ValueHolder vs. Proxy Indirection

Question

Hi all,
We have a new project and we must decide what kind of indirection we'll use. I analyzed both indirection types, and noticed that ValueHolder is simpler but kind of 'pollutes' the code, OTOH proxy indirection is transparent but it needs to have interfaces for each indirection class. I don't know if we choose the elegant and transparent proxy indirection, or the simplicity of ValueHolder indirection.
What do you use? Why not use Proxy? Why not use ValueHolder?
Should I mix ValueHolder and Proxy Indirection?
Thanks,
Henrique Viecili

Answer 1

Henrique,
Either option is functional and both illustrate the challenges of transparent 1:1 indirection in Java. In order to truly make these lazy load relationships transparent you must do one of the following:
1 - Introduce a proxy class (value holder)
2 - Use Java's dynamic proxy
3 - use byte-code weaving/enhancement (ASM, BCEL, CGLIB, ...)
Going forward, starting with TopLink JPA, we are offering dynamic byte code weaving. This involves the classes having the proxy introduced when the persistent class is loaded. This is the best way as it does not require any additional proxy class or interfaces in your model and does not add any additional compile/build steps to development.
In the current releases of Oracle TopLink I would recommend using the ValueHolder approach for 1:1 as a private field and then wrap it in get/set methods that provide access to only the real related object type. Using this pattern you can more easily move to our dynamic weaving solution without the ValueHolder with the least amount of change.
Proxy indirection does have some performance impact.
Doug