SERIAL and BIGSERIAL data types

Informix®

Informix supports the SERIAL, BIGSERIAL data types to produce automatic integer sequences:
  • SERIAL can produce 32 bit integers (INTEGER)
  • BIGSERIAL can produced 64 bit integers (BIGINT)
  • SERIAL8 is a synonym for BIGSERIAL
Steps to use serials with Informix:
  1. Create the table with a column using SERIAL, or BIGSERIAL.
  2. To generate a new serial, no value or a zero value is specified in the INSERT statement:
    INSERT INTO tab1 ( c ) VALUES ( 'aa' )
    INSERT INTO tab1 ( k, c ) VALUES ( 0, 'aa' )
  3. After INSERT, the new value of a SERIAL column is provided in SQLCA.SQLERRD[2], while the new value of a BIGSERIAL value must be fetched with a SELECT dbinfo('bigserial') query.

Informix allows you to insert rows with a value different from zero for a serial column. Using an explicit value will automatically increment the internal serial counter, to avoid conflicts with future INSERT statements that are using a zero value:

CREATE TABLE tab ( k SERIAL); -- internal counter = 0
INSERT INTO tab VALUES ( 0 ); -- internal counter = 1
INSERT INTO tab VALUES ( 10 ); -- internal counter = 10
INSERT INTO tab VALUES ( 0 ); -- internal counter = 11
DELETE FROM tab; -- internal counter = 11
INSERT INTO tab VALUES ( 0 ); -- internal counter = 12

Microsoft™ SQL Server

Microsoft SQL Server IDENTITY columns:

  • When creating a table, the IDENTITY keyword must be specified after the column data type:
    CREATE TABLE tab1 ( k INTEGER IDENTITY, c NVARCHAR(10) )
  • You can specify a start value and an increment with "identity(start, incr)":
    CREATE TABLE tab1 ( k INTEGER IDENTITY(100,2), ...
  • A new number is automatically created when inserting a new row:
    INSERT INTO tab1 ( c ) VALUES ( 'aaa' )
  • To get the last generated number, Microsoft SQL SERVER provides the SCOPE_IDENTITY() function.

    The @@IDENTITY global T-SQL variable is not recommended, as it is scope-less.

  • To put a specific value into an IDENTITY column, the SET command must be used:
    SET IDENTITY_INSERT tab1 ON
    INSERT INTO tab1 ( k, c ) VALUES ( 100, 'aaa' )
    SET IDENTITY_INSERT tab1 OFF 

Informix SERIALs and SQL Server IDENTITY columns are quite similar; the main difference is that SQL Server does not allow you to use the zero value for the identity column when inserting a new row.

Starting with version 2012, Microsoft SQL Server supports sequences:
-- To create a sequence object:
CREATE SEQUENCE myseq START WITH 100 INCREMENT BY 1;

-- To get a new sequence value:
SELECT NEXT VALUE FOR myseq;

-- To find the current sequence value (last generated)
SELECT convert(bigint, current_value) FROM sys.sequences WHERE name = 'myseq';

-- To reset the sequence with a new start number:
ALTER SEQUENCE myseq START WITH 100;

Solution

Note: For best SQL portability when using different types of databases, consider using sequences as described in Solution 3: Use native SEQUENCE database objects.
To emulation Informix serials with SQL Server, you can use three different solutions:
  1. Native SQL Server IDENTITY columns (default).
  2. Insert triggers based on sequences (requires SQL Server 2012 and +).
  3. Insert triggers based on the SERIALREG table (for SQL Server prior to 2012).
The method used to emulate SERIAL types is defined by the ifxemul.datatype.serial.emulation FGLPROFILE parameter:
dbi.database.dbname.ifxemul.datatype.serial.emulation = {"native"|"trigseq"|"regtable"}
  1. native: uses IDENTITY columns.
  2. trigseq: uses insert triggers with sequences (CREATE SEQUENCE).
  3. regtable: uses insert triggers with the SERIALREG table.

The default emulation technique is "native".

The serial types emulation can be enabled or disabled with the folllowing FGLPROFILE entries:
dbi.database.dbname.ifxemul.datatype.serial = {true|false}
dbi.database.dbname.ifxemul.datatype.serial8 = {true|false}
dbi.database.dbname.ifxemul.datatype.bigserial = {true|false}

The native IDENTITY-based solution is faster, but does not allow explicit serial value specification in insert statements; the other solutions are slower but allow explicit serial value specification in INSERT statements.

Important: The trigger-based solutions are provided to simplify the conversion from Informix, but are slower as the solution uses IDENTITY columns. To get best performances, we strongly recommend that you use native IDENTITY columns instead of triggers.

Disabling automatic serial retrieval for SQLCA.SQLERRD[2]

SERIAL emulation can be totally disabled by setting the ifxemul.datatype.serial FGLPROFILE entry to false:
dbi.database.dbname.ifxemul.datatype.serial = false

For Informix compatibility, after an INSERT statement, the ODI drivers automatically execute another SQL query (or do a DB client API call when possible), to get the last generated serial, and fill the SQLCA.SQLERRD[2] register. This results in some overhead that can be avoided, if the SQLCA.SQLERRD[2] register is not used by the program.

When SERIAL emulation is required (to create temp tables with a serial column during program execution), and the SQLCA.SQLERRD[2] register does not need to be filled, (typically because you use your own method to retrieve the last generated serial), you can set the ifxemul.datatype.serial.sqlerrd2 FGLPROFILE entry to false. This will avoid the automatic retrieval of last serial value to fill SQLCA.SQLERRD[2]:

dbi.database.dbname.ifxemul.datatype.serial.sqlerrd2 = false

See also db_get_last_serial().

1. Using the native serial emulation

In order to use the "native" emulation, make sure that the following FGLPROFILE entry is not defined (it is the default), or is set the entry to "native":

dbi.database.dbname.ifxemul.datatype.serial.emulation = "native"

In database creation scripts, all SERIAL[(n)] data types must be converted by hand to INTEGER IDENTITY[(n,1)] data types, while BIGSERIAL[(n)] data types must be converted by hand to BIGINT IDENTITY[(n,1)] data types.

Tables created from the BDL programs can use the SERIAL data type. When a BDL program executes a CREATE [TEMP] TABLE with a SERIAL column, the database interface automatically converts the "SERIAL[(n)]" data type to "INTEGER IDENTITY[(n,1)]".

In BDL, the new generated SERIAL value is available from the SQLCA.SQLERRD[2] variable. This is supported by the database interface which performs a "SELECT SCOPE_IDENTITY()". However, SQLCA.SQLERRD[2] is defined as an INTEGER, it cannot hold values from BIGINT identity columns. If you are using BIGINT IDENTITY columns, you must retrieve the last generated serial with the SCOPE_IDENTITY() SQL function.

By default (see SET IDENTITY_INSERT), SQL Server does not allow you to specify the IDENTITY column in INSERT statements; You must convert all INSERT statements to remove the identity column from the list.

For example, the following statement:
INSERT INTO tab (col1,col2) VALUES (0, p_value)
must be converted to:
INSERT INTO tab (col2) VALUES (p_value)
Static SQL INSERT using records defined from the schema file (DEFINE rec LIKE tab.*) must also be reviewed:
INSERT INTO tab VALUES (rec.*) -- will use the serial column
must be converted to:
INSERT INTO tab VALUES rec.* -- without parentheses, serial column is removed

Since 2.10.06, SELECT * FROM table INTO TEMP with original table having an IDENTITY column is supported: The database driver converts the Informix SELECT INTO TEMP to the following sequence of statements:

SELECT selection-items INTO #table FROM ... WHERE 1=2
SET IDENTITY_ INSERT #table ON
INSERT INTO #table ( column-list ) SELECT original select clauses
SET IDENTITY_ INSERT #table OFF

See also temporary tables.

2. Using the trigseq serial emulation (SQL Server 2012 and +)

In order to use the serial emulation based on triggers and sequences, make sure that all database users creating tables in program have permissions to create/drop sequences and triggers.

Define the FGLPROFILE entry to enable "trigseq" serial emulation:
dbi.database.dbname.ifxemul.datatype.serial.emulation = "trigseq"

In database creation scripts, all SERIAL[(n)] data types must be converted to INTEGER data types, BIGSERIAL must be converted to BIGINT and you must create one trigger for each table. To know how to write those triggers, you can create a small Genero program that creates a table with a SERIAL column. Set the FGLSQLDEBUG environment variable and run the program. The debug output will show you the native trigger creation command using a sequence.

Tables created from the BDL programs can use the SERIAL data type. When a BDL program executes a CREATE [TEMP] TABLE with a SERIAL column, the database interface automatically converts the "SERIAL[(n)]" data type to "INTEGER" and creates the insert triggers. When using BIGSERIAL[(n)], the column is converted to a BIGINT.

Important:
  • SQL Server does not allow you to create triggers on temporary tables. Therefore, you cannot create temp tables with a SERIAL column when using this solution.
  • SELECT ... INTO TEMP statements using a table created with a SERIAL column do not automatically create the SERIAL triggers in the temporary table. The type of the column in the new table is INTEGER. Similarly, a BIGSERIAL column becomes BIGINT.
  • When a table is dropped, all associated triggers are also dropped.
  • INSERT statements using NULL for the SERIAL column will produce a new serial value, instead of using NULL:
    INSERT INTO tab ( col1, col2 ) VALUES ( NULL, 'data' )
    This behavior is mandatory in order to support INSERT statements which do not use the serial column:
    INSERT INTO tab (col2) VALUES ('data')

    Check if your application uses tables with a SERIAL column that can contain a NULL value.

3. Using the regtable serial emulation (SQL Server versions prior to 2012)

Note: This solution is supported for SQL Server versions prior to 2012, if your server is a SQL Server 2012 or +, consider using the "trigseq" emulation instead.

In order to use the serial emulation based on triggers and the SERIALREG table, make sure that all database users creating tables in program have permissions to create/drop triggers.

Then, prepare the database and create the SERIALREG table as follows:

CREATE TABLE serialreg (
    tablename VARCHAR(50) NOT NULL,
    lastserial BIGINT NOT NULL,
    PRIMARY KEY ( tablename )
)

The SERIALREG table and columns have to be created with lower case names, since the SQL SERVER database is created with case sensitive names, because triggers are using this table in lower case.

Define the FGLPROFILE entry to enable "regtable" serial emulation:
dbi.database.dbname.ifxemul.datatype.serial.emulation = "regtable"

In database creation scripts, all SERIAL[(n)] data types must be converted to INTEGER data types, BIGSERIAL must be converted to BIGINT and you must create one trigger for each table. To know how to write those triggers, you can create a small Genero program that creates a table with a SERIAL column. Set the FGLSQLDEBUG environment variable and run the program. The debug output will show you the native trigger creation command using the SERIALREG table.

Tables created from the BDL programs can use the SERIAL data type. When a BDL program executes a CREATE [TEMP] TABLE with a SERIAL column, the database interface automatically converts the "SERIAL[(n)]" data type to "INTEGER" and creates the insert triggers. When using BIGSERIAL[(n)], the column is converted to a BIGINT.

Important:
  • The serial production is based on the SERIALREG table which registers the last generated number for each table. If you delete rows of this table, sequences will restart at 1 and you will get unexpected data.
  • SQL Server does not allow you to create triggers on temporary tables. Therefore, you cannot create temp tables with a SERIAL column when using this solution.
  • SELECT ... INTO TEMP statements using a table created with a SERIAL column do not automatically create the SERIAL triggers in the temporary table. The type of the column in the new table is INTEGER. Similarly, a BIGSERIAL column becomes BIGINT.
  • When a table is dropped, all associated triggers are also dropped.
  • INSERT statements using NULL for the SERIAL column will produce a new serial value, instead of using NULL:
    INSERT INTO tab (col1,col2) VALUES ( NULL,'data')
    This behavior is mandatory in order to support INSERT statements which do not use the serial column:
    INSERT INTO tab (col2) VALUES ('data')

    Check if your application uses tables with a SERIAL column that can contain a NULL value.