PEGO
G020: Lazy Backend Preparation
A compiled grammar need not prepare every execution backend before its first parse. PEGO delays derived bytecode, VM metadata and the recognition graph until an operation needs them, then caches the prepared state.
Mechanism and Invariants
Program owns distinct sync.Once guards for the module, recursive VM, iterative
VM and recognizer. Module compiles bytecode once; rule prepares the selected
VM's metadata from it. recognizer builds a separate graph and caches either
its program or its setup error. Grammar-less programs cannot construct closure
or recognition graphs. The ordinary closure parse graph is built eagerly during
compilation.
Scope and State
Default engine setup for closure recognition and both VMs. Generated Go Node/typed Go and TypeScript embed their machinery and are outside scope. This is first-use preparation, not per-rule matching work.
Reference Control
No same-source eager-preparation control or private opt-in alternative is maintained.
Evidence and Implementation
There is no numbered optimization record or isolated timing/allocation measurement. Warm parse results do not quantify avoided setup, and first-use timing must separate preparation from steady-state matching.
Source owners: Program.BytecodeOnce,
Program.recOnce, vmOnce, ivmOnce and Module,
Program.BytecodeModule, Program.Recognizer and rule.
The comparison policy distinguishes same-source controls, historical changes and full checkpoints.