1575338 - Underlines intermittently render without ink-skipping, during reftest runs

Status: RESOLVED FIXED

(Core :: Layout: Text and Fonts, defect, P3)

Description

[Daniel Holbert [:dholbert]]

For some tests, underlines occasionally render without ink-skipping, despite the fact that we've enabled text-decoration-skip-ink by default (and its default value causes ink-skipping). This is causing intermittent test-failures.

For now we're handling this by adding the CSS text-decoration-skip-ink:none in these tests (in https://hg.mozilla.org/mozilla-central/rev/97d2c2478643 and in commits on subsequent intermittent bugs). I'm filing this bug here to cover the underlying issue (which we don't really understand at this point).

Comment 1

[Daniel Holbert [:dholbert]]

Attachment: Bug 1575338: Turn off text-decoration-skip-ink for WPT auto-margins-001.html. r?jfkthame

This test's reference case occasionally (intermittetly) fails to do ink
skipping, so let's just turn off the ink-skipping preference for now to
avoid triggering too much intermittent orange.

Comment 2

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Seems like we shouldn't be turning on new features that add new non-deterministic behavior like this. I'd think we should figure this out before saying this feature is ready to ship.

Comment 3

[Jonathan Kew [:jfkthame]]

Yes, the intermittent nature of the issue is a bit worrying. We need to understand what's happening here.

Comment 4

[Daniel Holbert [:dholbert]]

This would be a good candidate for a Pernosco investigation, given that it's reproducible on linux.

Comment 5

[Daniel Holbert [:dholbert]]

Or rather, bug 1574031 is probably a good candidate (it's more frequent so probably easier to catch).

Comment 6

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/243643e83bfc
Turn off text-decoration-skip-ink for WPT auto-margins-001.html. r=jfkthame

Comment 7

[Daniel Holbert [:dholbert]]

oops, the patch that I just landed (annotating auto-margins-001.html) was meant for the more-specific bug 1574306. Bug number copypaste error on my part, noticed slightly too late.

No big deal; adding leave-open so we can keep this bug open & tracking the underlying cause.

Comment 8

[Alexandru Michis [:malexandru]]

https://hg.mozilla.org/mozilla-central/rev/243643e83bfc

Comment 9

[Jonathan Kew [:jfkthame]]

(In reply to Daniel Holbert [:dholbert] from comment #4)

This would be a good candidate for a Pernosco investigation, given that it's reproducible on linux.

Has this been seen on debug jobs, or is it opt-only? I can't seem to find any debug failures.... and I tried retriggering a bunch of test runs on https://treeherder.mozilla.org/#/jobs?repo=try&revision=c32707702c8a155ea8424b85298944c6adb14067&selectedJob=262633927 (which does not include the patch here to disable skip-ink), but so far I haven't seen a debug failure, which AIUI means there's nothing for pernosco to get hold of. :(

Comment 10

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Seems like it might be worth running one of the failing tests under valgrind to see if it reports reading uninitialized memory.

Comment 11

[David Baron :dbaron: (⌚️UTC-4, no longer working on Mozilla)]

Maybe related to bug 1574066 comment 7?

Comment 12

[Cameron McCormack (:heycam)]

Daniel, with bug 1574066 landed (well, it's on autoland), could you do a try run re-enabling skip-ink in those tests you disabled it for, and retrigger them a bunch of times to see if they still intermittently fail?

Comment 13

[Daniel Holbert [:dholbert]]

Yup, I'll do that. Thanks!

Comment 14

[Daniel Holbert [:dholbert]]

https://treeherder.mozilla.org/#/jobs?repo=try&revision=0dda703eb537712dbfe9ca023fbbdfcd84ebe58f

Comment 15

[Daniel Holbert [:dholbert]]

Filtered Try run links, showing formerly-affected Linux testrun jobs that contain the tests where we'd previously disabled ink skipping:
https://treeherder.mozilla.org/#/jobs?repo=try&revision=0dda703eb537712dbfe9ca023fbbdfcd84ebe58f&searchStr=linux%2Cx64%2Copt%2Cru3
https://treeherder.mozilla.org/#/jobs?repo=try&revision=0dda703eb537712dbfe9ca023fbbdfcd84ebe58f&searchStr=linux%2Cwr4

Looking good, so I'm going to proceed with reenabling ink-skipping for those tests.

(Note that we still have ink-skipping disabled in some tests as of its initial landing, due to bug 1573723, but that bug is less worrisome because it's a deterministic, subtle fuzziness issue rather than an ink-not-being-skipped-at-all issue.)

Comment 16

[Daniel Holbert [:dholbert]]

Attachment: Bug 1575338 followup: Backed out changeset 243643e83bfc to reenable ink-skipping in WPT auto-margins-001.html.ini. (no review, just un-annotating a test)

We are pretty sure that this test no longer needs ink skipping to be disabled,
as of bug 1574066.

Comment 17

[Pulsebot]

Pushed by [e-mail]:
https://hg.mozilla.org/integration/autoland/rev/68447f87ca05
followup: Backed out changeset 243643e83bfc to reenable ink-skipping in WPT auto-margins-001.html.ini. (no review, just un-annotating a test)

Comment 18

[Stefan Hindli [:stefan_hindli]]

https://hg.mozilla.org/mozilla-central/rev/68447f87ca05

Comment 19

[Intermittent Failures Robot]

1 failures in 4530 pushes (0.0 failures/push) were associated with this bug in the last 7 days.

Repository breakdown:

• mozilla-central: 1

Platform breakdown:

• windows7-32: 1

For more details, see:
https://treeherder.mozilla.org/intermittent-failures.html#/bugdetails?bug=1575338&startday=2019-09-23&endday=2019-09-29&tree=all