The problem with XmlTransient and XmlAttribute is the definition of priorities: whether ignoral is to occur before joining getter/setter pair, renaming or afterwards. So exact resulting combination needs to be carefully defined – order matters. For example, one could see @XmlTransient as applied to remove the whole property; which is not what test is expecting. And with renaming it all gets rather muddy.
As to rewrite, I mean that existing JaxbAnnotationIntrospector has enough fundamental flaws that it must be significantly rewritten. For example it relies on JDK property descriptors which only look for public methods, which obviously won't work when annotations can indicate "publicness". Code has many other issues too but that's the obvious one from last changes I made.
On plus side, jackson native pieces also need to be rewritten, if for slightly different reasons – here the issue is that getters and setters are not linked as there is no logical concept of property; so this should be added, and from thereon things can both work better and implementation can be simplified.
Given this it should be possible to reduce code duplication a bit, as it may be possible to have more shared discovery code.
Of course if a patch can be found for this particular test I am all for it, even if we will have to do more drastic rewrite down the road.