Inside Go Channels: hchan Struct, the GMP Scheduler, and the Send Path

A developer went through Go's runtime/chan.go source line by line to document what actually happens under the hood when you execute ch <- val, rather than relying on the surface-level explanation that channels let goroutines communicate. The write-up starts with the GMP scheduler model — goroutines, Ms (threads), and Ps (processors) — explaining why it makes concurrency in Go so cheap, then moves into the fields of the hchan struct and the three-step logic behind a channel send.

One notable detail covered is how, for an unbuffered channel send, the buffer is skipped entirely in favor of a direct stack-to-stack handoff between goroutines. The piece also walks through the nil-channel trick for disabling a select case at runtime, worked examples of pipelines, fan-out/fan-in, and worker pools, a deadlock the author had to debug, and guidance on when sync.Mutex is a better fit than channels along with practical use of the race detector.

This kind of source-level breakdown has direct practical value for engineers writing production Go code: understanding channel behavior and performance characteristics at the runtime level, rather than just the API surface, sharpens the ability to diagnose deadlocks and race conditions. The author is specifically seeking community feedback on the hchan interpretation, noting that part of the runtime remains genuinely debated among Go practitioners.