FAQ #

Which scheduler should I use? #

Use global_queue for the simplest behavior and work_stealing for most performance-sensitive work. Use lane_extension only when you need an explicit blocking lane.

How many workers can I configure? #

The current hard cap is 64 workers because the non-empty-worker bitmap uses a single u64.

How do I run the full test suite? #

See TESTING.md. The shortest full gate is:

zig build test-all --summary all

How do I reproduce the Tokio and Go comparisons? #

Run:

git submodule update --init --recursive
zig build bench-runtime-compare -Doptimize=ReleaseFast -- --output benchmarks/results/runtime-comparison-latest.json

The scripts bootstrap a pinned local Go toolchain from third_party/go so the results come from the vendored runtime source tree rather than the host Go installation.

Where do the benchmark artifacts live? #

• CI-safe suite: benchmarks/results/latest.json
• scale suite: benchmarks/results/scale-latest.json
• direct runtime comparison: benchmarks/results/runtime-comparison-latest.json

Why are the direct runtime numbers different from older docs? #

Older notes often compared this pool to Tokio or Go indirectly or used a different host state. The current direct snapshot in RUNTIME_COMPARISON.md comes from a same-host, same-command run against vendored Tokio and Go source trees.