To pick up the latest bug fixes to ANGLE's Metal backend, let's roll ANGLE to top-of-tree as of 2022-03-21. Note that this depends on just-filed anglebug.com/7115 and https://chromium-review.googlesource.com/3537512 . (It was possible to patch the update-angle script to do this, but it's better to just pick up the proper fix from upstream.)
<rdar://problem/90639134>
Rolling to 2022-03-22 instead: e7f29440f02514126f30350dcd7132b68f7954d7 https://chromium.googlesource.com/angle/angle/+/e7f29440f02514126f30350dcd7132b68f7954d7
Created attachment 455404 [details] Patch
Note that there are important steps to take when updating ANGLE. See https://trac.webkit.org/wiki/UpdatingANGLE
Created attachment 455441 [details] Patch
Fixed CMake build in the latest patch.
Added Bug 238196 to the list that this blocks. Have confirmed on Bug 238118 that this roll definitely fixes more multisampled rendering bugs compared to WebKit ToT.
Comment on attachment 455441 [details] Patch Great work, thanks!
The layout test failures are definitely unrelated to this ANGLE roll. CQ'ing.
Committed r291757 (248788@main): <https://commits.webkit.org/248788@main> All reviewed patches have been landed. Closing bug and clearing flags on attachment 455441 [details].
Note: the EWS found the following failure with this patch: --- EWS has detected layout test failure on iOS-15-Simulator-WK2-Tests-EWS while testing Patch 455441 for Bug 238171. Full details are available at: https://ews-build.webkit.org/#/builders/68/builds/11320 Layout test failure: - editing/spelling/spellcheck-async-remove-frame.html (test history) Test history: https://results.webkit.org/?suite=layout-tests&test=editing/spelling/spellcheck-async-remove-frame.html --- I'm positive that this flake is unrelated to this patch - please don't revert this ANGLE roll in response to it. Thanks.
I've done a TOT build of WebKit and tested via the run-safari script, and still see the failures on some of the three examples with antialiasing, such as: https://threejs.org/examples/#webgl_materials_cubemap_dynamic where the background isn't shown. Tools/Scripts/build-webkit --release Tools/Scripts/run-safari Not sure how to verify that my built version is indeed being used, but I have checked that the files in my working copy do have the changes from this roll.
(In reply to Simon Taylor from comment #12) > I've done a TOT build of WebKit and tested via the run-safari script, and > still see the failures on some of the three examples with antialiasing, such > as: > https://threejs.org/examples/#webgl_materials_cubemap_dynamic where the > background isn't shown. I should have specified this is on my M1 Pro MBP on latest macOS 12.3.
Simon: there's a bug in the build products and run-safari script, and run-safari isn't picking up your locally built WebKit. Please use run-minibrowser.
This fix shipped with Safari 15.5 (all platforms).