Hi Robin,

On Jul 20, 2006, at 6:42 AM, Robin M. Roos wrote:

Hi Eric et. al.

If you look at the spec section 18.10 the "implements" element is
used to notate that a (potentially concrete) persistent class
implements a persistent interface.  That means I can navigate to
or query the Interface type and arrive at the persistent instance.
 This is definitely a concrete developer-provided instance.

Not necessarily. It could be a developer- or implementation-provided instance.

Then from Section 18.21 the "implements" element contains
"property" elements which are used to map persistent fields of the
concrete class to the, if I may use the word here, "abstract"
properties of the interface.

[I actually believe the "property" element in this context, within
"implements" should be a simpler XML element which maps the name
from the interface to a fully defined property in this
persistence-capable, but the DTD reuses the same property element
here.  This potentially stops one class from implementing two
interfaces and mapping the "description" property of each of them
to the same persistent field.  It would have to define the field
twice if "description" was an (abstract) property in both
interfaces.]

Java has this same restriction. A class that implements multiple interfaces must have identical implementations for methods from the multiple interfaces that have the same signature. There's no way to disambiguate the methods of one interface from the others. Do other languages allow different behavior?

The big concern I have is the requirement for the interface to
provide methods at all.  Properties can be of many types, and we
could insist on get/set (default) is/set (boolean)
add/remove/iterator (List/Collection/Set) put/get (Map) etc., but
it is more flexible and not more dangerous for there to be no
requirement for any specific method convention.  The developer
will put appropriate methods on the interface to suit his object
model, but JDO does not need to know what they are in order to
satisfy its obligations to the object model.

That's true, and there is no restriction placed on non-persistent methods of persistent interfaces. Just don't declare them or use them in queries. But if persistent properties are used in queries, the implementation needs to know how to map them to the underlying query language. Since we don't know what add/remove/iterator are supposed to do, they cannot be used in queries.

The extreme case is a marker interface called Product.  You could
navigate an association to a Product.

Let's get specific about how to map the association. Any suggestions?

You could iterate the Extent
of Product.  You could have JDOQL queries which navigate to
Product and through its (abstract) properties as required.  But if
the Product interface has no methods

As mentioned above, you can have business methods that are not the concern of the persistence layer but are only in the domain model. In this case, it doesn't make sense for the JDO implementation to create an implementation of the persistent interface (you knew that, but I thought it worth mentioning). That's why we also support persistent abstract classes, in which the business behavior can be implemented solely using persistent properties.

you must cast the result to a
concrete class before you can really use the (concrete
implementing type) Product instance you get back.  This is an
extreme example, but I just want to illustrate that JDO should not
force a requirement on the object model which is not necessary for
persistence purposes.

Agree.

Naturally this perspective is different to Craig's use-case of
having the JDO 2.0 implementation perform the function of
"persistent dynamic proxy" factory.  I had thought we'd
successfully merged these two use-cases.

Me too.

Craig

Kind regards, Robin.



On Thu Jul 20 06:07:05 PDT 2006, Erik Bengtson <[EMAIL PROTECTED]>
wrote:

Robin,

I understand that you want to be able to allow user providing
their own concrete
classes. I'm not sure if this is spec compliant, but JPOX for
instance supports
user and JDO vendor concrete implementations. That means users
are not obliged
to invoke pm.newInstance() if they already have their own class,
however JPOX
requires that the user concrete class to implement
PersistenceCapable.

Regards,


Quoting "Robin M. Roos" <[EMAIL PROTECTED]>:

Hello All

Firstly please note my email address is [EMAIL PROTECTED] and I
no
longer see traffic on the JDO 2.0 Expert Group mailing list.

I was discussing JDO 2.0 with a colleague who wants to represent
collection-based relationships between persistent interfaces.  He
was confused by the spec's requirement for get/set is/set methods
to support persistent properties, and I was confused too until I
vuaguely rememberd Craig and me taking opposing sides on what a
persistent interface actually represents.

My view was that a persistent interface represented a collection
of persistent fields which were provided by all
persistence-capable implementing classes, with field names in the
interface mapped to field names in each persistent class.  This
scheme allowed JDOQL to navigate through interface types, without
requiring any specific method signatures in the interface (it
might be a marker interface) and without requiring identical
field
names in each implementing class.

I understood Craig's view was that a persistent interface was
something which could be instantiated on demand by the JDO
implementation without a developer-provided concrete
implementation class.

I thought we compromised on a best-of-both-worlds basis.  But I
now see that persistent interfaces must define get/set is/set
methods.  How then do we accommodate a collection relationship
without exposing the Collection instance and breaking
encapsulation?  I might want an interface with methods
addEmployee(), removeEmployee(), and employees():Iterator.  In
JDOQL I would like to be able to use "contains" semantics against
any persistent class which implements this persistent interface.
If we could do this, then we could type more of the domain model
using interfaces instead of abstract classes.  The application
domain is financial trading, but I would have thought this
capability would be streets ahead of anything planned by the JPA.

I'm appending a brief excerpt from the JDO 2.0 spec regarding the
interface element for easy reference.

Please confirm whether or not what I want to achieve is supported
by JDO 2.0 (I now believe it is not supported).  Additionally are
there any plans for a maintenance revision which might add
support
for such a capability?  Personally I would like to see the
newInstance() capability deprecated for persistent interfaces,
but
that's just me and I could never see its value.  In reality we
would have to find a way of supporting both.

I hope you are all well.

Kind regards, Robin.




JDO 2.0, February 28, 2006 p217:

ELEMENT interface
The interface element declares a persistence-capable interface.
Instances of a vendorspecific type that implement this interface
can be created using the newInstance(Class persistenceCapable)
method in PersistenceManager, and these instances may be made
persistent.

The JDO implementation must maintain an extent for persistent
instances of persistencecapable classes that implement this
interface.

The requires-extent attribute is optional. If set to “false”, the
JDO implementation does not need to support extents of
factory-made persistent instances. It defaults to “true”.

The attribute name is required, and is the name of the interface.
The attribute table is optional, and is the name of the table to
be used to store persistent instances of this interface.

The detachable attribute specifies whether persistent instances
of
this interface can be detached from the persistence context and
later attached to the same or a different persistence context.
The
default is false.

Persistent fields declared in the interface are defined as those
that have both a get and a set method or both an is and a set
method, named according to the JavaBeans naming conventions, and
of a type supported as a persistent type.

The implementing class will provide a suitable implementation for
all property access methods and will throw JDOUserException for
all other methods of the interface.

This element might contain property elements to specify the
mapping to relational columns.

Interface inheritance is supported.







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!

Attachment: smime.p7s
Description: S/MIME cryptographic signature

Reply via email to