Bug 250561 – REGRESSION(257650@main): Multiple regressions on input-events-get-target-ranges* WPT input event tests

• Status: RESOLVED
• Resolution: FIXED
• Component: UI Events

[Ahmad Saleem]

Created attachment 464482 [details]
Regressing WPT Tests

Hi Team,

While going through STP161 release and noticing passes, I noticed that we regressed in Input Events test cases:

I don't know whether it will work or not:

https://wpt.fyi/results/input-events?diff&filter=ADC&run_id=5076022454648832&run_id=5119896183111680

But I am attaching screenshot as well.

Thanks!

[Ahmad Saleem]

Our failing reduced with exact numbers matching now with Chrome so pretty sure this is regression introduced by myself while doing some "Blink Merges" and if I have to guess, it would be this:

https://github.com/WebKit/WebKit/commit/31be36f212e4872dfd9761736dba5814453ac8a2

[Alexey Proskuryakov]

Thank you!

[Radar WebKit Bug Importer]

<rdar://problem/104227951>

[Karl Dubost]

It's easier to read by changing the order here.
https://wpt.fyi/results/input-events?diff&filter=ADC&run_id=5119896183111680&run_id=5076022454648832&view=subtest

[Ahmad Saleem]

Do we have any regression tool similar to mozregression to automate this so I can confirm my culprit from Comment 01? Appreciate if someone can point me to right direction?

[Alexey Proskuryakov]

Yes, Tools/Scripts/bisect-builds. It hasn't been used in a while though, so could have bit rotted.

[Ahmad Saleem]

(In reply to Alexey Proskuryakov from comment #6)
> Yes, Tools/Scripts/bisect-builds. It hasn't been used in a while though, so
> could have bit rotted.

From where I can download old builds in order to confirm that above commit is culprit?

Or if you know someone in Apple or WebKit team, who could do this or guide me?

[Ahmad Saleem]

(In reply to Ahmad Saleem from comment #7)
> (In reply to Alexey Proskuryakov from comment #6)
> > Yes, Tools/Scripts/bisect-builds. It hasn't been used in a while though, so
> > could have bit rotted.
> 
> From where I can download old builds in order to confirm that above commit
> is culprit?
> 
> Or if you know someone in Apple or WebKit team, who could do this or guide
> me?

https://webkit.org/build-archives/#mac-ventura-x86_64%20arm64 <-- This does not have old builds.

[Alexey Proskuryakov]

We don't show them on the webpage, but we store them for a long time. In any case, bisect-builds downloads on its own.

Please reach out in Slack or email if it's not working - this bug is probably not the right place to discuss bisection.

[Ryosuke Niwa]

Sigh... who the fuck wrote this test? It's totally incomprehensible.

[Ryosuke Niwa]

Regression ranges:
Release 160 (Safari 16.4, WebKit 18615.1.14.3)
257350@main

Release 161 (Safari 16.4, WebKit 18615.1.17.6)
257522@main

[Ryosuke Niwa]

257522@main
392 pass 48 fail

257612@main
392 pass 48 fail

257634@main
392 pass 48 fail

257646@main
392 pass 48 fail

257649@main
392 pass 48 fail

257650@main
260 pass 180 fail

257652@main
260 pass 180 fail

257657@main
260 pass 180 fail

257703@main
260 pass 180 fail

257884@main
260 pass 180 fail

258246@main
260 pass 180 fail

258970@main
260 pass 180 fail

[Ryosuke Niwa]

The above is results for input-events/input-events-get-target-ranges-joining-dl-elements.tentative.html?Backspace

[Ryosuke Niwa]

input-events-get-target-ranges-deleting-in-list-items.tentative.html?Backspace,ol has the same regression point:

257649@main
238 pass 76 fail

257650@main
219 pass 95 fail

[Ryosuke Niwa]

Reverting the culprit in https://bugs.webkit.org/show_bug.cgi?id=250700