Lowering to Bytecode

Why lower?

• AST is ideal for analysis; bytecode is ideal for execution
• linear IR reduces execution complexity
• deterministic output improves debugging and testing

Core contract

• node-kind -> deterministic instruction sequence
• preserve evaluation order
• preserve side effects
• preserve errors where possible

Stack-bytecode pattern

• emit operands before operators (postfix/RPN)
• unary: [operand] OpNot/OpNeg
• binary: [left] [right] OpAdd/OpSub/...
• identifiers become LOAD, not special-cased pushes

Calls and scope

• lower arguments in source order
• CALL name arity
• blocks with locals: ENTER_SCOPE then body then EXIT_SCOPE

Determinism goals

• same AST always emits same instructions
• roundtrip checks: AST -> BC -> BC1 -> AST behavior
• clear failure on unknown node kinds

Why this matters

• wrong order often returns “valid” but wrong answers
• wrong scoping leaks locals
• stable lowering allows you to evolve to register IR later

Quick checks

• 1 + 2 * 3 becomes postfix emit order
• (a + b) / c emits ADD before DIV
• f(1, x, 2) emits three arg loads, then call
• block enters/exits scope explicitly