I've been looking at a few other ways to accomplish wrapping Spark, and none look good. I could generate Java functions, which seem to be inherently serialisable, but that defeats the point of using ABCL. I also considered an approach using Java to call into ABCL, but then I've got to ensure that the ABCL functions are distributed to each node, defeating the point of using Spark. @Alessio Stalla, your most recent github commit suggests that things might have moved along to the point that function and closure serialisation is working. Are things stable enough to give it a try? Not this week, but perhaps on Monday? I'm happy to debug anything that I turn up. There's a new project on the horizon and I'd like to use the Java compatibility angle to try and get Lisp into it. If I'm lucky, I may not even have to mention lisp, depending on how/if the contract terms work out.