Asynchronous GPU compiler errors with graph provenance; authoring diagnostics.
Whether WgpuNodePipelineStore.dispose has run.
Distinct WGSL modules compiled so far.
Pre-sizes the frame (the sized-before-recording discipline): resolves —
and on first sight compiles — the program of every "node" item in
items, and grows the store's own strided uniform buffer to hold one
block per node draw per view. Returns whether any node draw may record
this frame. Runs material accessors (material.graph), so the
renderer re-checks its own disposal after calling it.
Destroys the store's buffers and drops everything else (§83). Idempotent.
Records one surface node draw into the frame's pass, or skips it on the frame's §61 terms (unresolvable texture, missing vertex stream, latched graph). Returns the stencil reference now in effect.
Uploads the frame's packed node uniform blocks — one writeBuffer,
beside the renderer's own, before the submit that reads them (queue
order). A frame that recorded no node draw uploads nothing.
Drops every reference without destroying anything — the device-loss path (§61): the allocations belong to a device that no longer exists.
Draws one §70 "graph" effect (RFC 0001) in its own pass and encoder —
the WGSL twin of GL's #renderGraphEffect, bound by renderEffect's
contract: everything that will not resolve skips the effect before
anything is recorded, and colorView is called only once the draw is
certain (an on-screen destination's getCurrentTexture must not be
acquired for an effect that then skips).
One renderer's node pipeline store — the registry interface implemented (
wgpu-node-registry.tsowns each member's contract). Created per device by the registered factory; forgotten on device loss, disposed with the renderer.