fix cts_runner

[turbocrime]

Connections

#6839

plus two more issues, revealed after the above issue was fixed.

Description

cts_runner could not execute conformance tests due to broken deno state management.

this pr restores the functionality of cts_runner, by

• fix panic in compute_pass and render_pass by using state.borrow::<super::Instance>() instead of state.borrow::<wgpu_core::global::Global>(), which is not present
• fix panic in cts_runner main by providing Permissions struct to js_runtime state
• fix test failure due to incorrect ser/de of ColorWrites

Testing

cts_runner can now be manually executed.

i've been using a script to execute tests, adapted from the deleted cts.yml workflow

cts_runner/cts.sh

#!/bin/sh

set -euo pipefail

if [[ "$PWD/cts_runner" != "$(dirname -- "$(readlink -f -- "$0")")" ]]; then
  echo "Run this script from the repository root."
  exit 1
fi

cd cts

if [[ "$(git rev-parse HEAD)" != "$(cat ../cts_runner/revision.txt)" ]]; then
  echo "CTS revision mismatch."
  cat ../cts_runner/revision.txt
  exit 2
fi

grep -v '^//' ../cts_runner/test.lst | while IFS=$' \t\r\n' read test; do
  echo "=== Running $test ==="
  cargo run --manifest-path ../cts_runner/Cargo.toml --frozen -- ./tools/run_deno --verbose "$test"
done

Checklist

• Run cargo fmt.
• Run taplo format.
• Run cargo clippy. If applicable, add:
  • --target wasm32-unknown-unknown
  • --target wasm32-unknown-emscripten
• Run cargo xtask test to run tests.
• Add change to CHANGELOG.md. See simple instructions inside file.

[FrankenApps]

Please fix your commits to only include the changed lines, otherwise this is impossible to review.

[turbocrime]

Please fix your commits to only include the changed lines, otherwise this is impossible to review.

sorry for messy diff - still drafting pr. will mark as 'ready' soon

[turbocrime]

permissions must be available in state before bootstrap. for cts_runner it's static, so i moved it here into the extension macro.

[turbocrime]

super::Instance wraps a wgpu_core::global::Global and is also actually present in the runtime state. it's used in locations that continued to pass tests after 77a83fb and using it consistently seems to correct the test panics.

[turbocrime]

several op functions in command_encoder, render_pass, and compute_pass use some gfx_ok and gfx_put macros.

i didn't add or remove any calls, but i noticed use seems to have become more inconsistent over time.

[turbocrime]

the bitflags macro and serde macros seem to be sensitive to ordering.

previously, the generated deserializer would expect a javascript string, but it's clearly a numeric type.

[turbocrime]

this parameter is unused in this function, but it is part of the denoland implementation of deno_webgpu, and it's expected to be present by 01_webgpu.js in this repository

wgpu/deno_webgpu/01_webgpu.js

Lines 3867 to 3870 in 61b7063

	const { err } = op_webgpu_render_pass_end(
	commandEncoderRid,
	renderPassRid,
	);

[turbocrime]

this parameter is unused in this function, but it is part of the denoland implementation of deno_webgpu, and it's expected to be present by 01_webgpu.js in this repository

wgpu/deno_webgpu/01_webgpu.js

Lines 4383 to 4386 in 61b7063

	const { err } = op_webgpu_compute_pass_end(
	commandEncoderRid,
	computePassRid,
	);

[turbocrime]

uncertain how to add this to the changelog.

by intent, this is just fixing internal tools.

practically, i think the state management changes and the parameter changes probably fixed some real outwards issues, but i haven't looked for examples.

[cwfitzgerald]

You can add it to the testing/internal category as we don't know of any direct problems this solves.

[cwfitzgerald]

(Approved wgpu side)