Bug 33835 – [Gtk] svg/custom/use-instanceRoot-event-bubbling.xhtml fails

• Status: RESOLVED
• Resolution: FIXED
• Component: SVG

[Xan Lopez]

Parts of its state machine are constructed with very quick timeouts (100ms), which is a good way of making a test flakey :) (and, for instance, fails in the GTK+ bots).

[Eric Seidel (no email)]

http://trac.webkit.org/browser/trunk/LayoutTests/svg/custom/use-instanceRoot-event-bubbling.xhtml
Added by WildFox 15 months ago.

Next time I see it fail, I'll link to the failure diff.

[Nikolas Zimmermann]

This is true, the test is flakey as all <use> tests are, because of a problem with instanceRoot() - if the renderer is not attached yet, we're just returning 0, instead of asking the document to update style.

This is only one part of the issue, there is a race condition with event listeners, that currently still cause shadow tree reclones, I've fixed all these issues, and stability seems fine:

run-webkit-tests --repeat-each 50 -p `ls -l use-*.svg | sed -e "s/.*\ use/svg\/custom\/use/" | xargs`
Testing 49 test cases, repeating each test 50 times.
svg/custom ..................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................................
100.25s total testing time

all 2450 test cases succeeded

[Eric Seidel (no email)]

So you have a patch to fix this? Or one was already landed?  In which case, the revision link would be useful. :)

[Nikolas Zimmermann]

Created attachment 46969 [details]
Initial patch

Ran run-webkit-tests -p svg --tolerance 0, no new regressions. Ran all use* testcases with --repeat-each 50 -p, saw no leaks and/or timing issues.
Reenabling all affected testcases on all platforms. Let's hope <use> is stable now, when scripted from JS.

This is also a huge performance benefit when working with <use> + event listeners.

[Nikolas Zimmermann]

*** Bug 32519 has been marked as a duplicate of this bug. ***

[Eric Seidel (no email)]

Attachment 46969 [details] did not build on mac:
Build output: http://webkit-commit-queue.appspot.com/results/200841

[Nikolas Zimmermann]

Created attachment 46970 [details]
Updated patch

Fix release builds, thanks to ews :-)

[Eric Seidel (no email)]

Attachment 46970 [details] did not build on mac:
Build output: http://webkit-commit-queue.appspot.com/results/201057

[Nikolas Zimmermann]

Created attachment 46973 [details]
Updated patch v2

Damn, got it wrong - ASSERT_UNUSED requires 2 params - fixed.

[Nikolas Zimmermann]

Created attachment 47053 [details]
Updated patch v3

Fix issues ap mentioned on IRC.

[Alexey Proskuryakov]

Comment on attachment 47053 [details]
Updated patch v3

r=me. I didn't review SVG parts very deeply.

>     // creation, as we can't wait for the lazy creation to happen if ie. JS wants to

This comment is old - but it sounds like that should be "e.g.", not "i.e."

[Nikolas Zimmermann]

Landed in r53564.

[Nikolas Zimmermann]

Still fails on Gtk, skipped there - works fine on other platforms so it must be a Gtk problem. The setTimeout(, 100) calls are gone, so it must be another issue. Marking as Gtk specific bug.

[Eric Seidel (no email)]

Comment on attachment 47053 [details]
Updated patch v3

Obsoleting this patch since it was landed.

[Nikolas Zimmermann]

Can any gtk person run this test, and produce the actual/diffs.txt files and attach them here?

[Nikolas Zimmermann]

Changed component to SVG, so it shows up in my all-svg-bugs search.

[Zan Dobersek]

The test's no longer failing, so closing the bug.
http://test-results.appspot.com/dashboards/flakiness_dashboard.html#group=%40ToT%20-%20webkit.org&showAllRuns=true&tests=svg%2Fcustom%2Fuse-instanceRoot-event-bubbling.xhtml