Performance

[guybedford]

Jco has prioritised correctness over performance. As adoption grows, we will need to investigate performance more carefully, both for code generation and for host call overhead.

The first step here would be to obtain suitable performance benchmarks. The conformance test suite itself may be suitable for treating as a benchmark, but more dedicated benchmark components with predictable runtime performance would be beneficial here.

With benchmarks in place it would then be possible to explore:

• The performance benefits of low-level bindgen implementation
• Code generation optimizations
• The viability and benefit that might be incurred from native implementations of WASI, either through NAPI or through direct host integration via uvwasi

There is potentially a lot of low-hanging fruit from a performance perspective, and it has not been a major priority yet. Hopefully this issue can point towards steps to take as users seek more investment here.