Datastore identity strategy

[Craig L Russell <Cr...@Sun.COM>]

Javadogs,

I took an action item to resolve whether the TCK needs different  
metadata files to describe the different strategies that databases  
use for generating primary key values.

The spec says,

<spec>
•The value “sequence” specifies that a named database sequence is  
used to
generate key values for the table. If sequence is used, then the  
sequence
attribute is required.
•The value “identity” specifies that the column identified as the key  
column is
managed by the database as an identity type.
</spec>

In the TCK, we only use strategy="identity" and the issue is whether  
databases that don't support the SQL 2003 construction "GENERATED  
ALWAYS AS IDENTITY" can use this same strategy.

My reading of the specification is that the identity strategy can be  
used for any database as long as the database is configured to  
generate the keys. Such constructions as PostgreSQL SERIAL and Oracle  
sequences can be used, meaning that the SQL for the table declares  
that the database, not the JDO implementation, is responsible for  
generating the keys.

The effect of this interpretation is that the TCK is valid for all  
databases that are capable of generating the primary key values  
(which I think is all databases of interest to us).

Therefore, there is no need to change the TCK to accommodate  
databases that don't support the "GENERATED ALWAYS AS IDENTITY". A  
suitable SQL table definition can be used for all databases to  
implement the identity strategy.

I propose that we clarify this with a change to the specification,

<proposed>
•The value “sequence” specifies that a named database sequence is  
used to generate key values for the table. If sequence is used, then  
the sequence attribute is required. The JDO implementation uses the  
named sequence in insert statements for the table.

•The value “identity” specifies that the column identified as the key  
column is managed by the database as an identity type, regardless of  
how the database generates the values, e.g. using the SQL 2003  
IDENTITY column type or using triggers and stored procedures with a  
sequence. The JDO implementation does not provide primary key values  
for the table.
</proposed>

I'm also trying to figure out if we need the following:

<spec>
•The value “autoassign” specifies that the column identified as the  
key column
is managed by the database to automatically increment key values.
</spec>

Would anyone notice if we removed this?

Craig

Craig Russell
Architect, Sun Java Enterprise System http://java.sun.com/products/jdo
408 276-5638 mailto:Craig.Russell@sun.com
P.S. A good JDO? O, Gasp!