Compiling Stellogen

[engboris]

What Stellogen does is mainly scheduling basic term interactions in some order/way. We could imagine a virtual machine (BIM = Basic Interaction Machine) with elementary term unification and compile Stellogen programs to it.

Could be interesting to look at the Warren Abstract Machine (WAM).

Don't hesitate to share your opinion.

[engboris]

Another possibility it to compile to Wasm because it's the new trend

[bmorphism]

I've been exploring what a Zig-based BIM could look like, motivated by bridging Stellogen to zig-syrup (OCapN transport + propagator networks).

Why Zig?

• Compiles to both native and Wasm from same source (addresses your Wasm comment)
• Zero-cost C FFI (for embedding in other systems)
• No GC, predictable memory (arena per execution)
• Comptime codegen for instruction dispatch tables

Sketch: ~20 opcodes

PUSH_VAR idx       ; variable onto stack
PUSH_TERM arity    ; construct term from stack
UNIFY              ; core operation: unify top two, push substitution or FAIL
FUSE star_a star_b ; ray unification between stars (Stellogen's fusion)
EXTERN fn_id arity ; FFI escape to registered C/Zig function

The key insight: BIM is simpler than WAM because Stellogen is deterministic (no backtracking/choice points). The scheduler just needs to pick the next fusion candidate.

FFI Bridge Example

// Register pure functions from zig-syrup
vm.register("hsl_to_rgb", hslToRgbBridge);
vm.register("golden_spiral", goldenSpiralBridge);

(extern hsl_to_rgb 3)
(constellation color_from_bci [phi valence fisher]
  (hsl_to_rgb (* phi 137.5) (sigmoid valence) (sigmoid fisher)))

Concrete Offer

I can contribute:

1. BIM spec as docs/BIM.md (~20 opcodes, memory model, scheduler modes)
2. Zig VM core (~800 LOC, arena allocator, unification, work queue)
3. OCaml emitter (emit_bim.ml) once the instruction set is agreed

Questions for you:

• Binary bytecode or text-based (for debugging)?
• Scheduler modes: OS threads or cooperative fibers for parallel?
• Wasm first (playground) or native first (FFI)?

Happy to discuss further or start with a small PR!

[engboris]

@bmorphism Thank you for your interest. The original idea was that when computation happens, we sometimes would like to know what step were done. Think of an automaton running. It would be nice to see each state transition. In some sense, Stellogen is just orchestrating some basic term unification steps so I thought of a sort of intermediate language and all Stellogen expressions would be translated into such series of term unification steps.

I'm curious about what is done for other similar languages like miniKanren.

I don't have doubts about the relevance of Zig but I'm actually not sure what a compiler would actually bring beyond some debugging features and such. At this point, I'm mostly interested in prototyping and experiments.

Since this is about prototype and not performance or real application, I would say a text-based bytecode would be relevant. I have no opinion for the other questions.