Safe Points

Safe points are the only places where a staged hot reload update can become the current program version. Vela does not interrupt arbitrary bytecode instructions to replace function bodies.

Purpose

Safe points give the host a predictable boundary for code replacement. They prevent partially executed functions from observing a different instruction stream or registry halfway through a call.

Common host safe points include:

end of event
tick boundary
between queued jobs
explicit runtime.check_reload()

Old And New Calls

When a safe point accepts an update, only future calls use the new version. Active calls keep their old CodeObject and old metadata snapshot until they return.

This means a long-running event handler can finish with the code it started with, while the next event handler uses the updated code.

Host Responsibilities

The host should place safe points at boundaries where repeated work naturally returns control to Rust. It should not expose unbudgeted infinite script loops; execution budgets still apply while old frames are running.

Debugging And Reporting

Safe-point reports identify whether a staged update was applied, rejected, or not present. Hosts should log these reports with source labels so operators can connect a rejection to the file or deployment that produced it.