This is about a very minor improvement but interesting research in the JVM internals.

It turns out that the code our annotation processor generates for all Loggers is - of course - quite a lot of code being generated as there are a couple of methods, constants and formatters for each possible error, warning and exceptions.

What we missed is that by having our non-core modules inherit their Log interface from the one in hibernate-search-engine the generated code is duplicating all definitions from the main module, rather than similarly generating a logger implementation which inheirts from the main one.

Beyond the reasons of JVM metadata, I also believe this "convenience" of inheriting from the same module is actually not a good idea for other reasons so I'll be splitting the logger of hibernate-search-engine into a parent contract meant to be shared across modules, vs one which is meant to be private for this unit of code.