Max Lvshadowsocks-libev 与 shadowsocks-android 核心维护者，SIP 规范主导者，meow-rs 内核创立者

作者发言

GitHub·2026.10.10 14:05:13（UTC+8）

Max Lv：meow-rs 多路复用 WebSocket 修复已本地复核通过，补充说明后待 CI 通过合并

Max Lv 复核 meow-rs PR #745（稳定多路复用 WebSocket 连接）：在 macOS 本地验证新增 ws_test、mux 库与 api_test 的 DNS 用例通过，fmt、三路 clippy、rustdoc 干净。他在原提交上追加了一个小提交 dc72528：为 ws.rs 的 16 KiB 单次 poll_write 上限与 64 KiB tungstenite 写缓冲上限命名并加 const 断言关联，更新过时的 ws_config() 文档注释；muxcool.rs 的 MAX_UDP_PAYLOAD 改为从 MAX_PAYLOAD 派生。两点记录、不要求改动：write_packet 对超大报文返回 Err 会让 meow-tunnel 驱逐该 UDP NAT 条目（下一个包重新拨号建流），与 SS/SOCKS5/Trojan/anytls 现有行为一致，比丢掉整个 mux 会话好；vless/conn.rs 的纯 VLESS UDP writer 仍只守 u16::MAX，值得单独检查 Xray 的 packet reader 是否也施加 8 KiB 上限。CI 在新 head 上通过后即 squash-merge。

作者原文@madeye

Reviewed, thanks. Verified locally on macOS: the new ws_test, mux lib and api_test DNS cases pass, and fmt / three-way clippy / rustdoc are clean.

I pushed one small follow-up commit (dc72528) on top of yours rather than round-tripping:

• ws.rs: named the 16 KiB per-poll_write cap and the 64 KiB tungstenite write-buffer ceiling (WS_WRITE_CHUNK, WS_MAX_WRITE_BUFFER) with a const assert tying them together, and refreshed the ws_config() doc comment that still said every write is a single relay-buf chunk.
• muxcool.rs: MAX_UDP_PAYLOAD now derives from MAX_PAYLOAD since both come from Xray's buf.Size.

Two notes for the record, no change requested:

• Returning Err from write_packet for an oversized datagram makes meow-tunnel evict that UDP NAT entry (next packet redials a mux stream). That matches what the SS/SOCKS5/Trojan/anytls writers already do, and it is far better than losing the whole mux session, so I kept it.
• The plain VLESS UDP writer in vless/conn.rs still only guards u16::MAX; worth checking separately whether Xray's packet reader applies the same 8 KiB bound there.

Will squash-merge once CI is green on the new head.