Hi Michael,
On May 2, 2005, at 7:31 AM, Michael Watzek wrote:
Hi Craig,
please find my comments inline:
The ri11 implementation does not declare a FOStore specific PM implementation. Following up your proposal, we would add a new class FOStorePM extending PMImpl. We would move all newQueryXXX methods to that class and make PMImpl abstract (removing all all newQueryXXX methods).Hi Michael, Thanks for this analysis. On Apr 28, 2005, at 11:16 AM, Michael Watzek wrote:What if we move the method newQueryTree from PMFImpl to the FOStorePMF class?Hi,
Micheal Bouschen raised the issue to get rid of the compile time dependencies between ri11 runtime and ri11 query. The benefit would be:
1) Runtime and query could be separated into 2 different projects. 2) ri11 runtime could be executed with another query implementation. 3) ri11 query could be executed with another runtime implementation.
Below, I summerize what code changes have to be done to achieve issue 2).
There are only 2 dependencies from runtime implementation to query implementation:
- PersistenceManagerFactoryImpl calls Tree constructor - PersistenceManagerImpl calls QueryImpl constructors
Similarly, we can move the nine newQueryXXX methods from PMImpl to FOStorePM.
Right. When I wrote this, I had forgotten that there was no FOStorePM! But you have accurately reflected the idea.
We could still keep the QueryFactory approach and move the interface from Runtime to Query. As an advantage, FOStore would not call constructors of the query implementation. Instead, it would create a query factory (using "Class.newInstance") and call the newQueryXXX methods. Thus, FOStore would not have compile time dependencies to the query implementation but only to the factory interface. Does this make sense?
Yes. Go for it.
If we break this dependency, we could keep the dependency from Query to Runtime. And not implement part 3 as outlined below. I think part 3 is a lot of work and still doesn't remove the dependencies.Actually, it is not so much work. I already know all locations in the query implementation which have to be changed. The change is mostly to replace the PM instance by a RuntimeContext instance.
(The proposal below still requires the Query to know about StateManagerInternal, FieldManager, and StoreManager which I guess are still defined in Runtime.)
However, we'd still have dependencies to FieldManager and StoreManager. We would not have dependencies to StateManager. The query implementation needs a StateManager instance to invoke "provideField". The two calls "PM.findStateManager" and "SM.provideField" can both be implemented in "RuntimeContext.provideField". For this reason, RuntimeContext would not declare method "findStateManager" (in contrast to my original proposal) and thus, would not depend on StateManager.
Changes 2) and 3) will aproximately take a day. Please let me know which direction to go.
I expect the changes for 3) will be more difficult and take more time for review to make sure we get it right. So I'd like to defer this exercise for now.
So the new proposal is to break out ri11 into core20 (model plus utility): no dependenciesOk.
enhancer20: depends on coreOk.
runtime20: depends on coreOk.
query20: depends on runtime and coreOk.
fostore20: depends on core, runtime, and queryOk.
So the end result will be the five projects buildable and testable against the api11 project library. We will then have a task to migrate these projects to depend on api20 instead.
Thanks,
Craig
Thanks, MichaelCraig
The proposal is to introduce a new runtime interface QueryFactory which defines methods creating QueryTree instances and Query instances, e.g.
public interface QueryFactory { QueryTree createTree(); Query createQuery(...); }
This interface is implemented by query implementations. The ri11 query implementation calls the Tree and QueryImpl constructors.
Additionally, ri11 runtime defines a new PMFInternal property containing the class name of the QueryFactory implementation. The property may be specified by a Properties instance which is passed to JDOHelper.getPersistenceManagerFactory, or it may be set calling a setter on the PMFInternal instance. The PMFInternal instance looks up the implementation class and calls newInstance() in order to create a QueryFactory instance. The class object is kept in an instance variable. The Tree and QueryImpl constructor calls in PersistenceManagerFactoryImpl and PersistenceManagerImpl are replaced by corresponding calls on the QueryFactory instance.
Constraints on the QueryFactory implementation class: It must implement a public no argument constructor.
Below, I summerize what code changes have to be done to achieve issue 3).
The ri11 query implementation does not have references into runtime implementation packages, such as "org.apache.jdo.impl.pm" or "org.apache.jdo.impl.state". However, it has some references into runtime specific interfaces like PersistenceManagerInternal (PMI) and StateManagerInternal (SMI):
- PMI.findStateManager(...) - PMI.loadClass(...) - PMI.assertIsOpen() - PMI.getStoreManager() - PMI.getCurrentWrapper() - SMI.provideField(...)
Additionally, there is a dependency to the FieldManager interface: As query implementation calls method SMI.provideField(...) having a field manager parameter, it provides an implementation for that interface and passes an instance of that implementation.
The proposal is to define a new query interface RuntimeContext defining all of the methods above, e.g.
interface RuntimeContext
{
StateManager findStateManager(PersistenceCapable pc);
Class loadClass(String name, ClassLoader given) throws ClassNotFoundException;
void assertIsOpen();
StoreManager getStoreManager();
PersistenceManager getCurrentWrapper();
void provideField(int fieldNumber, FieldManager fieldManager, boolean identifying);
}
This interface is implemented by runtime implementations. A query instance has an instance of this interface which it gets through a parameter of query factory method "createQuery", e.g.
QueryFactory.createQuery(..., RuntimeContext context);
All calls of the methods above are replaced by corresponding method calls on RuntimeContext instance.
Drawbacks of this proposal: It does not eliminate the dependency to the FieldManager interface.
Advantages of this proposal: It gathers all runtime calls in an single instance implementing a small API, rather than having calls on 2 different instances each of which implements a big API (PMI defines 23 methods, SMI defines 33 meethods).
Regards, Michael -- ------------------------------------------------------------------- Michael Watzek [EMAIL PROTECTED] Engineering GmbH mailto:[EMAIL PROTECTED] Buelowstr. 66 Tel.: ++49/30/235 520 36 10783 Berlin - Germany Fax.: ++49/30/217 520 12 http://www.spree.de/ -------------------------------------------------------------------
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!
-- ------------------------------------------------------------------- Michael Watzek [EMAIL PROTECTED] Engineering GmbH mailto:[EMAIL PROTECTED] Buelowstr. 66 Tel.: ++49/30/235 520 36 10783 Berlin - Germany Fax.: ++49/30/217 520 12 http://www.spree.de/ -------------------------------------------------------------------
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
