VM #

VCL ships a bytecode compiler and VM for fast, deterministic execution. The VM matches interpreter behavior and is selectable via --vm in the CLI.

Pipeline #

1. Parse + link: parse → linkProgram
2. Lower to IR: lowerProgram
3. Compile to bytecode: compileIR
4. Execute: runBytecode(bytecode, handlerName, runtime)

Handler naming #

Event handlers are exported as on_<event>. For HTTP events, http GET /health becomes on_http GET /health. Use the same name when invoking runBytecode.

Runtime surface #

The VM expects the same runtime shape as the interpreter: { fetch, send, store, log, now? }.

Example #

import { parse } from '../dist/parser/index.js'
import { linkProgram } from '../dist/module/linker.js'
import { lowerProgram } from '../dist/ir/index.js'
import { compileIR } from '../dist/vm/compiler.js'
import { runBytecode } from '../dist/vm/vm.js'

const ast = parse(source)
const linked = linkProgram(ast, process.cwd())
const bc = compileIR(lowerProgram(linked))

const result = runBytecode(bc, 'on_http GET /health', {
  fetch: () => [],
  send: () => {},
  store: () => {},
  log: console.log,
})

Scheduling #

Scheduled handlers (every 5 minutes: ...) are lowered to functions you can call manually. Hook them to your scheduler of choice and pass them through the VM runner the same way.

Determinism #

The VM is designed to be deterministic given the same runtime inputs. Provide a fixed now() for repeatable tests.