Often it's found that we bump tika version while the setups using oak which are supposed to provide dependencies for tika do not bump dependencies version.
That situation often leads to hidden errors that get detected very late as by definition we are expected to not fail on a text extraction (at best log a DEBUG log in *BinaryTextExtractor.
With current maven tests we don't hit that situation as maven takes care of pulling in dependencies transitively. We do miss out on having an OSGi based test which can kind of server 2 purposes:
- notify us when we do bump tika version
- serve as a reference for versions of dependencies that can work