When parsing dependencies of plugins, the existing code is comparing numbers with text. That can never produce a positive result.
I'm unsure if Spark has any plugins that defines dependencies, so the impact of this bug could be neglible.