Bug 142208 – REGRESSION (r180907): inspector/timeline tests frequently time out or fail

• Component: Web Inspector
• Status: NEW

[Brent Fulgham]

The following layout tests are failing on Windows due to timeouts. According to test history, they actually seem to timeout or be very flaky on all platforms.

inspector/timeline/debugger-paused-while-recording.html
inspector/timeline/exception-in-injected-script-while-recording.html

[Radar WebKit Bug Importer]

<rdar://problem/20017459>

[Alexey Proskuryakov]

This looks like a regression from <http://trac.webkit.org/r180907>:

https://webkit-test-results.appspot.com/dashboards/flakiness_dashboard.html#showAllRuns=true&tests=inspector%2Ftimeline%2Fexception-in-injected-script-while-recording.html

Should we roll out?

[Brent Fulgham]

(In reply to comment #2)
> This looks like a regression from <http://trac.webkit.org/r180907>:
> 
> https://webkit-test-results.appspot.com/dashboards/flakiness_dashboard.
> html#showAllRuns=true&tests=inspector%2Ftimeline%2Fexception-in-injected-
> script-while-recording.html
> 
> Should we roll out?

I think so. if we leave it skipped I think it may just join the dozen other inspector-related bugs and timeouts that have been left unaddressed.

[Mark Lam]

(In reply to comment #3)
> (In reply to comment #2)
> > This looks like a regression from <http://trac.webkit.org/r180907>:
> > 
> > https://webkit-test-results.appspot.com/dashboards/flakiness_dashboard.
> > html#showAllRuns=true&tests=inspector%2Ftimeline%2Fexception-in-injected-
> > script-while-recording.html
> > 
> > Should we roll out?
> 
> I think so. if we leave it skipped I think it may just join the dozen other
> inspector-related bugs and timeouts that have been left unaddressed.

I disagree.  I added a test inspector/timeline/exception-in-injected-script-while-recording.html which is based on inspector/timeline/debugger-paused-while-recording.html.  Note that the original test was previously excluded, and I fixed a bug for it and unskipped the test (see http://trac.webkit.org/changeset/180901).

I was warned that the test was seen to be flaky before, but I since I fixed one issue for it, and I wasn't able to reproduce the flakiness locally, I unskipped the test.

The proper remedy is to restore the skipping inspector/timeline again.  I will do this while I take a second look at why the tests can flake now that I have examples of their flakiness to study.

[Mark Lam]

I've restored the skipping of the inspector/timeline directory in r180922: <http://trac.webkit.org/r180922>.

[Mark Lam]

From the bots, it looks like the flaky failures are a lot more reproducible if we run with webkit2 instead of webkit1.  I was able to reproduce this locally.

[Joseph Pecoraro]

These tests now ASSERT under InspectorTimelineAgent::willComposite. Likely due to that not handling nesting well.