Hi Andrus;
Thanks for getting back to me on this one. Your comments do make sense
and I think that in order to maintain focus on the "persistence
framework" and maintain consistency, it may be best to remove
"readNestedProperty" on CDO.
cheers.
(Andrus)
Sorry for delayed reply.
The current set of accessors 'readProperty'/'readPropertyDirectly' exists for
the benefit of the framework. 'readNestedProperty' is in a different category -
it is a utility method and is never used by Cayenne internally (not counting
unit tests). FWIW, I'd be willing to deprecate 'CDO.readNestedProperty' and
just keep its static version in the utility class. So an inconsistency you
pointed to - 'readProperty' being limited to mapped properties, while
'readNestedProperty' being a mix of mapped and unmapped properties, to me is
purely a *naming* inconsistency.
With that in mind us not having the suggested extra methods in Cayenne
shouldn't prevent the customizations that you have in mind ... I think (?) You
can always create a custom common superclass with a custom path navigation
strategy, ignoring 'readNestedProperty' if it gets in the way (MyDataObject
extends CayenneDataObject), but I am not convinced Cayenne should be the place
for this code. It is a persistence framework, and staying focused on
persistence is important IMO.
We had a discussion in the context of 'cayenne-lifecyce' recently on what should be in
Cayenne core, and what is an "extension". I think here we are down to this
choice also.
Let me know if I am still missing the point of your suggestion.
(Andrew)
I've been playing around with trying to override some behaviours in the area of
"read property" on the data objects, but I think I have come to the conclusion
that the way it is rigged-up makes this a little difficult. I'd like to suggest a change
(which is probably more appropriate for 3.1 owing to it being a non-trivial change in
behaviour) and I would be interested in any feedback. Here is the text of the ticket...
---
There is an issue that a data object (DO) uses the "readProperty()" method in its accessors such as
"getStartTimestamp()"/"getArtist()" etc... The "readNestedProperty()" does some extra things like
using reflection to get at non-modelled accessors, but "readProperty()" does not. This is an inconsistency.
Simply adding the additional reflection to "readProperty()" is not a good idea
because in the case where an object is not yet related to the model, an infinite loop can
result. Particularly in the case where a data object is not yet added to an object
context, the logical condition around stopping this infinite loop is not able to be
identified.
My suggestion is to add a protected "readPropertyStored()" which will be used by the accessors such as
"getStartTimestamp()"/"getArtist()". This method will ~not~ use reflection, but "readProperty()" will do the
additional reflection if necessary. The "readPropertyStored()" will continue to invoke "readPropertyDirectly()".
In addition, the "extra reflection" from "readProperty()" would be serviced
through two additional methods on the data object;
readPropertyDerived(..)
For the case of the "readNestedProperty(..)" the use of reflection into an
object which is ~not~ a data object will be serviced through;
readNestedPropertyFromNonDataObject(..)
Together these changes will allow for consistenecy in the 'read property'
behaviour and will also allow third parties to more easily extend Cayenne's
'read property' behaviors in order to support more sophisticated 'read
property' behaviour.
This is a change in behavior and generated DAO classes would need to be
modified to use 'readPropertyStored()' instead of 'readProperty()'.
--
Andrew Lindesay
www.silvereye.co.nz
