Speed up macOS CI by cross-compiling Intel tests on Apple Silicon and sharding with nextest
DanielLockyer · x · 2026-11-17
Daniel Lockyer shared a practical CI speedup for the lumen project: the macos-15-intel job was by far the slowest JIT task, with 4 Intel cores spending 160–270s just compiling. His PR moves the compile to Apple Silicon, packages the test binaries via cargo nextest archive, then has 3 Intel shards download and run with --partition count:N/3 — no Rust toolchain needed on the Intel runners.
It works because lumen has no build.rs, embeds V8 fixtures with includestr!, and reads no files at runtime; subprocess tests use currentexe(), which is nextest-compatible. The screenshot shows the sharded run finishing in 3m49s vs a 4m39s baseline. A useful pattern for teams stuck with slow legacy Intel runners.
More from coding & agent
- Qwen4-Max goes live on Lebrel, tops DeepSWE coding benchmark at 7% below list price — Alibaba_Qwen · 2026-11-17
- Pageloop Launches a Docs Agent That Auto-Captures Screenshots and Updates Stale Docs — prasanna_says · 2026-11-17
- WingLLM: Go daemon doubles 7B LLM speed to 2.54 tok/s on a 2GB GPU via speculative decoding — rbm3267 · 2026-11-17
- Developer tip: use MCP to pass UI templates alongside fetched data, not just raw results — js_craft_hq · 2026-11-17
- OutfitterBench tests LLM agents on 200-day rental business, exposing commitment failures — Jie Huang · 2026-11-17
- Craftax framework evolves code policies via feedback cascade, reaching final stage 50% of time — Víctor Gallego · 2026-11-17