Testing

• internal/** unit tests validate implementation-level behavior.
• e2e/** contains isolated Neva modules with Go harnesses for language and standard-library regressions.
• examples/** are executable user-facing documentation; all examples belong to one module and must compile together.
• internal/runtime/**/*_bench_test.go measures native runtime functions and runtime primitives in isolation.
• benchmarks/** measure explicit language-level and composed runtime performance questions.

Run the smallest meaningful scope while iterating, then widen validation when a change crosses compiler, runtime, or public standard-library boundaries. Generated tests should carry a short intent comment. Runtime benchmarks never substitute for unit tests.

E2E and Examples

Each e2e/ package is an independent Neva module. Run focused e2e packages while iterating; use a broader e2e run when the change crosses module or standard-library boundaries.

examples/ is one module. A single example run still requires all examples to compile, so examples must remain readable, deterministic where their topology requires it, and suitable as user documentation.

Benchmarks

Benchmarks have three tiers:

• atomic: one primary component under test;
• simple: a small composition for one focused scenario;
• complex: a multi-domain, more realistic pipeline.

Keep the layout flat by tier: benchmarks/<tier>/<pkg>_<component>[_<context>]/main.neva. Support wiring is allowed only when necessary to make a standalone program valid and should be documented. Use the Go benchmark harness for iteration by default; internal Neva loops are reserved for deliberate throughput measurements.