The goal really is to avoid classpath dependencies to Lucene, so that Hibernate Search can be used without the application relying on Lucene directly at all. That would be useful for various integrators that need to provide guarantees of stable APIs, and cannot provide such guarantees for Lucene.

We would not need a common (backend-agnostic) list of available tokenizers, tokenfilters, charfilters, and parameters; it's perfectly fine if those are still backend-specific (so we wouldn't address , at least not yet).