[help wanted] Benchmark harness proving the 10x-vs-JS claim (criterion + Node baseline)

[abhicris]

Context

The repo tagline claims 10x faster than JS solvers. We need a reproducible benchmark harness to back that up — it becomes a marketing artifact and a regression tripwire. criterion is the canonical Rust benchmarking crate; we can compare the Rust Across decoder against the JS reference implementation (@across-protocol/sdk) on the same fixture set.

Scope

Medium-effort ticket.

• Add criterion = "0.5" as a dev-dep.
• Create benches/intent_decoding.rs that, for each decoder, benchmarks:
  • Parsing N=100 fixtures from tests/fixtures/.
  • Cold-path (first parse) vs warm-path (steady-state).
• In the same PR, add benches/compare_js/ — a tiny Node script (npm i @across-protocol/sdk pinned) that decodes the same fixtures and emits ns/op.
• benches/README.md documents: cargo bench for Rust, node benches/compare_js/index.js for JS. Include a generated table of the last-measured numbers.
• CI job .github/workflows/bench.yml — runs cargo bench --no-run to ensure bench code at least compiles on every PR. Actual perf numbers run on-demand, not per PR.

Acceptance criteria

• cargo bench runs successfully locally.
• JS comparison script runs and produces a number in the same units (ns/op or μs/op).
• benches/README.md shows a table with at least one data point per decoder and the 10x ratio (or an honest correction if it's not 10x).
• CI job green.

Reference: resolver#1 is the existing Benchmark UniswapX/Across/CoW fill-latency issue — this ticket is complementary (decoder latency, not fill latency). PR #3 shows the Across decoder structure under test.

Estimated effort

M (1–2 days) — bulk of time is setting up the JS side reproducibly.

— kcolbchain / Abhishek Krishna

[abhicris]

This issue/PR has had no activity for 21 days and is being closed automatically by the kcolbchain PR pipeline.

If it's still relevant, reopen or leave a comment and we'll re-triage. For the full policy see pipeline docs.