Tracking issue for rfcs/0000-scriptc-task-runtime-adapter.md (status Draft, number 0000 pending maintainer assignment). Opened per rfcs/README.md step 1 and IMPL-EVAL finding 1 on the drafting PR.
Provenance
Proposal in one paragraph
Compile CPU-bound tasks written in restricted TypeScript to static native binaries with scriptc and dispatch through the existing executable TaskType. Measured through the production dispatch path: end-to-end p50 108.0 → 64.0 ms (40.7%), executor-side 50.6 → 6.8 ms (7.4×), peak RSS 17×, fanout throughput ~2.2× — but compute runs at JS speed, not native (10M-iter loop: scriptc 225 ms ≈ V8 219 ms vs Rust 53 ms / WASM 54 ms). Recommendation: phased — recipe + experimental customAdapters adapter now; first-class scriptc TaskType gated on upstream maturity (vercel-labs/scriptc#173 JSR support, npm handling out of experimental, post-0.0.x stability, signing story).
Lifecycle
Related: #1679 (monty sandboxed Python — same adapter seam, in-process-sandboxed quadrant), arch-debt workers-non-deno-task-sandbox-boundary.
Tracking issue for
rfcs/0000-scriptc-task-runtime-adapter.md(status Draft, number 0000 pending maintainer assignment). Opened perrfcs/README.mdstep 1 and IMPL-EVAL finding 1 on the drafting PR.Provenance
.llm/runs/claude-harness-profile-rfc-benchmark-shzhgv--scriptc-rfc-bench/, benchmark data + pinned environment manifest in the run dir)deepseek-v4-flash-0731, evaluated head13e212e8, verdict comment on docs(rfc): scriptc task runtime adapter — dispatch-path benchmark + RFC #1678) — quantitative claims independently recomputed from raw JSONL; source-seam claims verified at head.Proposal in one paragraph
Compile CPU-bound tasks written in restricted TypeScript to static native binaries with scriptc and dispatch through the existing
executableTaskType. Measured through the production dispatch path: end-to-end p50 108.0 → 64.0 ms (40.7%), executor-side 50.6 → 6.8 ms (7.4×), peak RSS 17×, fanout throughput ~2.2× — but compute runs at JS speed, not native (10M-iter loop: scriptc 225 ms ≈ V8 219 ms vs Rust 53 ms / WASM 54 ms). Recommendation: phased — recipe + experimentalcustomAdaptersadapter now; first-classscriptcTaskType gated on upstream maturity (vercel-labs/scriptc#173 JSR support, npm handling out of experimental, post-0.0.x stability, signing story).Lifecycle
tracking-issueto this issue, fold in the evaluator's wording nit — "~57 ms" dispatch share reads 58–62 ms per subject in the raw data)rfcs/README.mdcustomAdaptersexample (docs-only), and the maturity-gate checklist for first-class promotionRelated: #1679 (monty sandboxed Python — same adapter seam, in-process-sandboxed quadrant), arch-debt
workers-non-deno-task-sandbox-boundary.