Re: SCO and FCO treatment of String, Locale, etc.

[Craig L Russell]

Hi Ilan,

On Jul 24, 2006, at 12:10 AM, Ilan Kirsh wrote:

Hi Craig,

Probably this discussion is mainly theoretical because most JDO implementations do not support system immutable types as FCO, and even in implementations that do support this, the default is to use SCO, and this default is rarely changed (if at all). Maybe the only application that tries to use this optional feature is the JDOTCK, in which some immutable type fields are defined with embedded=false in the metadata. In my opinion the suggested fix in [1], at least for the fields that are defined with embedded=false is the right solution.

I had not understood that the tck tests in question defined embedded=false. Changing this to "default" to allow the implementation to choose would then be better. I'll take another look at the test cases.

Unfortunately the suggested fix to the specification might be insufficient because it will cause conflicts in many other places. For instance, what should return JDOHelper.getPersistenceManager (field) when a FCO immutable type field is shared by two different PersistenceManagers?

This is not portable behavior in any case. The specification could define this to return null always for these cases. These instances should not be associated with a PersistenceManager.

What should happen when such a field value is passed as a query argument (according to the spec: "If a persistent instance associated with another PersistenceManager is passed as a parameter, JDOUserException is thrown during execute()"), etc.

Since there is no ambiguity as to the semantics of passing such a reference (there is no PersistenceManager that owns the instances) no exception needs to be thrown.

The most logical solution IMO is to treat such FCO instances as any other FCO, and a user that chooses to use them with embedded=false (again, very rare) will have to use clone.

Thanks,

Craig

It might be better to treat such FCO instances as special cases that are not owned by a PersistenceManager, but are shared among all PersistenceManagers. This treatment is consistent with what at least one object database does (having a special database representation for fixed precision integers etc.)

Regards,

Ilan

  ----- Original Message -----
  From: Craig L Russell
  To: JDO Expert Group ; Apache JDO project
  Sent: Monday, July 24, 2006 7:01 AM
  Subject: SCO and FCO treatment of String, Locale, etc.


  Javadogs,

















  <spec 6.4.3>
  Immutable Object Class types

  classes, and may choose to support these instances as SCOs or FCOs:
  •package java.lang: Boolean, Character, Byte, Short, Integer, Long,
  Float, Double, and String;
  •package java.util: Locale, Currency.
  •package java.math: BigDecimal, BigInteger.

  ed as SCOs or FCOs.
  </spec 6.4.3>


  <proposed 6.4.3>
  Immutable Object Class types

  classes, and may choose to support these instances as SCOs or FCOs:
  •package java.lang: Boolean, Character, Byte, Short, Integer, Long,
  Float, Double, and String;
  •package java.util: Locale, Currency.
  •package java.math: BigDecimal, BigInteger.

  </proposed 6.4.3>


  [1]http://issues.apache.org/jira/browse/JDO-397?page=all


  Craig Russell

An issue has been raised [1] with regard to storage of instances of immutable system object classes, e.g. Integer, String, Locale. The specification calls for users of the API to not depend on whether these instances are stored as FCO or SCO. The intent of the specification is that applications should not depend on whether the instances are stored by themselves as persistent instances or stored as embedded within the domain class instances. Furthermore, the application should not depend on FCO behavior: uniquing of instances in the same PersistenceManager for the same instance. Finally, the application should not depend on guaranteeing SCO behavior: copying the instance upon commit and guaranteeing a distinct instance in memory upon reinstantiation. The rationale of this wording was specifically to allow an implementation of JDO to either store instances embedded in the "containing" domain class instance or as a nullable reference to a persistence-capable System class in the datastore, as is done by some object databases. The intent was not to impose restrictions on the application's use of the identical instance as the value of multiple fields of domain classes. A literal interpretation of this part of the specification would require users to guarantee that the same instance of any of the classes was never contained in instances managed by multiple PersistenceManagers. The effect of this interpretation is to require cloning for each instance of Integer, String, Locale, etc. used in domain classes. I believe that this interpretation should be disallowed by the specification, as it imposes a great burden on users. I would like the Expert Group to discuss this issue and straw proposal. JDO implementations must support fields that reference instances of immutable object Portable JDO applications must not depend on whether instances of these classes are treat- JDO implementations must support fields that reference instances of immutable object Portable JDO applications must not depend on SCO or FCO uniquing behavior, nor on the storage mechanism in the datastore. Portable applications may use the same instance of these classes as field values in any persistence-capable class instance. Architect, Sun Java Enterprise System http://java.sun.com/ products/jdo

  408 276-5638 mailto:[EMAIL PROTECTED]
  P.S. A good JDO? O, Gasp!

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

smime.p7s
Description: S/MIME cryptographic signature