docs(r13): the Go runtime for systems — example 44 + chapter 44 (Deep Dives)

[patterncatalyst]

Closes Part 12 (Deep Dives) with the Go counterpart to ch43's Rust deep dive: a single Go binary that observes its own runtime and prints self-computed ok/FAIL verdicts across four themes.

Example 44-go-runtime-for-systems (mode: local, langs: [go], stdlib only)

• sched — GMP scheduler: a rendezvous barrier (8 goroutines concurrently alive), a worker pool (every job once), and a no-yield-point spinner under GOMAXPROCS=1 proving signal-based (SIGURG) async preemption lets a starved ticker progress.
• gc — pacing: auto-GC off, one runtime.GC() → NumGC delta exactly 1; the same churn under GOGC=100 vs 800 and under a tight GOMEMLIMIT shows the pacer's CPU/memory trade-off.
• netpoll — 2000 goroutines block on pipe reads while OS threads stay bounded (<= 4*GOMAXPROCS+16): goroutines are not threads.
• knobs — effective GOMAXPROCS/NumCPU/GOGC/GOMEMLIMIT/GODEBUG.
• verify.lua matches each theme's fixed verdict label (PASS 16 / FAIL 0); numbers vary per run, labels don't.

Runtime facts current to Go 1.26 (source-verified vs go1.26.5): Green Tea GC on by default, container-aware GOMAXPROCS (since 1.25), the _Psyscall P-state removed (~30% lower cgo overhead), netpoller on epoll_pwait.

Chapter 44

To the spine (Part "Deep Dives"): Tools-used box, verbatim go excerpts (no tri-language codetabs), Figures 44.1 (GMP scheduler) + 44.2 (goroutines-not-threads netpoller contrast), verified footer with real gctrace/schedtrace captures and the GODEBUG=asyncpreemptoff=1 FAIL contrast (ticker_incr=0).

Gate

• --only 44 → 1 passed / 0 failed
• full --mode local → 98 passed / 0 failed / 3 skipped (skips are 38-otel, lgtm down)
• validate.py: OK

The implementer also caught + fixed a real goroutine launch-order livelock that only manifested under asyncpreemptoff=1 (spawn the ticker first so the spinner claims runnext).