James Liu

This looks generally pretty sound in terms of approach, but I'd like to see some backing numbers for this to show the impact.

IMO we should *not* be serializing window state in scenes. This feels like a massive footgun waiting to happen.

This is a bit of a tough one. There's so much logic in the scheduler now that replicating everything for it is now no simple task. This is why we...

#12398 should implement `WorldQuery` and `QueryData` for `EntityLocation`, which is just missing the generation from the entity, which should be accessible already.

The fact that accesskit now uses it's own async executor (https://github.com/AccessKit/accesskit/pull/337) means that a tokio or async-std runtime is now necessary to be bundled on Unix platforms. This also imposes...

Is there a reason why this cannot be done as an ecosystem crate? It seems the plugin level support on all of these other engines seem to operate just fine.

> Additionally, each `batch_and_prepare_render_phase` cannot run in parallel with each other due to `ResMut`. This is potentially avoidable by splitting the resource based on phase, though I'm not sure how...

When trying to accelerate `encase` encoding, I found that the memory copy and encoding costs aren't the bottleneck in the batching systems. It may come down to the fact that...

> Improve MeshUniform inverse matrix calculation performance This may be supplanted on platforms where compute shaders are present by #12773. > Faster MeshUniform serialization https://github.com/teoxoy/encase/pull/65 should make `encase` as fast...

Does the `Limits` and `Features` returned by the device state it has compute shader and storage buffer support? @pcwalton was operating under the assumption that if a device supports one,...