Memory handling scheme for render buffers

[Ichthyostega]

As a Lumiera developer,
I need a consolidated scheme for dealing with temporary and output buffers,
to be able to pass invocation from a Render Job into the Render Nodes.

consolidation of framework drafts

Several years ago, as part of the initial (stalled) efforts to build a Render Node invocation, and later the (still valid) foundation of the Player subsystem, two similar and partially overlapping handling protocols were defined:

• Buffer Provider Protocol
• Output Slot Protocol

The latter even re-uses a concept introduced for the former: the BuffHandle front-end to be used by client code.

The integration efforts for the »Playback Vertical Slice« however uncovered some discrepancies in the intended usage of these protocols: notably the OutputSlot provides a DataSink and expects that the client invokes DataSink::emit(FrameID, BuffHandle) when data is ready for output. Yet the Buffer Provider Protocol defines another function BuffHandle::emit() — and obviously the code in the Render Nodes will invoke the latter, since it can not even know when a given BuffHandle represents an output.

Another shortcoming unveiled by the integration effort is the very much prototype-y nature of the default implementation, which unfortunately was tied-in much too tightly with the interfaces. Notably, the BufferProvider relies directly on a BufferMetadata table attached as an implementation component to track the buffer's state transitions — which makes usage in a highly concurrent environment quite challenging.

The Concepts and protocols as such seem valid and should be retained; the task is rather to loosen the overly tight coupling to the default implementation and to change the interface design in a way to make both protocols better composable.

[Ichthyostega]

Attach to parent #1412 (clarify the role of Buffer Types).

[Ichthyostega]

Add subticket #1429 (control effort for global buffer allocator).

[Ichthyostega]

Add #1428 + #1430

(Render-buffer pools).