[github-actions]

This PR was opened by the Changesets release GitHub action. When you're ready to do a release, you can merge this and the packages will be published to npm automatically. If you're not ready to do a release yet, that's fine, whenever you add more changesets to main, this PR will be updated.

Releases

effect-machine@0.31.0

Minor Changes

• 1e2966a Thanks @cevr! - actor.subscribe, actor.client.subscribe and self.client.subscribe listeners now get every state written after they subscribe, each once, in write order, like the waits. A listener reads the states from the state ref's own publication, and a notification only wakes it. Before this change a supervised restart could deliver its restart state after a newer state of the new generation, a send from a listener callback could deliver the newer state to later listeners before the older one and call a listener again inside its own callback, a state whose entry crashed before its notification was never delivered, and a stop of an actor could drop the last write. Behavior changes: a listener that subscribes between a write and its notification, or inside another listener's callback, no longer gets that state (it was current when it subscribed; read getSnapshot for it). This includes a listener that a .spawn handler subscribes: it no longer gets the state that the handler runs for (0.30.0 delivered it); read the handler's state or getSnapshot() for it; a listener is not called again while its own callback runs, and gets the states its callback caused after it returns; the listeners of an actor get a write that a stop overtook before awaitExit completes; each subscribe call is its own registration, so the same function subscribed twice gets each write twice (the listeners were a set of functions before, which called it once), and each unsubscribe ends one registration. The ordering and terminal guarantees cover local actor listeners. An entity's self.client listener can still miss a write that the entity's deactivation overtakes, as in 0.30.0: the entity runtime does not ring its listeners at teardown.

• 1469452 Thanks @cevr! - actor.waitFor, actor.awaitFinal and actor.sendAndWait now end when the actor terminates. They fail with ActorStoppedError (the error awaitOutput already uses) when the actor stops, drains, defects, or reaches a final state, and no state from the start of the wait matched. Before this change a wait never resolved after the actor terminated without a match, so callers had to race it against awaitExit or a time limit. The waits now read the states from the state ref's own publication: the predicate sees the snapshot and then every later state write, each once, in write order and one call at a time, and the wait returns the first match. A brief state around the start of the wait is no longer lost, a state counts as reached once it is written (also when its entry crashes later, when a supervised restart follows, or when a stop comes before its notification), the predicate is not called again after it matches, and a predicate that sends an event cannot re-enter the wait. sendAndWait starts to observe before it sends, and does not send when the predicate throws on the snapshot. A predicate that throws on a later state now fails the wait with that defect instead of being ignored. A state restored by recovery now notifies subscribe listeners and waiters. A wait releases its listener and its subscription when it ends, also when it is interrupted at any point. The error channel of all three widens from never to ActorStoppedError: code that annotates them as Effect<State> must handle or declare the new error.