polyfill2 0.3.0

A high-performance Rust client for Polymarket's CLOB V2 API, with latency-optimized data structures and zero-allocation hot paths. This is a V2 migration fork of polyfill-rs (Julius Tranquilli), updated in April 2026 for the CLOB V2 cutover (new EIP-712 Order schema, new contract addresses, new endpoint URLs, new WebSocket message shapes).

polyfill-rs was originally created as a performance-focused Rust alternative to polymarket-rs-client, aiming to beat the benchmarks quoted in that project's README while maintaining zero-allocation hot paths. This fork carries those properties forward onto V2.

On zero-alloc: in this project, "zero-alloc" means zero allocations in the per-message handling loop after init/warm-up — i.e. the hot path never touches the heap, even though cold paths do. Order book paths that introduce new allocations by design:

• First time seeing a token/book (HashMap insert + key clone): src/book.rs:~788
• New price levels (BTreeMap node growth): src/book.rs:~409

Quick Start

Add to your Cargo.toml:

[dependencies]
polyfill2 = "0.1"

use polyfill2::{ClobClient, Side, OrderType};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {
    let client = ClobClient::new("https://clob-v2.polymarket.com");
    let markets = client.get_sampling_markets(None).await?;
    println!("Found {} markets", markets.data.len());
    Ok(())
}

Performance Comparison

Note: the numbers below were measured on upstream polyfill-rs against CLOB V1 (clob.polymarket.com) prior to the V2 migration. Computational paths (book updates, spread/mid, WS decode) are unchanged and should still hold; the networked end-to-end number has not been re-measured against V2 — #1 tracks re-measurement.

Real-World API Performance (with network I/O)

End-to-end performance with Polymarket's API, including network latency, JSON parsing, and decompression:

Performance vs polymarket-rs-client:

• 21.4% faster
• 32.5% more consistent
• 4.2x faster than Official Python Client

Benchmark Methodology: All benchmarks run side-by-side on the same machine, same network, same time using 20 iterations, 100ms delay between requests, /simplified-markets endpoint. Best performance achieved with connection keep-alive enabled. See examples/side_by_side_benchmark.rs in commit a63a170: https://github.com/floor-licker/polyfill-rs/blob/a63a170/examples/side_by_side_benchmark.rs for the complete benchmark implementation.

Computational Performance (pure CPU, no I/O)

Run the WS hot-path benchmark locally with cargo bench --bench ws_hot_path.

Key Performance Optimizations:

The 21.4% performance improvement comes from SIMD-accelerated JSON parsing (1.77x faster than serde_json), HTTP/2 tuning with 512KB stream windows optimized for 469KB payloads, integrated DNS caching, connection keep-alive, and buffer pooling to reduce allocation overhead.

Memory Architecture

Pre-allocated pools eliminate allocation latency spikes. Configurable book depth limiting prevents memory bloat. Hot data structures group frequently-accessed fields for cache line efficiency.

Architectural Principles

Price data converts to fixed-point at ingress boundaries while maintaining tick-aligned precision. The critical path uses integer arithmetic with branchless operations. Data converts back to IEEE 754 at egress for API compatibility. This enables deterministic execution with predictable instruction counts.

Measured Network Improvements

Credits

polyfill-rs was originally authored by Julius Tranquilli (floor-licker/polyfill-rs). This V2 fork (polyfill2) is maintained by onsails.