PEGO
M020: Character-Class Tests
A character class should not repeatedly walk its complete range table when a small comparison or ASCII bit lookup can decide membership.
Mechanism and Invariants
Closure compilation specializes common one- and two-range classes into captured constant comparisons. Larger closure classes use an ASCII bitmap and a copied range fallback. VM program preparation builds 128-bit ASCII masks once; non-ASCII characters retain range tests. Generated Go and TypeScript emit class comparisons from the grammar. Negation and out-of-input behavior must remain identical on every route; an ASCII mask does not replace Unicode membership.
Scope and State
Default class specialization applies in all six targets: closure, recursive VM, iterative VM, generated Go Node/recognition, typed Go and TypeScript. Generated direct Go bodies can also read decoded code points directly, a separate access optimization.
Reference Control
No universal same-source class-specialization control exists. A bitmap-only closure experiment was not adopted; its result does not describe the current small-range specialization.
Evidence and Implementation
Historical record 017 measures 1.3–4.6% faster VM recognition on four workloads. Record 031 measures 1.2–4.9% faster closure full parsing and 1.6–9.4% recognition. Record 011 includes generated inline comparisons among several changes. These backend-specific results do not isolate every current generated class path.
Evidence records: 011, 017, 031.
Source owners: classAccept, newASCIISet and VM class tests, class test emission, TypeScript class test emission.
The comparison policy distinguishes same-source controls, historical changes and full checkpoints.