This is an interesting architecture, can you provide some more details on how exactly that works? E.g. do you use a c++ library that interprets scheme?
yeah, I use S7 Scheme. It's a complete scheme interpreter in one C file, super easy to embed, quite similar to Clojure linguistically. Basically all "business" logic is in Scheme, all OS/UI interaction is in C or C++. For me, it's amazing, because I write music apps and I can reuse my model code across desktop, phone, in Max/MSP or PureData, in web assembly, everywhere. Being able to prototype in Max is awesome. My model layer only knows about music, the high-level user actions (ie "play note", not "click button"), and Scheme. It doesn't know anything about widgets, phones, etc. All boundaries are crossed with the C FFI layer (which is really nice in S7) forcing a strict ports-and-adapters approach. There absolutely is upfront cost, but the liberation of being totally decoupled from any framework or vendor in my model code is awesome. I chose S7 because I'm doing music, but you could do similar with Guile, Chicken, Gambit, Embedded Common Lisp, etc. And Scheme is so minimal that if I needed to switch Scheme implementations it would not be a big deal - I would only be rewriting the adapter layer.
Interesting, one of the most successful companies I've seen in my tech due diligence work was doing the same thing in high end scientific computing. They had written their own DSL, and a whole containing application layer in C that could run on any target platform, with GUI adapters for windows, unix, osx, etc. They were doing very well. And they could get amazing scientists to work for them super-productively because the DSL was designed around the scientists needs.
This sounds awesome, glad to hear it's working well for you.
One question that comes to mind for the business stuff is security. Basically, are you accepting Scheme code from untrusted users, or is it just a bundled part of the app?
The reason I'm asking is that that run-time flexibility can completely bite you in the ass if someone can submit code that does unacceptable, unanticipated things. For example, I had an engagement with a company to review their main application, that they host. The issue was that while the app was in Java, it accepted user templates in (IIRC) BeanShell. One quick System.exit() in a template as a proof of concept, and Tomcat came thundering down.
The concept still works of course if you restrict the language accepted to some minimal DSL.
Interesting. This actually sounds very similar to game developers embedding Lua or V8 for various logic. I've used Lua myself in a similar way, embedded in a high perf application, with a user script driving the logic.
