Changelog

[0.3.1] - 2026-09-29

use Solve now generates an overridable child_spec/1, allowing apps to be added directly to supervision trees as permanent workers.
Default child IDs include the app module and configured name, allowing differently named instances of the same app under one supervisor.

Automatically restore Lookup subscriptions after outages and named app restarts, using owner-scoped watching and capped exponential backoff.
Solve.Lookup.status/1 to read cached connection status and the optional handle_solve_connection_changed/3 callback for lifecycle-aware consumers.
Configurable finite cold-acquisition timeouts and recovery backoff through :lookup_timeout and :lookup_recovery.

Warm remote Lookup reads, pushed updates, and cleanup no longer make synchronous liveness or name-resolution RPCs; they use owned process monitors instead.
Remote names stay pinned to their acquired app until invalidation and reacquisition.
Lookup retains subscription intent across outages. Solve.Lookup.unsubscribe/1,2 cancels pending recovery as well as active subscriptions; cleanup and raw unsubscribe do not cancel recovery intent.
Migration: Manual and helper-mode consumers must forward owned DOWN messages and private {:solve_lookup, kind, payload} recovery messages to Solve.Lookup.handle_message/1. Auto mode handles these messages automatically. See manual message handling.
Refresh development dependencies and Elixir/OTP and CI tooling while retaining Elixir 1.18 support.

Correct global atom-key name classification.
Fence recovery and cancellation races so stale work cannot restore canceled subscriptions.
Notify data consumers after fresh snapshots and event routes are installed, even when recovered values are unchanged.

Solve.unsubscribe/2 and /3 to stop receiving raw updates without stopping the controller.
Solve.Lookup.unsubscribe/1 and /2 to release a process-local lookup subscription and its cached value. A later lookup subscribes again.
Solve.Collection.new/1 to build an ordered collection from {id, value} pairs.
Solve.Lookup.cleanup/0 to clear cached data for app instances that have stopped.

Breaking: Explicit app dispatch requires Solve.dispatch(app, target, event, payload). Pass %{} when no payload is needed; implicit /2 and /3 remain available in controller context.
Breaking: Acquire lookup targets through solve or collection before forwarding updates. Solve.Lookup.handle_message/1 no longer creates subscriptions or seeds the cache from unsolicited or versionless updates. Forward complete runtime envelopes rather than rebuilding them from values.
Controller startup now times out after five seconds by default. Set the app's :controller_start_timeout option for slower initialization.
Reduce overhead of repeated cached lookup reads.

Refresh lookup values and event handlers correctly after controller replacement or app restart.
Keep collection dependencies consistent during item changes, reordering, and replacement.
Correct collection handling of nil values and distinct numeric keys such as 1 and 1.0; reject invalid reorder operations.
Avoid restarting slow controllers when a subscription times out.
Clean up controller processes when their app stops or initialization fails.
Reject invalid controller dependency definitions with clearer errors.

Resubscribe dependent controllers when a dependency controller is replaced after a params change, instead of leaving them attached to the stopped process and missing all further updates. Applies to both singleton dependencies and collection items replaced under the same id.

Add Solve-style controller handle_info fallback for non-Solve messages.
Allow controller handle_info arities from /2 through /5, with access to state, dependencies, callbacks, and init params.
Document controller handle_info usage in the README and module docs.

Keep Solve internal messages reserved from controller-defined handle_info clauses.
Validate controller handle_info arity at compile time.