Update to wgpu-native v0.14.2.3

[almarklein]

From scanning the changes in the .h files:

• Do we use errorfilter en errortype somewhere? It has a new "internal" field.
• Default limits :/
• Handling of features.
• Buffer getsize and getusage
• Texture getheight and a lot more.
• WGPUProcComputePassEncoderDispatch -> WGPUProcComputePassEncoderDispatchWorkgroups
• same for WGPUProcComputePassEncoderDispatchIndirect
• Use wgpuFree. Not for logging. But needed for result of wgpuSurfaceGetSupportedFormats and wgpuSurfaceGetSupportedPresentModes.
• Logcallback now has userdata
• wgpuRenderPassEncoderDrop, wgpuSurfaceDrop, and more drop functions.

Other tasks:

• Make sure that the examples work.
• Make sure that the tests pass.
• Make sure that the pygfx tests pass locally too.
• Re-release wgpu-native for the rows_per_image-hotfix.
• Bump to the new wgpu-native.
• Document changes in the release notes.

Copy from the changelog:

• To use the default min_binding_size in create_bind_group_layout, it should be None instead of zero.
• If the depth-stencil texture has not room for stencil data, the stencil_read_mask and stencil_write_mask fields in the DepthStencilState struct passed to create_render_pipeline() must be set to 0.
• In WGSL, @stage(compute) must now be @compute. Same for vertex and fragment.
• In WGSL, the list of reserved words has been extended, including e.g. mod, matrix and ref.
• In WGSL, smoothStep is now smoothstep.

[Korijn]

I went over your work so far and didn't see anything out of the ordinary.

[almarklein]

This is close to finished, but I'm stuck on an error related to uploading 3D texture data. I'm starting to think this might be a bug in wgpu-core.

thread '<unnamed>' panicked at 'Unable to write texture: Transfer(BufferOverrun { start_offset: 0, end_offset: 329728, buffer_size: 12288, side: Source })', src/device.rs:743:6

See gfx-rs/wgpu-native#237

[almarklein]

Having some trouble getting it to work in pygfx, related to validations failing for depth-stencil:

thread '<unnamed>' panicked at 'Unable to end render pass:
RenderPassError { scope: SetPipelineRender((12, 1, Metal)), inner: RenderCommand(IncompatiblePipelineRods) }',
src/command.rs:298:10

[almarklein]

PyGfx is up and running! 🚀 But ... Sigh ... now the wheel building fails: "This does not look like a platform wheel"

[almarklein]

Nevermind, fixed it.

[almarklein]

Ready from my end!

[Korijn]

PyGfx is up and running! 🚀 But ... Sigh ... now the wheel building fails: "This does not look like a platform wheel"

So, it didn't fail silently! That's a good thing. :) Good job!

[almarklein]

So, it didn't fail silently! That's a good thing. :)

Well, the Linux build was the only one that failed. For some reason the Windows and MacOS build passed, even though they suffered from the same problem.