Repository navigation
HermitCrab: #546 follow-ons (compiled slot-rule sharing, ParallelApplySlots) are unbuilt and unmeasured in work steps #608
Description
Claude-drafted.
Two follow-ons to PR #546 (shared trailing slot batches) are not built and have no work-step measurement: sharing compiled slot rules across templates, and a design for ParallelApplySlots, whose layered breadth-first queue needs its own design.
Status: not started. Waits on #546 landing.
What we know:
- perf(hermitcrab): share trailing analysis template slot batches #546's C# gain is counted in slot batches on samples, not in work steps on full lists: Sena 31.4% fewer (47,072 to 32,287), Mbugwe 19.3% fewer (72,467 to 58,486).
- The port measured the same batch sharing in work steps: Sena 11.5% less work, Mbugwe 18.1% less (1,364-word partial), Aweti 0.35%, none on one-template grammars.
- perf(hermitcrab): share trailing analysis template slot batches #546 covers sequential, untraced walks without an alternative limit.
ParallelApplySlotsis outside it.
Not known: the work-step effect of compiled-rule sharing; what a ParallelApplySlots design needs.
Proposed change (C#): measure the work-step effect of compiled-rule sharing on the full word lists, then design ParallelApplySlots. A deterministic work counter on full lists comes first (lead 7 of the 2026-10-21 round).
Related: PR #546 (open), #566 (open). PanGloss: sillsdev/PanGloss#110 (the pointer issue for this one, open); sillsdev/PanGloss#66 (open) ports the batch sharing of #546 and #566 only, not these follow-ons.
Next: once #546 lands, count compiled-rule sharing in work steps on the full lists.
Evidence -- where the numbers come from
- C# batch counts and port work steps: ledger row 26 (Sena 47,072 to 32,287 is 31.4%; Mbugwe 72,467 to 58,486 is 19.3%; port Sena 11.5%, 31.6% fewer steps, 448,573 batches avoided; Mbugwe 18.1% on a 1,364-word partial; Aweti 0.35%).
- The round's own wording: "The C# gain is measured in slot batches on samples, not in a work counter on full lists."
ParallelApplySlotsexists today atAnalysisAffixTemplateRule.cs#L96(called from line 71).
Activity
[-]HermitCrab #546 follow-ons: share compiled slot rules across templates; design ParallelApplySlots[/-][+]HermitCrab: #546 follow-ons (compiled slot-rule sharing, ParallelApplySlots) are unbuilt and unmeasured in work steps[/+]