对若干优秀开源项目的源码深度分析——AI 辅助阅读、人工核对行号，每个代码引用都可点击直达固定 commit 的精确源码行。

• Aeron — a high-performance, low-latency messaging transport for streaming data over UDP and IPC.
• colibrì — a single-file, zero-dependency C engine that runs the 744B-parameter GLM-5.2 MoE on a consumer machine by streaming experts from disk.
• Ninja — a small build system built with a singular focus on speed.
• nuthatch (self-built) — a from-scratch streaming MoE inference engine I wrote to test whether a usage-learned expert cache beats naive LRU / OS caching (it does — and cross-architecture). 源码分析（11 章） · 仓库。
• upb — a small, fast C implementation of Protocol Buffers.
• vLLM — a high-throughput, memory-efficient inference and serving engine for LLMs.