Module render_plan 

Compiled, immutable audio render plan — Phase 2 of LOCKLESS.md.

A RenderPlan is the flattened, topologically ordered form of the engine’s per-cycle task graph plus the AudioIO port network. It is built on the control thread (see the builder contract in LOCKLESS.md), published with arc_swap, and executed by the cycle executor with count-up dependency counters. Nothing in it is behind a mutex: while a plan is alive it never mutates.

Key invariants (checked by RenderPlan::verify after every compile):

• Single-producer chains. Every arena buffer has at least one writer, and multiple writers of the same buffer always form a dependency chain (e.g. an input Sum node writes a track input, then the track task adds clip audio in place). Two nodes that could run concurrently never write the same buffer.
• Topological order. Except for forced nodes (feedback loops, broken deliberately — the same situation today’s finished-flag scheduler resolves with its force-progress fallback), every edge points from a lower node index to a higher one, so sequential execution is a plain iteration and the arena can be split safely at each output index.

Structs

RenderPlan

A compiled, immutable render plan. Owns the whole buffer arena.

Enums

Op

What a plan node does when executed.

Type Aliases

BufferId

Index into RenderPlan::buffers.

NodeId

Index into RenderPlan::nodes.

PlanSlot

The atomically published plan slot. The dispatcher pulls with load_full() at cycle start; the builder stores newly compiled plans.

SharedPlan

A plan shared across the dispatcher, workers and hardware drivers.