Struct RenderSession 

pub struct RenderSession { /* private fields */ }

Persistent batch render session. See the module docs in render::RenderSession for lifetime, thread-affinity, and config-invariance guarantees. Persistent batch render session. Keeps a Bevy App (and its RenderDevice plus PSO cache) alive across multiple render() calls, amortizing per-episode cold-init cost.

Thread affinity

RenderSession must be created, used, and dropped on the same thread. It holds a bevy::App which owns GPU resources that are not safe to move across threads. The !Send + !Sync marker is enforced via PhantomData<*const ()>.

Config invariant

The RenderConfig (resolution, lighting, near/far, fov) is fixed at new(). All render() calls must use requests whose render_config matches; heterogeneous configs are rejected.

Phase 1 limitation

Each render() call must contain homogeneous requests (same object_dir and object_rotation). Heterogeneous calls return BatchRenderError::InvalidConfig. Hold a single RenderSession and call render() once per episode to amortize setup across episodes.

Implementations

impl RenderSession

pub fn new(render_config: &RenderConfig) -> Result<Self, RenderError>

Build the App, run plugin finish()/cleanup(), and perform one warmup update() so Startup systems run and the wgpu device + adapter are initialized. The first render() call still pays PSO compilation for the specific mesh/material combination; subsequent calls reuse the cache.

pub fn render( &mut self, requests: &[BatchRenderRequest], ) -> Result<Vec<BatchRenderOutput>, BatchRenderError>

Render a homogeneous batch of viewpoints (same object + rotation + config). Returns outputs in request order.

On BatchRenderError::DeviceLost, the returned error signals that the wgpu device was lost mid-render. This call produced no output; any outputs from earlier render() calls on this session are still valid. Recovery: drop this RenderSession and construct a new one.