PEGO
A070: Typed Output Bootstrap
A tiny typed result can retain a large first slab containing discarded siblings. Independent early allocations keep common small results from immediately paying that slab footprint.
Mechanism and Invariants
For each owner, the first three objects/terminals allocate independently before object slabs grow from 8 to 256 cells. The first nonempty small list also allocates independently before list slabs grow from 64 to 1,024 elements. Later slab maxima and exact-size large lists remain unchanged. Fresh ownership preserves append isolation, spans and shared result values; returned chunks are not pooled. Full/bounded liveness sweeps keep legacy helpers and are measured separately.
Scope and State
Applied by default to generated Go typed output, both direct construction and Node conversion. Ordinary Node parse/recognition, closure, VMs and TypeScript are unchanged.
Reference Control
Private generation option disableTypedOutputBootstrap emits the legacy
first-slab strategy. Object/list slab controls can further isolate ownership.
The legacy same-source comparator retains bootstrap state fields, so it is not
layout-identical to historical main.
Evidence and Implementation
Record 098 measures direct discard-heavy retained output from about 2.1 MB to at most a few KiB after cloning returned Text. Conversion retained little in both routes. Ordinary XML Bytes costs about 2.3% versus the same-generator legacy route, with additional allocations; small JSON improves. These scoped tradeoffs supported adoption and do not establish a general speedup.
Source owners: typed allocation helper emission, typed allocator state,
disableTypedOutputBootstrap,
retained-output fixture.
The comparison policy distinguishes same-source controls, historical changes and full checkpoints.