the article, that MySQL will have virtual columns in MySQL 6.

The syntax is virtually identical to that found in Oracle databases. For example, here’s what suppose to work in the future:

1
2
3
4
5

CREATE TABLE salary
( salary_id    INT UNSIGNED PRIMARY KEY AUTO_INCREMENT
, salary       DOUBLE NOT NULL
, bonus        DOUBLE
, compensation DOUBLE AS (salary + bonus));

It’s virtually (excuse the pun) how it would work in Oracle Database 11g. I wonder what else we see move over into MySQL. Just a note, virtual columns moved from Oracle RDB (a prior acquisition from Digital Equipment Corporation in 1996) to the Oracle Database 11g.

Equivalent syntax in Oracle Database 11g would be like this for concatenation:

1
2
3
4
5

CREATE TABLE employee
( employee_id NUMBER
, first_name  VARCHAR2(20)
, last_name   VARCHAR2(20)
, full_name   VARCHAR2(41) AS (first_name || ' ' || last_name));

And like this for a math operation (like the one noted above for a future release of MySQL):

1
2
3
4
5

CREATE TABLE salary
( salary_id    NUMBER       CONSTRAINT pk_salary PRIMARY KEY
, salary       NUMBER(15,2) CONSTRAINT nn_salary_01 NOT NULL
, bonus        NUMBER(15,2)
, compensation NUMBER(15,2) AS (salary + bonus));

Finally, the Oracle 11g XE Beta has arrived. Actually, it appears I missed it when Justin announce it. Anyway, it is available for download here. It’s now complete and downloadable, the links are here.

A student encountered a connection problem with a PHP application that failed to resolve to the database. The steps to validate this are to check the PHP credentials, which are in this older post. Once you’ve done that, you should do:

1. Check the tnsnames.ora file contents, they should have a hostname value not an IP address. When you’ve not set your machine hostname in Microsoft Windows host file (mine in the example is: McLaughlin7x64), Oracle opts for the IP address.

XE =
  (DESCRIPTION =
    (ADDRESS = (PROTOCOL = TCP)(HOST = McLaughlin7x64)(PORT = 1521))
    (CONNECT_DATA =
      (SERVER = DEDICATED)
      (SERVICE_NAME = xe)
    )
  )

2. Check if the Oracle TNS (Transparent Network Substrate) validates with this command:

tnsping xe

3. If the prior step fails, check to see if you’re listener is running and that it’s configuration file looks more or less like this sample. If it doesn’t, you should edit this listener.ora file and start the listener process through the services dialog.

LISTENER =
  (DESCRIPTION_LIST =
    (DESCRIPTION =
      (ADDRESS = (PROTOCOL = IPC)(KEY = EXTPROC1521))
      (ADDRESS = (PROTOCOL = TCP)(HOST = McLaughlin7x64)(PORT = 1521))
    )
  )

4. You should then be able to connect like this:

sqlplus username/password@xe

Hope this helps a few people.

Writing an anonymous block to conditionally drop tables and sequences got very old. I figured it was time to simply put a procedure that would simplify it. Avoiding repetition was important too, so it supports an invoker rights model. This mimics the IF EXISTS syntax available in MySQL.

You need to create this as the SYSTEM user and then grant execute permission on the procedure to PUBLIC. Then, you need to create a couple public synonyms.

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

-- Create a procedure to simplify dropping objects.
CREATE OR REPLACE PROCEDURE drop_ifexists
( pv_type   VARCHAR2
, pv_table  VARCHAR2 ) AUTHID CURRENT_USER IS
 
  -- String for DDL command.
  sql_text  VARCHAR2(2000);
 
  -- Declare a parameterized cursor.
  CURSOR find_object
  ( cv_type   VARCHAR2
  , cv_table  VARCHAR2 ) IS
    SELECT   uo.object_name
    ,        uo.object_type
    FROM     user_objects uo
    WHERE    uo.object_name = UPPER(cv_table)
    AND      uo.object_type = UPPER(cv_type);
 
BEGIN
 
  -- Open the cursor with the input variables.
  FOR i IN find_object(pv_type, pv_table) LOOP
 
    -- Check for a table object and append cascade constraints.
    IF i.object_type = 'TABLE' THEN
      sql_text := 'DROP '||i.object_type||' '||i.object_name||' CASCADE CONSTRAINTS';
    ELSE
      sql_text := 'DROP '||i.object_type||' '||i.object_name;
    END IF;
 
    -- Run dynamic command.
    EXECUTE IMMEDIATE sql_text;
 
  END LOOP;
 
END drop_ifexists;
/

After creating the procedure in the SYSTEM user schema, you should run these Data Control Language (DCL) commands:

GRANT EXECUTE ON drop_ifexists TO PUBLIC;
CREATE PUBLIC SYNONYM drop_ifexists FOR system.drop_ifexists;

Assuming you have a table named MESSAGE, you can call the drop_ifexists procedure to conditionally delete it as follows below. The same procedure works with any object in a user’s schema.

EXECUTE drop_ifexists('table','message');

The only problem comes with types because they can have dependents. When they have dependents you can’t drop them until you’ve removed all the dependents. There are two types of dependents types and collection types.

Since discovering those may be difficult for new users, here’s another invoker rights function to discover user defined types. You should note that you’ll need to create the types before you try to compile the code (create the stored program in the Oracle database).

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

-- Create a SQL structure.
CREATE OR REPLACE TYPE dependent_type_obj IS OBJECT
( dependent_type  VARCHAR2(10)
, type_name       VARCHAR2(30));
/
 
-- Create a SQL collection of a user-defined data structure.
CREATE OR REPLACE TYPE dependent_type_tab IS TABLE OF dependent_type_obj;
/
 
-- Create a procedure to discover type dependents.
CREATE OR REPLACE FUNCTION type_dependents
( pv_type_name  VARCHAR2 ) RETURN dependent_type_tab AUTHID CURRENT_USER IS
 
  -- Declare a counter for the collection variable.
  c NUMBER := 1;
 
  -- Declare a return type variable.
  list DEPENDENT_TYPE_TAB := dependent_type_tab();
 
  -- Declare a parameterized cursor for dependent types.
  CURSOR find_dependent_types
  ( cv_type  VARCHAR2 ) IS
    SELECT   'STANDALONE' AS dependent_type
    ,        at.type_name
    FROM     all_types at INNER JOIN all_type_attrs ata
    ON       at.type_name = ata.type_name
    WHERE    at.owner = USER
    AND      ata.attr_type_name = UPPER(cv_type)
    UNION ALL
    SELECT   'COLLECTION' AS dependent_type
    ,        act.type_name
    FROM     all_types at INNER JOIN all_coll_types act
    ON       at.type_name = act.elem_type_name
    WHERE    at.owner = USER
    AND      act.elem_type_name = UPPER(cv_type); 
 
BEGIN
 
  -- Loop through all returns and add them to the return collection.
  FOR i IN find_dependent_types(pv_type_name) LOOP
    list.EXTEND;
    list(c) := dependent_type_obj(i.dependent_type, i.type_name);
	c := c + 1;
  END LOOP;
 
  -- Return the list.
  RETURN list;
END type_dependents;
/

You should also make the following grants and synonyms:

GRANT EXECUTE ON system.type_dependents TO PUBLIC;
CREATE PUBLIC SYNONYM type_dependents FOR system.type_dependents;
GRANT ALL ON system.dependent_type_obj TO PUBLIC;
GRANT ALL ON system.dependent_type_tab TO PUBLIC;

Then, you can call the type_dependents function to find any dependent user-defined data types. If so, you must drop them from the bottom or most specialized straight up the type tree to the root node data type.

SELECT *
FROM   TABLE(type_dependents('&type_name'));

Somebody raised the question about writing stored functions while we were discussing scalar subqueries against COMMON_LOOKUP tables. Common look up tables store collections of possible <OPTION> elements. They’re basically generalized tables that contain a set of smaller tables, where the row sets make up a list of unique values.

When you write a function that must return one and only one row, the SELECT-INTO syntax is ideal in Oracle because it automatically raises an exception when the query returns no row or two or more rows. MySQL doesn’t raise an automatic exception when a SELECT-INTO fails to return a row but it does raise an ERROR 1722 when two or more rows are found.

Here are sample implementations written in Oracle’s PL/SQL and MySQL’s SQL/PSM languages. If you’re an Oracle developer and new to MySQL, the biggest oddity may be the delimiter, or it may be the sizing of formal parameters. Hopefully, its not the lack of formal declaration and exception blocks in SQL/PSM. For those coming from MySQL, the OR REPLACE command or dynamic sizing of formal parameter list values are the big changes.

Oracle PL/SQL Function

The Oracle PL/SQL function is very straightforward and helped by pre-defined exceptions for both the no data found and too many rows found errors. Formal parameters inherit their physical size at run time from the calling scope program.

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

-- Create or replace existing function with new implementation.
CREATE OR REPLACE FUNCTION get_common_lookup_id
( pv_table   VARCHAR2
, pv_column  VARCHAR2
, pv_type    VARCHAR2 )
RETURN NUMBER IS
 
  -- Declare a return variable.
  lv_return  NUMBER;
 
BEGIN
 
  -- Query data and assign it to a local variable.
  SELECT   common_lookup_id
  INTO     lv_return
  FROM     common_lookup
  WHERE    common_lookup_table = pv_table
  AND      common_lookup_column = pv_column
  AND      common_lookup_type = pv_type;
 
  -- Return the value found.
  RETURN lv_return;
 
EXCEPTION
 
  -- Handle errors.
  WHEN NO_DATA_FOUND THEN
    RAISE;
  WHEN TOO_MANY_ROWS THEN
    RAISE;
 
END;
/

MySQL SQL/PSM Function

The lack of an implicitly raised exception when a SELECT-INTO structure fails to return a row adds several steps to a SQL/PSM function. You must define a custom exception and an if-block to raise the exception, which is trigger by returning a null value into the lv_return local variable.

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

/* Conditionally drop the function before trying to create it. */
DROP FUNCTION IF EXISTS get_common_lookup_id;
 
/* Change the delimiter to write a stored procedure into the database. */
DELIMITER $$
 
/* Create a function that replaces a scalar subquery. */
CREATE FUNCTION get_common_lookup_id
( pv_table   VARCHAR(30)
, pv_column  VARCHAR(30)
, pv_type    VARCHAR(30))
RETURNS INT UNSIGNED
BEGIN
 
  /* Declare a return variable. */
  DECLARE lv_return  INT UNSIGNED;
 
  /* Declare a local variable for a subsequent handler. */
  DECLARE no_data_found CONDITION FOR SQLSTATE '99001';
 
  /* Query data and assign it to a local variable. */
  SELECT   common_lookup_id
  INTO     lv_return
  FROM     common_lookup
  WHERE    common_lookup_table = pv_table
  AND      common_lookup_column = pv_column
  AND      common_lookup_type = pv_type;
 
  /* The local variable is only null when no rows are returned by the implicit cursor. */
  IF lv_return IS NULL THEN
    SIGNAL no_data_found SET MESSAGE_TEXT = 'Result was no rows found.';
  END IF;  
 
  /* Return the value found. */
  RETURN lv_return;
 
END;
$$
 
/* Reset the delimiter to the default to run other programs. */
DELIMITER ;

Hope this help solve your problems.

Somebody wanted another example of how to leverage a true/false condition from a PL/SQL stored function in PHP. The first key is that you write the function as if you were using it in SQL not PL/SQL. That means you return a NUMBER data type not a PL/SQL-only BOOLEAN data type.

Here’s the schema-level PL/SQL function:

CREATE OR REPLACE FUNCTION like_boolean
( a NUMBER, b NUMBER ) RETURN NUMBER IS
 
  /* Declare default false return value. */
  lv_return_value NUMBER := 0;
 
BEGIN
 
  /* Compare numbers and return true for a match. */
  IF a = b THEN
    lv_return_value := 1;
  END IF;
 
  /* Return value. */
  RETURN lv_return_value;
 
END;
/

Here’s the PHP that leverages the PL/SQL in an if-statement on line #24:

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

<?php
  // Capture local variables when provided.
  $thingOne = (isset($_GET['thingOne'])) ? $_GET['thingOne'] : 1;
  $thingTwo = (isset($_GET['thingTwo'])) ? $_GET['thingTwo'] : 1;
 
  // Open a connection.
  if(!$c = oci_connect("student","student","localhost/xe"))
  {
    die;
  }
  else
  {
 
    // Parse a statement.  
    $s = oci_parse($c,"BEGIN
                         :returnValue := LIKE_BOOLEAN(:thingOne,:thingTwo);
                       END;");
 
    // Bind input and output values to the statement.
    oci_bind_by_name($s,":returnValue",$returnValue);
    oci_bind_by_name($s,":thingOne",$thingOne);
    oci_bind_by_name($s,":thingTwo",$thingTwo);
 
    // Execute the statement.
    if (@oci_execute($s))
    {
      // Print lead in string.
      print "[".$thingOne."] and [".$thingTwo."] ";  
      if ($returnValue)
        print "are equal.<br />";
      else
        print "aren't equal.<br />";
    }
 
    // Clean up resources.
    oci_close($c);
  }
?>

If you run into a parsing error, which is infrequent now. You can wrap the multiple row PL/SQL anonymous block call with this function. It strips tabs and line returns. Alternatively, you can put all the lines of PL/SQL on a single line.

  // Strip special characters, like carriage or line returns and tabs.
  function strip_special_characters($str)
  {
    $out = "";
    for ($i = 0;$i < strlen($str);$i++)
      if ((ord($str[$i]) != 9) && (ord($str[$i]) != 10) &&
          (ord($str[$i]) != 13))
        $out .= $str[$i];
 
    // Return pre-parsed SQL statement.
    return $out;
  }

If you run into a parsing problem on Oracle XE 10g, you can wrap the PL/SQL call like the following. Alternatively, you can place the entire anonymous PL/SQL block on a single line without embedded tabs or return keys..

10
11
12
13

  $s = oci_parse($c,strip_special_characters(
                    "BEGIN
                       :returnValue := LIKE_BOOLEAN(:thingOne,:thingTwo);
                     END;"));

Hope that answers the question and helps some folks.

A few years ago I wrote a couple articles showing how to use PHP to connect to an Oracle Database 10g Express Edition instance. They’re still there on Oracle’s Technical Network but the source files are missing. It appears that Oracle may have migrated the articles to a new server but failed to migrate the source files.

Don’t forget that you’ll need to Install the Zend Community Server. Then, you need to create an IDMGMT1, IDMGMT2, and IDMGMT3 users and run the create_identity_db2.sql or create_identity_db3.sql seeding script for both database validation models. You’ll find links to the original articles, the source code in zip files. The newer version with CSS is the IDMGMT3 code.

The CSS sign-on form looks like this:

The portal page looks like this:

The add new user page looks like this:

The source files are as follows:

1. Database Authentication #1 – IdMgmt1.zip
2. Database Authentication #2 – IdMgmt2.zip
3. Database Authentication #3 – IdMgmt3.zip includes the CSS but works with the same write-up as Idmgmt2.

Hope this helps those who wanted the files.

We had a discussion today about how you can manage legacy code that you can’t change. For example, how do you access a PL/SQL function in PHP that returns a PL/SQL table of record structures? PL/SQL tables, index-by tables, or associate arrays are one in the same dependent on the release documentation. They’ve been available since Oracle 7.3 (roughly 15+ years).

You’ve a handful of solutions but I think the best is to wrap it in a Pipelined Table function (more on that in this older post). Here’s an example of such a package, wrapper function, and PHP program calling the wrapper function (command-line only PHP sample code).

Let’s say you have the following type of legacy package specification and body:

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

-- Create the package specification.
CREATE OR REPLACE PACKAGE lib IS
  /* Define a record structure. */
  TYPE movie_title_record IS RECORD
  ( title    VARCHAR2(60)
  , subtitle VARCHAR2(60));
 
  /* Define an associative array of a package record structure. */
  TYPE movie_title_table IS TABLE OF movie_title_record
  INDEX BY BINARY_INTEGER;
 
  /* Define a prototype of a package function. */
  FUNCTION get_movie
  ( pv_title VARCHAR2 ) RETURN lib.movie_title_table;
 
END lib;
/
 
-- Create the package body.
CREATE OR REPLACE PACKAGE BODY lib IS
 
  /* Implement the package function. */
  FUNCTION get_movie
  ( pv_title VARCHAR2 ) RETURN lib.movie_title_table IS
 
    /* Declare a counter variable. */
    lv_counter INTEGER := 1;
 
    /* Declare an instance of the package nested table and initialize it. */
    lv_table   LIB.MOVIE_TITLE_TABLE := lib.movie_title_table();
 
    /* Define a parameterized cursor to read values from the ITEM table. */  
    CURSOR c ( cv_partial_title VARCHAR2 ) IS
      SELECT   i.item_title
      ,        i.item_subtitle
      FROM     item i
      WHERE    i.item_title LIKE '%'||cv_partial_title||'%';
 
  BEGIN
 
    /* Read the contents of the parameterized cursor. */
    FOR i IN c (pv_title) LOOP
 
      /* Extend space, assign values from the cursor to the record structure
         of the nested table, and increment counter. */
      lv_table.EXTEND;
      lv_table(lv_counter) := i;
      lv_counter := lv_counter + 1;
 
    END LOOP;
 
    /* Return PL/SQL-scope nested table. */
    RETURN lv_table;
 
  END get_movie;
 
END lib;
/

You can wrap the lib package’s get_movie function with a schema-level function provided you convert the older associative array to a PL/SQL-scope nested table. You can do that in two steps. The first requires that you create a wrapper package specification, like the following example. The second step requires you to write a conversion wrapper function, shown later.

The table is dependent on the named record structure from the lib, and as such the packages are now tightly coupled. This is not uncommon when you can’t fix a vendors legacy code set.

1
2
3
4
5
6
7

CREATE OR REPLACE PACKAGE wlib IS
 
  /* Define a nested table of a package record structure. */
  TYPE movie_title_table IS TABLE OF lib.movie_title_record;
 
END wlib;
/

The wrapper function also converts the Oracle Database 7.3 forward data type to an Oracle Database 8.0.3 data type, and then pipes it into a SQL aggregate table. SQL aggregate tables are valid call parameters in the SQL-context. The TABLE function converts the collection of record structures into an inline view or derived table, as you’ll see a little farther along.

You should note that the return type of this function differs from the original package-level get_movie function. The former uses an associative array defined in the lib, while the latter uses a nested table defined in the wlib package.

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

CREATE OR REPLACE FUNCTION get_movie
( pv_title VARCHAR2 ) RETURN wlib.movie_title_table
PIPELINED IS
 
  /* Define a PL/SQL-scope associative array (Available Oracle 7.3). */
  lv_table_source LIB.MOVIE_TITLE_TABLE;
 
  /* Define a PL/SQL-scope nested table (Available Oracle 8.0.3). */
  lv_table_target WLIB.MOVIE_TITLE_TABLE := wlib.movie_title_table();
 
BEGIN
 
  /* Assign the results of a PL/SQL-scope call to the package function. */
  lv_table_source := lib.get_movie(dbms_assert.simple_sql_name(pv_title));
 
  /* Read the contents of the PL/SQL-scope nested table into a PIPE ROW. */    
  FOR i IN 1..lv_table_source.COUNT LOOP
    lv_table_target.EXTEND;
    lv_table_target(i) := lv_table_source(i);
    PIPE ROW(lv_table_target(i));
  END LOOP;
 
END;
/

You can test this exclusively in SQL*Plus with the following formatting and query. The TABLE function translates the returned array into an inline view or derived table for processing.

-- Format columns for display with SQL*Plus.
COLUMN title    FORMAT A20 HEADING "Movie Title"
COLUMN subtitle FORMAT A20 HEADING "Movie Subtilte"
 
-- Select the contents of the schema-level function in a SQL-context.
SELECT *
FROM   TABLE(get_movie('Star'));

If you’re using my sample code from the Oracle Database 11g PL/SQL Programming book, you should see:

Movie Title          Movie Subtilte
-------------------- --------------------
Star Wars I          Phantom Menace
Star Wars II         Attack of the Clones
Star Wars II         Attack of the Clones
Star Wars III        Revenge of the Sith

The following is a simple command-line PHP program that calls the wrapper function. It calls the wrapper function, which calls the lib.get_movie() function, and it converts the PL/SQL data type from an associative array (Oracle 7.3+ data type) to a nested table (Oracle 8.0.3+ data type). The nested table is defined in the wlib library, which supplements rather than replaces the original lib library.

The last thing that the wrapper function does is transform the associative array result into a nested table before placing it in the pipe (this process is known as a Pipelined Table function). Only nested table and varray data types may be piped into a SQL aggregate table. Then, the external programming language can manage the output as if it were a query.

Here’s the PHP program:

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

<?php
  // Connect to the database.
  if ($c = oci_connect("student","student","localhost/orcl"))
  {
    // Parsing a statement into a digestable SQL statement.
    $s = oci_parse($c,"SELECT * FROM TABLE(get_movie('Star'))");
 
  // Execute the parsed statement.
  oci_execute($s,OCI_DEFAULT);
 
  // Store control variable for the number of columns returned by the statement.
  $columns = oci_num_fields($s);
 
  // Find the number of columns, loop through them, and write their column name metadata.
  for ($i = 1; $i <= $columns; $i++) {
 
    // Print the column names, also known as field names.
    print oci_field_name($s,$i);
 
    // Define a variable.
    $line;
 
    /* Check whether a variable is declared and pad it.
     *   The numeric literal is for my convenience because the maximum size
     *   of possible returns is known. In a real situation, you'd use the 
     *   following str_pad() call:
     *
     *   str_pad($temp,oci_field_size($s,$i),"-") */
    if (!isset($line)) 
      $line .= str_pad($temp,15,"-");
    else
      $line .= " ".str_pad($temp,20,"-");
 
    /* One of the values requires a double tab to line up, otherwise this
       practice is unwise. */    
    if ($i < $columns)
      print "\t\t"; }
 
  // Print line return for the header and a line to mimic Oracle SQL*Plus output.
  print "\n";
  print $line .= "\n";
 
  // Process row-by-row data returned when data is returned.
  while (oci_fetch($s))
  {
    // Process column-by-column data returned for each row.
    for ($i = 1; $i <= $columns; $i++) {
      print oci_result($s,$i);
 
    if ($i < $columns) print "\t"; }
 
      // Print line return for the row of data returned.
    print "\n";
  }
 
  // Release resources.
  oci_close($c);
 
  // Explicitly free any resources.
  oci_free_statement($s);
  oci_free_cursor($c);
  }
?>

Assuming you call this callOracle.php, you can call it from the command-line with this syntax:

php callOracle.php

It prints, like it would in SQL*Plus:

TITLE           SUBTITLE
--------------- --------------------
Star Wars I     Phantom Menace
Star Wars II    Attack of the Clones
Star Wars II    Attack of the Clones
Star Wars III   Revenge of the Sith

Hope that helps those working with legacy Oracle code.

Sometimes I’m amazed at things that come up. A student wondered why the sequences were incrementing by two when they’re defined to increment by one. It turns out that they were using Oracle APEX to create SQL statements to build a table, constraints, and a auto-numbering sequence trigger. Before executing the code, they’d copy it into their re-runnable script that created their schema.

Here’s an example of code that was generated by APEX for a table:

-- Create the table.
CREATE TABLE onesy
( onesy_id NUMBER
, onesy_text VARCHAR2(20));
 
-- Add the primary key constraint.
ALTER TABLE onesy ADD CONSTRAINT onesy_seq PRIMARY KEY (onesy_id);
 
-- Add a database trigger.
CREATE OR REPLACE TRIGGER onesy_trg 
BEFORE INSERT ON onesy
FOR EACH ROW
BEGIN
  :NEW.onesy_id := onesy_s1.NEXTVAL;
END;
/

This works in APEX because it doesn’t create forms that call onesy_seq.NEXTVAL but they did create that logic in their forms. The INSERT statement would look like:

INSERT INTO onesy VALUES (onesy_seq.NEXTVAL, 'One');

Therefore, the INSERT statement incremented the trigger by one and the database trigger incremented it by one. The result is that sequences two-step, which isn’t effective or the desired behavior.

After I explained the two-step problem, they asked if they could only call the trigger when the primary key value was null. While they could do that like this:

1
2
3
4
5
6
7
8
9

-- Add a database trigger.
CREATE OR REPLACE TRIGGER onesy_trg 
BEFORE INSERT ON onesy
FOR EACH ROW
WHEN (NEW.one_id IS NULL)
BEGIN
  :NEW.onesy_id := onesy_s1.NEXTVAL;
END;
/

The problem is that this type of trigger doesn’t stop other possible problems. While it prevents two-stepping the sequence, it doesn’t prevent two other errors.

One possible error that isn’t managed in this scenario is the use of numeric literals beyond the next value of the sequence. It writes the row but eventually the sequence catches up to the higher value and a production insert would fail. It would raise the following exception.

INSERT INTO onesy (onesy_text) VALUES ('Eight')
*
ERROR at line 1:
ORA-00001: UNIQUE CONSTRAINT (STUDENT.PK_ONE) violated

Another possible error can occur when you use a bulk insert operation. Assuming you’re inserting 500 rows at a go, you query the maximum value of the onesy_id column and then create an array of 500 numbers. Then, you perform the bulk INSERT statement. The next call to the trigger would raise another ORA-00001 unique constraint error.

Yes, you could lock the table before you perform the bulk operation. After the bulk operation you would drop and recreate the sequence with a new value equal to the maximum value in the column, and unlock the table. This limits concurrency of operation. You could treat these bulk operations as off-line transactions (batch processing) and it would work nicely.

You could also implement a policy that no bulk operations provide generated column values that link to a sequence. Beyond it’s impracticality to manage, that type of restriction does limit the benefit of bulk operations.

The students wanted a solution. So, here’s my take on a trigger that prevents collision with values above the next sequence value. It assumes that bulk operations will be performed as batch processing where you can disable this trigger.

This trigger disallows numeric literals, logs any attempts to use them, and stops processing when an INSERT statement tries to use anything other than the .NEXTVAL of the sequence. It will only work in an Oracle Database 11g database because the context of using a sequence_name.CURRVAL in a comparison isn’t supported in prior releases. The onesy table is renamed the one table in the example.

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

-- Create a sequence for table ONE that starts with 1 and increments by 1.
CREATE SEQUENCE msg_s1;
 
CREATE OR REPLACE TRIGGER one_t1 
BEFORE INSERT ON one
FOR EACH ROW
DECLARE
 
  /* Define an autonomous transaction scope to the trigger. */
  PRAGMA AUTONOMOUS_TRANSACTION;
 
  /* Declare a local exception raised when a .CURRVAL pseudo column for a sequence
     is called before a .NEXTVAL for the same sequence in the same session. */
  no_sequence_in_scope EXCEPTION;
  PRAGMA EXCEPTION_INIT(no_sequence_in_scope,-08002);
 
BEGIN
 
  /* Check if surrogate key is provided and the sequence not out of transaction scope. */
  IF :NEW.one_id IS NOT NULL AND NOT :NEW.one_id = one_s1.CURRVAL THEN
 
    /* Write message when sequence value is a numeric literal not a sequence
       generated value but a one_s1.NEXTVAL was previously called in the session.
       Commit after write or information is lost because it throws an user-defined
       exception. */
    INSERT INTO msg VALUES (msg_s1.NEXTVAL,'ID value less or greater than .NEXTVAL ['||:NEW.one_id||']['||:NEW.one_text||'].');
    COMMIT;
 
    /* Stop processing by throwing exception. */
    RAISE_APPLICATION_ERROR(-20002,'ID provided by calling scope is not next sequence value ['||:NEW.one_id||']['||:NEW.one_text||'].');
 
  ELSIF :NEW.one_id = one_s1.CURRVAL THEN
 
    /* Do nothing, calling scope is correct with a one_s1.NEXTVAL sequence call. */
    NULL;
 
  ELSE
 
    /* Increment sequence and assign a value when one isn't provided, like a NULL value. */
    :NEW.one_id := one_s1.NEXTVAL;
 
  END IF;
 
EXCEPTION
 
  /* Handle a no sequence in scope error. */
  WHEN no_sequence_in_scope THEN
 
    /* Write and commit log message for error. */
    INSERT INTO msg VALUES (msg_s1.NEXTVAL,'ID provided by calling scope is invalid ['||:NEW.one_id||']['||:NEW.one_text||'].');
    COMMIT;
 
    /* Stop processing by throwing an exception. */
    RAISE_APPLICATION_ERROR(-20001,'Not a sequence generated value ['||:NEW.one_id||'].');
 
END;
/

Since anonymous transaction triggers are tricky, it’s important to note that the message writing requires two commits. One before raising the exception when the .CURRVAL is in session scope and another in the exception handler before raising the error. If you forget those COMMIT statements, this is a sample of the error stack:

INSERT INTO one VALUES (one_s1.nextval,'Six')
                        *
ERROR at line 1:
ORA-06519: active autonomous TRANSACTION detected AND rolled back
ORA-06512: at "STUDENT.ONE_T1", line 31
ORA-04088: error during execution OF TRIGGER 'STUDENT.ONE_T1'

The trigger raises the following type of exceptions for an offending INSERT statement. The first occurs when the sequence is valid in the session scope, like:

DECLARE
*
ERROR at line 1:
ORA-20001: NOT a SEQUENCE generated VALUE [1].
ORA-06512: at "STUDENT.ONE_T1", line 48
ORA-04088: error during execution OF TRIGGER 'STUDENT.ONE_T1'
ORA-06512: at line 15

The second occurs when the sequence isn’t valid in the session scope.

INSERT INTO one VALUES (401,'Nine')
            *
ERROR at line 1:
ORA-20002: ID provided BY calling scope IS NOT NEXT SEQUENCE VALUE [401][Nine].
ORA-06512: at "STUDENT.ONE_T1", line 24
ORA-04088: error during execution OF TRIGGER 'STUDENT.ONE_T1'

A value that’s below the current high-watermark of the sequence raises a unique constraint, like this:

INSERT INTO one VALUES (1,'Eight')
*
ERROR at line 1:
ORA-00001: UNIQUE CONSTRAINT (STUDENT.PK_ONE) violated

The following is a script with all the necessary code components to test the example.

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
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140

-- Conditionally drop tables and sequences.
BEGIN
  FOR i IN (SELECT object_name, object_type FROM user_objects WHERE object_name IN ('ONE','ONE_S1','MSG','MSG_S1')) LOOP
    IF i.object_name = 'TABLE' THEN
      EXECUTE IMMEDIATE 'DROP '||i.object_type||' '||i.object_name||' CASCADE CONSTRAINTS';  
    ELSE
      EXECUTE IMMEDIATE 'DROP '||i.object_type||' '||i.object_name;  
    END IF;
  END LOOP;
END;
/
 
-- Create a table ONE.
CREATE TABLE one (one_id NUMBER CONSTRAINT pk_one PRIMARY KEY, one_text VARCHAR2(20));
 
-- Create a sequence for table ONE that starts with 1 and increments by 1.
CREATE SEQUENCE one_s1;
 
-- Create a table MSG.
CREATE TABLE msg (msg_id NUMBER, msg_text VARCHAR2(80));
 
-- Create a sequence for table ONE that starts with 1 and increments by 1.
CREATE SEQUENCE msg_s1;
 
-- Create or replace trigger.
CREATE OR REPLACE TRIGGER one_t1 
BEFORE INSERT ON one
FOR EACH ROW
DECLARE
 
  /* Define an autonomous transaction scope to the trigger. */
  PRAGMA AUTONOMOUS_TRANSACTION;
 
  /* Declare a local exception raised when a .CURRVAL pseudo column for a sequence
     is called before a .NEXTVAL for the same sequence in the same session. */
  no_sequence_in_scope EXCEPTION;
  PRAGMA EXCEPTION_INIT(no_sequence_in_scope,-08002);
 
BEGIN
 
  /* Check if surrogate key is provided and the sequence not out of transaction scope. */
  IF :NEW.one_id IS NOT NULL AND NOT :NEW.one_id = one_s1.CURRVAL THEN
 
    /* Write message when sequence value is a numeric literal not a sequence
       generated value but a one_s1.NEXTVAL was previously called in the session.
       Commit after write or information is lost because it throws an user-defined
       exception. */
    INSERT INTO msg VALUES (msg_s1.NEXTVAL,'ID value less or greater than .NEXTVAL ['||:NEW.one_id||']['||:NEW.one_text||'].');
    COMMIT;
 
    /* Stop processing by throwing exception. */
    RAISE_APPLICATION_ERROR(-20002,'ID provided by calling scope is not next sequence value ['||:NEW.one_id||']['||:NEW.one_text||'].');
 
  ELSIF :NEW.one_id = one_s1.CURRVAL THEN
 
    /* Do nothing, calling scope is correct with a one_s1.NEXTVAL sequence call. */
    NULL;
 
  ELSE
 
    /* Increment sequence and assign a value when one isn't provided, like a NULL value. */
    :NEW.one_id := one_s1.NEXTVAL;
 
  END IF;
 
EXCEPTION
 
  /* Handle a no sequence in scope error. */
  WHEN no_sequence_in_scope THEN
 
    /* Write and commit log message for error. */
    INSERT INTO msg VALUES (msg_s1.NEXTVAL,'ID provided by calling scope is invalid ['||:NEW.one_id||']['||:NEW.one_text||'].');
    COMMIT;
 
    /* Stop processing by throwing an exception. */
    RAISE_APPLICATION_ERROR(-20001,'Not a sequence generated value ['||:NEW.one_id||'].');
 
END;
/
 
-- Define a schema collection of strings.
CREATE OR REPLACE TYPE one_index AS TABLE OF VARCHAR2(20);
/
 
-- Define a schema collection of strings.
CREATE OR REPLACE TYPE one_list AS TABLE OF VARCHAR2(20);
/
 
 
-- Anonymous block bulk testing program.
DECLARE
  -- Declare two scalar collection variables.
  lv_index ONE_INDEX := one_index();
  lv_list  ONE_LIST := one_list('One','Two','Three','Four','Five');
BEGIN
 
  -- Initialize the collection of the ONE_INDEX collection.
  lv_index.EXTEND(lv_list.COUNT);
  FOR i IN lv_list.FIRST..lv_list.LAST LOOP
    dbms_output.put_line('['||i||']['||lv_list(i)||']');
    lv_index(i) := i;
  END LOOP;
 
  -- Perform a trivial example of a bulk insert.  
  FORALL i IN lv_list.FIRST..lv_list.LAST
    INSERT INTO one VALUES (lv_index(i),lv_list(i));
 
END;
/
 
-- Testing INSERT statements until sequence value increments beyond bulk insert values.
INSERT INTO one VALUES (6,'Six');
INSERT INTO one VALUES (one_s1.nextval,'Seven');
INSERT INTO one VALUES (1,'Eight');
INSERT INTO one VALUES (401,'Nine');
INSERT INTO one VALUES (NULL,'Ten');
 
COLUMN msg_id   FORMAT 999 HEADING "Msg|ID"
COLUMN msg_text FORMAT A72 HEADING "Msg Text"
 
-- 
SELECT * FROM one;
SELECT * FROM msg;
 
-- Anonymous block bulk testing program.
DECLARE
  -- Declare two scalar collection variables.
  lv_list  ONE_LIST := one_list('Twenty-one','Twenty-two','Twenty-three','Twenty-four','Twenty-five');
BEGIN
 
  -- Perform a trivial example of a bulk insert.  
  FORALL i IN lv_list.FIRST..lv_list.LAST
    INSERT INTO one VALUES (NULL,lv_list(i));
 
END;
/
 
-- Query tables.
SELECT * FROM one;
SELECT * FROM msg;

If I’ve fat fingered any typing or made logical errors, please let me know.

The last day of Oracle Open World 2010. My focus today was on attending OracleDevelop and JavaOne.

The weather picked up today and it was a nice warm Indian Summer day. There were lots of tourists out too. The photo is taken of the building where you board the Powell Street cable car. There was a lot of walking with the way the events are dispersed among the Moscone South, Moscone West, Marriott Hotel, Westin Hotel, Hilton Hotel, W Hotel, and Hotel Nikko. I can’t quite remember how many times I walked back and forth across the 6 blocks between the Moscone centers and Union Square hotels. I can tell you that we went only twice to Mel’s Drive-in.

It was amazing to see how quickly the various conference expedition centers shutdown, packed up, and had their materials shipped out. We had to step over all the plywood that protected tiles and carpets to attend events.

Day 4 also brought a smaller audience for venues. I’m not quite sure if they left earlier or slept in because they were out too late last night attending the Wednesday night event on Treasure Island. The reduced number of attendees was great for those of us who remained. You can see how few attended the .NET/Oracle hands-on lab in the Hilton, which made finding a nice spot easy. You can find the .NET/Oracle hands-on tutorial materials on the Oracle Technical site.

If you opt to use the tutorials, you may benefit from these hints. You should be able to avoid some of the issues that I ran into when working through the open labs. First, you should expand the Microsoft Studio to full screen. Second, you should look for context pop-ups attached to a small arrow at the top right corner of grids, et cetera. Lastly, there are a few small mistakes that you’ll need to work through. Look at the errors as an opportunity to think and experiment and they’re great basic .NET/Oracle tutorials.

Oracle Open World 2010 is done. Time to review the keynotes for those things that I missed while listening to them, and consider the new role of Tuxedo in the life of Oracle’s product stack. It’s also time to download and play with the MySQL 5.5 candidate release; and it’s time to kick off my shoes, put up my feet, and play with the technology again.

Tomorrow I turn my fate over to the airlines, and hope to arrive home on schedule.