All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=605839

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgxLfLoAoM


Bot(s) for this bug's original alert(s):

android-nexus5X

cc test owner.

bokan: bisect can not run for an error.

> Failed steps failed gathering reference values.performance test 1 of 5 failed gathering reference values.reading chartjson results

https://build.chromium.org/p/tryserver.chromium.perf/builders/android_nexus5X_perf_bisect/builds/109

https://chromeperf.appspot.com/group_report?bug_id=605839

In the graph, this test does not have enough sampling points after Apr/11. My guess is the recent versions of Chrome can not report results stably.

Can you help me to find out what happens there?

Interesting. Looks like there's nothing for "mean_input_event_latency" between Apr/11 and Apr/19 but individual test cases do have more data. In particular, ESPN has the regression at Apr 16. Could you try kicking off another bisect against an individual test case. I'll take a closer look later this week.

I manually bisected this down to:

  Eliminate pinch drift by removing touch radius from span calculation

  Previously, we included touch radius in our pinch span computation.
  This caused some drift between the users fingers and the web contents
  during pinch. Removing the touch radius fixes this problem.

  BUG= 590026 

  Review URL: https://codereview.chromium.org/1778063002

  Cr-Commit-Position: refs/heads/master@{#386377}

This patch would activate the zoom gesture sooner so I expect that's why we're seeing a slight movement in the metrics so this isn't a real regression. Assigning to tdresser@ to confirm my hunch but I think we can close this as WontFix.

Yup, your hunch is correct.