v0.5.2

[0.5.2] - 2026-04-17

Performance Gains

v0.5.2 represents a massive leap in raw throughput and memory efficiency. By eliminating heap allocations and consolidating the serialization/fragmentation pipeline, Webshocket now hits staggering numbers:

• RPC: ~45,000 RPCs/s at 1.1 ms latency (asyncio, 50 concurrent clients)
• Streaming: 289 MB/s on asyncio, 421 MB/s with winloop

Measured on AMD Ryzen 5 5600G, 14 GB RAM, Windows, Python 3.13. Results vary by hardware.

Changed

• Zero-Allocation Serialization: Replaced msgspec.encode() with encode_into() using a pre-allocated bytearray per connection, eliminating heap allocations on every send.
• Unified _send_rpc_response Fast Path: Merged _send_rpc_response and _send_rpc_packet into a single method that encodes directly into the per-connection buffer.
• Extracted _write Primitive: Consolidated all fragmentation logic into ClientConnection._write(). All send paths delegate to it.
• Broadcast & Publish Serialize-Once: Serialize the Packet once per wire format, then fan out pre-encoded bytes to all recipients.
• Wire Format Reduction: Enabled omit_defaults=True on Packet, reducing typical RPC payload size by ~20-35%.
• Client Frame Queue Hardening: Increased client _frame_queue capacity to 8192 for burst tolerance. Added graceful QueueFull handling (log + drop) to prevent process crashes at the sync/async boundary.
• Python 3.10 Support: Replaced typing.Self with a string literal return annotation. Lowered requires-python to >=3.10. Added 3.10 to CI matrix.

Added

• RateLimitError, ClientType, RPCErrorCode exported from webshocket top-level.
• DEFAULT_ENCODE_BUFFER_SIZE constant (512 bytes) in constant.py.
• Benchmark Suite: benchmark/benchmark.py with 2×2 grid (RPC throughput, RPC latency, stream throughput, stream depth) and benchmark/README.md with event loop selection guidance.

Fixed

• ReadTheDocs Build: .readthedocs.yaml now installs the package itself (method: pip, path: .), fixing empty API docs on the hosted site.