go-concurrency

Go Concurrency

Judge every change by its happens-before story: a claim that two events cannot race is either an edge you can name — channel operation, lock, WaitGroup, Once, atomic — or it is false.

shared state -> goroutine lifetime -> happens-before edges -> cancellation -> unblock and join -> bounds -> proof

Every goroutine has an owner, a signal it observes to stop, a guaranteed unblock for each blocking site, and a join point; a goroutine missing any of these outlives its purpose and leaks work, memory, or writes into freed assumptions. Worker pools and fan-outs carry explicit bounds, and timers and tickers have named stop owners.

Load goroutine leak profiles for goroutine-count growth, shutdown stalls, production liveness incidents, or changes to channel, lock, wait-group, and process-exit ownership.

For a delegated Decision or Review, or when the active artifact requires its result interface, load the shared specialist contract. Trace every changed spawn site to its join or explicit process-lifetime disposition. For interacting goroutines, exhaustive blocking-site coverage, or a decision/review handoff, record GoroutineStory{owner, stop, blocking_sites, unblock, join, bound, happens_before, proof} per path. A single local path can keep its judgment and evidence in the code or existing task artifact. Account for shared state, callers, cancellation, close ownership, timers, and synchronization identity. Complete when every story names an edge or a race and every blocking site has an unblock disposition. The race detector can reject a story; it cannot prove one. Use existing race, leak, synctest, and lint gates for mechanical evidence.

Load the review selector for fire-and-forget or blocked goroutines, publication or lock scope, unbounded fan-out or queues, and timer-driven coordination.