This is dynamically typed, but staged. A type error is something that doesn't doesn't survive compile time evaluation and or doesn't specialize to state machine form. I'm thinking that might be good enough as guard rail.
Conversation
Or.. add an actual type system. Playing with this is reshaping the way I think about managing my future "C life". I cannot escape C nor do I want to, but I can write a whole lot more in a Schemish language + ad hoc compile time checks than I originally thought was possible.
1
Doodled a bit to combine new SM compiler with the new interpreter for partial evaluation, but seem to have lost my stride after 2 weeks of putting out fires at work...
1
Two more weeks of exhaustion.. But picked it up again. Lure Scheme: added preliminary Erlang, JavaScript and GDB script backends, and started working on dataflow analysis (liveness) for the state machine compiler to implement state variable allocation across suspend points.
1
Still existential questions about what this thing is supposed to be. For now it's self-propelling. Something something Scheme, microcontrollers and test systems.
1
In related news, I contributed a Lua backend to Ribbit VM. I'd like to integrate that somehow. There is some overlap, though my main aim is compilation.
1
Anyways I think I know what I want: rebuild the ideas of Staapl (interactive tethered development), but build it on top of standard tools (GDB & C) using a lowest common denominator language (Scheme or Scheme subset) as a focal point.
1
Actually surprised I got compilation from Scheme to GDB script working with so little effort. It can now do functions, if/then/else and while loops (from named let or single-def letrec, with support for nesting).
1
1
Managed to encode anonymous functions - see gist. Unfortunately due to global variables only, there is no re-entrancy. E.g. nested 'map' won't work. Can't think of a simple solution, so will probably resort to inlining.
1
Possible solutions: use nested data structures on target, but this requires target memory. GDB can host arrays so could implement memory which could host a low level VM + GC, but that seems a bit far-fetched.
Replying to
This needed a break apparently... I know where to go next: build something with RTIC first, then see how to mesh the state machine compiler with something that maps to RTIC or some stripped down C version of the RTIC idea (use Cortex M NVIC as a scheduler).
2