2006984 - Tighten our expectations for WPT position-anchor-target-with-children.html to expect a fuzzy pass rather than intermittent failure

Status: RESOLVED DUPLICATE of bug 2007368

(Core :: Layout: Positioned, task)

Description

[Daniel Holbert [:dholbert]]

As noted in bug 1923151 comment 1, we have an unnecessarily permissive annotation on WPT position-anchor-target-with-children.html right now:

https://searchfox.org/firefox-main/rev/d47d53b6f23c510e9c0a7219fe4b99f9e063644a/testing/web-platform/meta/css/css-anchor-position/position-anchor-target-with-children.html.ini#1-2

[position-anchor-target-with-children.html]
  expected: [PASS, FAIL]

This annotation would let it completely break without turning treeherder orange. So let's tighten that to a fuzzy expectation, to have better protections against breakage/regressions.

(For me at least, the test renders consistently, just needing a little bit of fuzz around rounded bullets.)

Comment 1

[Daniel Holbert [:dholbert]]

Try run, to annotate my own local fuzzy-failure needs - let's see if any of our treeherder platforms need more than that:
https://treeherder.mozilla.org/jobs?repo=try&landoCommitID=169929

Comment 2

[Daniel Holbert [:dholbert]]

Attachment: Bug 2006984: Adjust failure annotation for position-anchor-target-with-children.html to be fuzzy rather than intermittent. r?#layout

Comment 3

[Daniel Holbert [:dholbert]]

Attachment: Bug 2006984: Remove apparently-unnecessary intermittent-failure allowance for WPT position-anchor-004.html. r?#layout

Comment 4

[Daniel Holbert [:dholbert]]

Attachment: fuzzy test-failure-log on dholbert's linux machine, for position-anchor-target-with-children.html

For the record, here's the fuzzy-failure that I get in position-anchor-target-with-children.html on my own linux ThinkStation (with "Found 84 pixels different, maximum difference per channel 11", differing pixels all being antialiasing around 3 of the 4 list-bullets).

Comment 5

[Daniel Holbert [:dholbert]]

Looks like I forgot to land this; but tnikkel ended up doing more substantial cleanup (including removing both .ini files in question here) in bug 2007368, so I'll dupe this to that bug.