Bug 280490
Summary: | REGRESSION (284225@main) : [ macOS iOS ] 2x imported/w3c/web-platform-tests/mathml/presentation-markup/operators/* are constant failures | ||
---|---|---|---|
Product: | WebKit | Reporter: | Anfernee Viduya <aviduya> |
Component: | MathML | Assignee: | Nobody <webkit-unassigned> |
Status: | NEW | ||
Severity: | Normal | CC: | fred.wang, harpsichen, ntim, webkit-bot-watchers-bugzilla, webkit-bug-importer |
Priority: | P2 | Keywords: | InRadar |
Version: | WebKit Nightly Build | ||
Hardware: | Unspecified | ||
OS: | Unspecified |
Anfernee Viduya
imported/w3c/web-platform-tests/mathml/presentation-markup/operators/size-and-position-of-stretchy-fences-with-default-font-001.html
imported/w3c/web-platform-tests/mathml/presentation-markup/operators/stretchy-largeop-with-default-font-2.html
are constant failures
HISTORY:
https://results.webkit.org/?suite=layout-tests&suite=layout-tests&test=imported%2Fw3c%2Fweb-platform-tests%2Fmathml%2Fpresentation-markup%2Foperators%2Fsize-and-position-of-stretchy-fences-with-default-font-001.html&test=imported%2Fw3c%2Fweb-platform-tests%2Fmathml%2Fpresentation-markup%2Foperators%2Fstretchy-largeop-with-default-font-2.html
DIFF:
imported/w3c/web-platform-tests/mathml/presentation-markup/operators/size-and-position-of-stretchy-fences-with-default-font-001.html
-PASS Inner binary operator should not affect position and size of outer fences.
-PASS Fences are stretched symmetrically with respect to the math axis
+FAIL Inner binary operator should not affect position and size of outer fences. assert_approx_equals: expected 42.640625 +/- 2 but got 39.796875
+FAIL Fences are stretched symmetrically with respect to the math axis assert_approx_equals: expected 71.640625 +/- 2 but got 68.1171875
(
(
x
imported/w3c/web-platform-tests/mathml/presentation-markup/operators/stretchy-largeop-with-default-font-2.html
-FAIL Test that non-stretchy largeops are bigger in display mode. assert_greater_than: width of largeop '⨇' expected a number greater than 6.640625 but got 2.640625
+FAIL Test that non-stretchy largeops are bigger in display mode. assert_greater_than: width of largeop '⫼' expected a number greater than 13.28125 but got 13.265625
DIFF LINK:
https://build.webkit.org/results/Apple-Sequoia-Release-AppleSilicon-WK2-Tests/284313%40main%20(83)/imported/w3c/web-platform-tests/mathml/presentation-markup/operators/size-and-position-of-stretchy-fences-with-default-font-001-diff.txt
https://build.webkit.org/results/Apple-Sequoia-Release-AppleSilicon-WK2-Tests/284313%40main%20(83)/imported/w3c/web-platform-tests/mathml/presentation-markup/operators/stretchy-largeop-with-default-font-2-diff.txt
REPRODUCIBILITY:
I was able to reproduce the issue using command.
run-webkit-tests <test_name> --iterations=10
DESCRIPTION:
In 284225@main changes were made in MathML layout.
Attachments | ||
---|---|---|
Add attachment proposed patch, testcase, etc. |
Radar WebKit Bug Importer
<rdar://problem/136794624>
EWS
Test gardening commit 284326@main (cf4c150d7ff5): <https://commits.webkit.org/284326@main>
Reviewed commits have been landed. Closing PR #34339 and removing active labels.
Frédéric Wang (:fredw)
These two tests involve operator stretching with the default math fonts available on the system, so the expectations should really be platform-dependent.
However, they were initially imported in https://commits.webkit.org/282098@main with a single platform-independent test `-expected.txt` files. They had the flaky expectation "ImageOnlyFailure Pass" (now removed), which does not really make sense here since these are testharness.js tests, not visual reftests. But probably that's a hint they were actually platform-dependent.
In https://commits.webkit.org/284225@main the `-expected.txt` files were adjusted. But this patch is unrelated to stretchy operators or fonts and the tests unrelated to OOF children, so I don't think the change should have affected the tests. What I believe happened is that expectations on another platform were taken instead as a reference. Testing on macOS Sequoia, I actually still get the very same expectations as before the patch.
So I think the right way to fix that is just to add platform-dependent expectations for these tests, as it should have been the case when they were initially imported. I'll talk to Harry about it.