errors in webgpu:shader,execution,expression,constructor,non_zero:*

[kainino0x]

Example: webgpu:shader,execution,expression,constructor,non_zero:abstract_array_elements:abstract_type="abstract-int";concrete_type="f32";length=1

EXCEPTION: Error: Unexpected validation error occurred: Error while parsing WGSL: :22:22 error: cannot assign 'array<i32, 1>' to 'array<f32, 1>'

Found in CTS->Dawn roll. It sounds like a test bug, I don't think this is supposed to be valid.

All of the following are failing with similar errors locally (M1 Mac, Chrome 125.0.6415.0 (Official Build) canary (arm64)):

webgpu:shader,execution,expression,constructor,non_zero:abstract_vector_splat:*
webgpu:shader,execution,expression,constructor,non_zero:abstract_vector_elements:*
webgpu:shader,execution,expression,constructor,non_zero:abstract_vector_mix:*
webgpu:shader,execution,expression,constructor,non_zero:abstract_matrix_elements:*
webgpu:shader,execution,expression,constructor,non_zero:abstract_matrix_column_vectors:*
webgpu:shader,execution,expression,constructor,non_zero:abstract_array_elements:*

From #3631

[kainino0x]

Tangentially(?) this test is unexpectedly passing on most configurations on Chrome's CQ: https://crbug.com/dawn/2514

[kainino0x]

IIUC fixed in #3659? @ben-clayton

[ben-clayton]

I believe so, but we might want to wait until the next roll confirms.

[kainino0x]

The roll did not remove the expectations: https://dawn-review.googlesource.com/c/dawn/+/184140/4/webgpu-cts/expectations.txt#1483

And locally they still fail with unroll_const_eval_loops disabled (and pass with it enabled).

[ben-clayton]

Argh. Okay. I'll take another look tomorrow. Sorry about that.

[kainino0x]

Thanks!

[ben-clayton]

Fixed.