Future Direction

Core Philosophy

"The lightweight nature of Lua + Modern Syntax + Observable GC/Coroutine Runtime"

A safe, predictable scripting layer optimized for game engine embedding.

Performance Targets

We continuously measure against our lua_baseline (which represents standard Lua 5.4 performance) to ensure our VM architecture remains competitive for real-time applications.

Metric	lua_baseline	Current Zephyr	Target
arithmetic (ns/op)	3~5	~8	4~5
coroutine resume	200~400 ns	593 ns	300~400 ns
host resolve	—	641 ns	< 300 ns

Roadmap

Short-term

#	Task	Notes
1	AOT compilation	bytecode → native object file, LLVM or QBE backend
2	JIT (hot path)	opcode_count threshold → ORC JIT, excluding coroutines
3	LSP inlay hints	textDocument/inlayHint — variable type hints
4	Regression tests	Split zephyr_parser.cpp / zephyr_gc_impl.cpp for TU unit tests

Mid-term

#	Task	Notes
5	Package manager	package.toml fetch / lock / publish
6	Standard library expansion	std/net, std/async, std/test
7	Enhanced REPL	multiline, history, tab completion

Long-term

#	Task	Notes
8	Self-hosting compiler	Compile Zephyr with Zephyr
9	WASM target	Emscripten or wasm32-unknown-unknown

AOT/JIT Backend Comparison

Backend	JIT	AOT	C++ Integration	Notes
LLVM	✅ ORC	✅	C++ native	Heavy build (~50MB)
Cranelift	✅	✅	Rust FFI needed	Lightweight, Rust toolchain
QBE	❌	✅	C lib	Extremely lightweight (~200KB)
MIR/libmir	✅	❌	C API	Red Hat, Lightweight JIT only

Conclusion: Pure game scripting VMs like Gravity/Lua are highly practical even without JIT. We recommend adding it only when actual bottlenecks mandate it.

Non-Goals

• Borrow checking / lifetime annotations
• Async/await paradigm (replaced entirely by coroutines)
• Metaprogramming beyond Reflection / dynamic dispatch