Performance

Current Implementation

• Value type: enum-based (not yet NaN-boxed)
• String interning: StringPool with Symbol(u32) keys
• Env: Vec<Binding> stack with Symbol keys (shared by interpreter + VM)
• Bytecode VM: stack-based, ~75 opcodes, hybrid mode (falls back to tree-walk)
• Async bytecode runtime: pollable continuation VM behind async-runtime; host futures, timers, and channels park on Tokio instead of blocking an OS thread
• Stack-slot locals: compiler resolves locals to slot indices, VM uses O(1) indexed access
• VM fn compilation caching: IonFn has fn_id: u64, VM has HashMap<u64, Chunk> cache
• Precompiled fn chunks: compile_program() returns (Chunk, FnChunkCache)

VM Optimizations

• Peephole optimizer: dead instruction removal with jump fixup
• Constant folding: literal BinOp/UnaryOp at compile time (int, float, string, bool)
• Dead code elimination: skip unreachable statements after return/break/continue
• Tail call optimization: compile-time in_tail_position detection + TailCall opcode

Future Plans (not yet implemented)

• NaN-boxing: 8 bytes per value, fits in register
• Arena allocation for AST nodes
• Small-dict optimization (<=8 entries: flat array)
• Frame pool for function calls

Benchmarks

• ion-core/benches/ion_bench.rs with criterion
• Targets: fib, loop, map/filter, string, match, comprehension, closures