82. Match Typed Direct Bodies to General Frame Layouts

The typed Go structural emitter now derives capture-slot layouts from the general emitter's metadata for eligible left-recursion, Operators Rule and owned repetition scopes. LR bodies use the authoritative shared Node-rule names; independent Operators Rule and repetition scopes use the same build, predicate, projection and slot-order rules as the general emitter. The collector also visits dead syntax, so omitted captures still have the expected nil slots. The runtime clears slots that are not live for a match. Additional required slots can widen a frame or pool class, so unchanged allocations in these workloads do not establish zero memory cost for every grammar. Node generation, recognition and TypeScript output are unchanged. Eligible recovery bodies use the separate structural route in record 83; unmatched recovery layouts remain on general dispatch.
The generation-only disableTypedFrameLayouts control retains the previous discovery and fallback route for same-generator comparisons. It selects no runtime branch. Layouts that still cannot be represented by the structural emitter remain on the general typed path.
The eight-case corpus checks direct/reference and conversion parity in both position units, including dead captures before and after live captures, infallible alternatives, element predicates, projections and lookahead. Additional checks compare collected slot metadata with the actual general emitter for value-building and recognition paths, including twins and dead syntax. Workload tests cover parser reuse, pool reuse and returned AST ownership.

Five alternating 300 ms pairs used Go 1.27.1 on darwin/arm64, Apple M3 Max, 16 CPUs. Matching ratios are medians of direct/general time ratios, with paired min–max ranges:

Workload	n	Unit	Median ratio	Paired range
Left recursion	1	CodePoints	0.965×	0.942–1.005×
Left recursion	1	Bytes	0.969×	0.927–1.017×
Left recursion	128	CodePoints	0.907×	0.888–0.937×
Left recursion	128	Bytes	0.934×	0.877–0.984×
Operators Rule	1	CodePoints	0.939×	0.921–1.007×
Operators Rule	1	Bytes	0.970×	0.953–1.042×
Operators Rule	128	CodePoints	0.904×	0.871–0.937×
Operators Rule	128	Bytes	0.945×	0.929–0.996×

Every paired ratio for the larger inputs is below parity; all small-input ranges cross parity. Allocation counts are unchanged at 4/11 for LR and 5/12 for Operators Rule at n=1/128, respectively; cumulative bytes are about 11.9/41.6 KB and 12.4/45.1 KB.

The separate generation benchmark includes generation, formatting and runtime embedding; grammar parsing is outside its timed loop. Ratios are medians of five paired direct/general ratios, with paired min–max ranges:

Fixture Median ratio Paired range

Left recursion 0.989× 0.964–1.061×

Operators Rule 0.972× 0.942–1.019×

Calculator 1.007× 0.988–1.105×

TypeScript 0.983× 0.960–1.007×

All paired ranges cross parity. Generated source is 109,223 versus 111,790 bytes for LR and 112,143 versus 113,845 bytes for Operators Rule; calculator and TypeScript source sizes are unchanged. Layout collection adds generation-only work, including slot-map allocations; these measurements do not establish a generation-time benefit or cost. The matching results are synthetic and do not establish a broad parser speedup.

Fixture	Median ratio	Paired range
Left recursion	0.989×	0.964–1.061×
Operators Rule	0.972×	0.942–1.019×
Calculator	1.007×	0.988–1.105×
TypeScript	0.983×	0.960–1.007×

Generate the direct and same-generator reference workload fixtures and check parity:

export PEGO_TYPED_FRAME_LAYOUT_DIR=/tmp/pego-typed-frame-layouts
go test ./internal/engine -run '^TestGeneratedTypedFrameLayoutWorkloads$' -count=1

Build the direct/general fixtures into separate binaries:

for workload in lr operatorsRule; do
  for variant in direct general; do
    fixture="$PEGO_TYPED_FRAME_LAYOUT_DIR/$workload/$variant"
    (cd "$fixture" && go test -c -o "$PEGO_TYPED_FRAME_LAYOUT_DIR/$workload-$variant.test" .)
  done
done

Alternate five paired runs of BenchmarkTypedLR and BenchmarkTypedOperatorsRule with -test.run='^$' -test.benchtime=300ms -test.benchmem, reversing direct/general order between pairs. The matching loop in record 81 can be reused with PEGO_TYPED_FRAMED_CUT_DIR="$PEGO_TYPED_FRAME_LAYOUT_DIR". The inputs use the same lrInput(n) and operatorsRuleInput(n) workloads: n=1/128 means two/129 LR groups and one/128 Operators Rule postfix groups after the base atom.

To compare generation separately, keep PEGO_REFERENCE_FRAME_LAYOUTS fixed before each timed run:

PEGO_REFERENCE_FRAME_LAYOUTS=0 go test ./internal/engine -run '^$' \
  -bench '^BenchmarkTypedFrameLayoutGeneration$' -benchtime=300ms -benchmem
PEGO_REFERENCE_FRAME_LAYOUTS=1 go test ./internal/engine -run '^$' \
  -bench '^BenchmarkTypedFrameLayoutGeneration$' -benchtime=300ms -benchmem

The generation fixtures parse their grammar before the timer and then measure Generate; alternate the reference setting across five pairs. Source sizes are reported as a benchmark metric, not as a generated-parser runtime measure.