Refactor validation of PluginDescriptors

[janhoy]

Was duplicated in each finder, now pluggable.
Did this through an AbstractPluginDescriptorFinder so everyone get the default behavior ootb.
To plug in your own validator you'd need to Override createPluginDescriptorFinder() in your PluginManager.

[coveralls]

Coverage increased (+0.3%) to 15.475% when pulling 014cec2 on cominvent:validator into e9126c4 on decebals:master.

[coveralls]

Coverage increased (+0.3%) to 15.475% when pulling 014cec2 on cominvent:validator into e9126c4 on decebals:master.

[coveralls]

Coverage increased (+0.3%) to 15.475% when pulling 014cec2 on cominvent:validator into e9126c4 on decebals:master.

[decebals]

My idea is to move the validation process from PluginDescriptorFinder to PluginManager, after we found a PluginDescriptor (https://github.com/decebals/pf4j/blob/master/pf4j/src/main/java/ro/fortsoft/pf4j/AbstractPluginManager.java#L753).
The simple solution is to create a protected method validatePluginDescriptor in AbstractPluginManager and to call this method for each PluginDescripotor instance (no more PluginDescriptorValidator).
What you say?

[coveralls]

Coverage decreased (-0.5%) to 17.397% when pulling 5d3ea84 on cominvent:validator into c6aa1f1 on decebals:master.

[janhoy]

Agree that this was a more elegant location for the validation. I also had to fix some of the validation tests that previously failed by explicitly calling validation from the test code.
Alternatively we could have created a separate test class for validation of incomplete descriptors.

[decebals]

I propose you to move the code from PluginDescriptorUtils in AbstractPluginManager and to delete:

• PluginDescriptorUtils class
• code that tests validity of PluginDescriptor from *PluginDescriptorFinderTest.java (revert the modifications)

You can add a testValidatePluginDescriptor in AbstractPluginManagerTest. In the future if we see that validatePluginDescriptor method is a candidate for an extension point (many overwrites for this method) we can create an interface PluginDescriptorValidator with a default implementation (DefaultPluginDescriptorValidator) but in this moment I think that it's not necessary to do this (we complicate the library and increase artificially the learning curve ).

Is it OK for you?

[janhoy]

Think I've accommodated most of it now. Found no AbstractPluginManagerTest so I created a DefaultPluginManagerTest

[coveralls]

Coverage increased (+5.8%) to 23.66% when pulling 391d101 on cominvent:validator into c6aa1f1 on decebals:master.

[decebals]

Thanks! Good work.

[janhoy]

Good solution, easy to plug own validation. In my scenario I accept a descriptor without a PluginClass since Solr already has another way of handling that.