Bug 306670 – [MacOS Debug]http/tests/webgpu/webgpu/api/validation/render_pipeline/resource_compatibility.html times out due to combinatorial complexity

• Status: NEW
• Component: WebGPU

[Diego De La Toba]

http/tests/webgpu/webgpu/api/validation/render_pipeline/resource_compatibility.html is a flaky text failure on MacOS Debug arm64.


HISTORY:
https://results.webkit.org/?suite=layout-tests&test=http%2Ftests%2Fwebgpu%2Fwebgpu%2Fapi%2Fvalidation%2Frender_pipeline%2Fresource_compatibility.html


DIFF URL:
https://build.webkit.org/results/Apple-Tahoe-Debug-AppleSilicon-WK2-Tests/306492%40main%20(673)/http/tests/webgpu/webgpu/api/validation/render_pipeline/resource_compatibility-pretty-diff.html


REPRODUCTION: 
Reproduced on macOS Tahoe Debug Tot with the following:

run-webkit-tests --debug --force --no-retry --iterations 5 http/tests/webgpu/webgpu/api/validation/render_pipeline/resource_compatibility.html

I am going to mark expectations as pass fail while this pends investigation.

[Radar WebKit Bug Importer]

<rdar://problem/169316951>

[Diego De La Toba]

Pull request: https://github.com/WebKit/WebKit/pull/57605

[EWS]

Test gardening commit 306536@main (84a28d97f8a3): <https://commits.webkit.org/306536@main>

Reviewed commits have been landed. Closing PR #57605 and removing active labels.

[Alexey Proskuryakov]

This test usually times out in debug (it takes 20+ seconds even in release), but yes, it sometimes fails with a text diff too. Seems to be a pretty old issue - I looked as far as August 2025, and it was happening then.

[Mike Wyrzykowski]

Text diff seems to be a timeout too, e.g., https://build.webkit.org/results/Apple-Tahoe-Debug-AppleSilicon-WK2-Tests/306530@main%20(675)/http/tests/webgpu/webgpu/api/validation/render_pipeline/resource_compatibility-pretty-diff.html:

- EXCEPTION: Error: endTestScope timed out

[Jonathan Bedard]

Seems pretty straight-forward, then. We need to break this test up.

[Karl Rackler]

Adding Timeout to test expectation

[EWS]

Test gardening commit 307796@main (67606735a7d5): <https://commits.webkit.org/307796@main>

Reviewed commits have been landed. Closing PR #58968 and removing active labels.

[Karl Rackler]

TIMEOUT
This consistent timeout goes back to Sequoia and range https://commits.webkit.org/compare/306551@main...306488@main 
Possibly: 306543@main by Chris. The GPURenderPassEncoder changes are the most direct match to a render pipeline resource compatibility test timing out.

[Mike Wyrzykowski]

I don't understand the difference to breaking this test up, duplicating much of its logic, and specifying a longer timeout?

These tests take very long to run due to testing each combination of possible options. We could test a subset but we'd lose coverage. 

Maybe a subset is more appropriate for EWS and leave the entire test for post-commit?