Models from code or code as a model?

Many model-driven engineering workflows take the existing code of a system as an input. Some of these include validation and verification, software modernisation or knowledge extraction. Current approaches take the code and produce a standalone model, instead of treating the code itself as a model and navigating through it on demand. For very large codebases this can be quite expensive to produce, and the resulting model can be too large to suit the simplest monolithic file-based stores. In this work we propose reusing a model that is already incrementally built for us and supports fast lookups: the internal indices maintained by our integrated development environments for code analysis and refactoring. In particular, we show an Epsilon driver that exposes the Eclipse JDT indices of one or more Java projects as a model, and compare its performance and usability against MoDisco for validating Java code with regards to a partial UML model.