PEGO
R060: Chunked Document Input
A small edit to a large Document should not copy the entire unchanged text suffix. Immutable chunks and indexed piece ranges replace full-buffer splicing above a size threshold.
Mechanism and Invariants
Documents below 4,096 logical positions stay flat. Growing across the threshold promotes once to piece storage; later shrinking does not demote. Edits split/replace ranges, and more than 512 pieces triggers compaction while keeping piece storage. Byte boundaries, malformed UTF-8 and code-point positions preserve the input contract. Ordinary batch and stream input stay contiguous.
Scope and State
Default eligible Document storage in closure and both VMs. Generated Go and TypeScript have no Document API.
Reference Control
Build tag pego_reference_document_pieces selects flat same-source input. It
isolates storage while retaining surrounding Document logic.
Evidence and Implementation
Record 093 measures CodePoints edit-plus-reparse about 19–22% faster, and long CSV about 25–30% faster, with lower allocation volume. Small edit-only cases can cost more, supporting the threshold; batch controls show 0.76–3.31% cost. These results are workload-specific. Design 051 defines ownership and boundary behavior.
Source owners: useDocumentPieces,
piece input, storage contract.
The comparison policy distinguishes same-source controls, historical changes and full checkpoints.