PEGO
G030: Constructor Dependency Analysis
Sample construction need not sweep every rule whenever one derivation estimate changes. Reverse dependencies and a bounded worklist propagate only changes that can affect callers.
Mechanism and Invariants
Analysis builds contiguous reverse edges, deduplicates repeated calls, seeds a worklist in explicit DFS postorder and reevaluates callers when minimal heights, byte lengths or coverage sets change. Acyclic callees run first; cycles converge to the same fixed point. Target numbering and exclusions for negative lookahead/recovery remain unchanged. This changes analysis work, not the grammar's matching semantics.
Scope and State
Default shared sample.New analysis requiring a grammar AST. All parser-backend
matching cells are not applicable: selecting closure or a VM does not run this
work in the matching loop. Standalone generated parsers do not perform it.
Reference Control
No maintained same-source old-sweep control exists. The focused comparison is historical parent versus candidate analysis.
Evidence and Implementation
Historical record 071 times analysis after grammar parsing. A caller-first depth-2,048 alias chain changes from 234.498 to 0.9909 ms, but B/op grows 8.6–13.0% across chain cases for dependency/worklist storage. Leaf-first ranges overlap parity. These are constructor-analysis scaling and startup tradeoffs, not parsing throughput or a bound on all cyclic grammars.
Evidence records: 071.
Source owners: analyze and reverse-dependency worklist.
The comparison policy distinguishes same-source controls, historical changes and full checkpoints.