mill.daemon

Logic around bootstrapping Mill, creating a MillBuildRootModule.BootstrapModule and compiling builds/meta-builds and classloading their RootModules so we can evaluate the requested tasks on the RootModule representing the user's build.mill file.

When Mill is run in daemon mode or with --watch, bootstrap state is cached in-memory across evaluations. The daemon-wide reusable bootstrap metadata lives in RunnerSharedState, while each launcher keeps its own evaluators and retained read leases in RunnerLauncherState.

When a subsequent evaluation happens, each meta-level attempts to re-use its corresponding cached frame to avoid unnecessary classloader recreation, worker churn, or re-evaluation. This should remain transparent to users while improving performance.

Per-launcher view of the current bootstrap/evaluation run.

Holds launcher-owned resources that cannot be shared daemon-wide: active evaluators, retained meta-build read leases that pin the shared frames the launcher is using, the user-task evaluation, and the launcher session.

Daemon-wide bootstrap cache shared across concurrent launcher runs.

Each stored frame represents reusable metadata from one level of build.mill evaluation: watches, classloaders, code signatures, classpaths, and worker caches that are safe to share between launchers. A frame whose bootstrap succeeded carries a RunnerSharedState.Frame.Reusable payload (which now also owns the per-classloader worker cache, so worker lifetime tracks classloader lifetime); a failed frame still publishes its watches so callers can use them to invalidate.

Logic around the "watch and wait" functionality in Mill: re-run on change, re-run when the user presses Enter, printing status messages, etc.