📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/14ba4178e40000

📍 Found significant differences after each of 2 commits.
https://pinpoint-dot-chromeperf.appspot.com/job/14ba4178e40000

Define basic Trichrome build targets. by torne@google.com
https://chromium.googlesource.com/chromium/src/+/bef92b9197b49def215aced79961f50e3b83544e
3496 → No values

Revert "Define basic Trichrome build targets." by thomasanderson@chromium.org
https://chromium.googlesource.com/chromium/src/+/f505cde61a75f566b4edd2043015a93e5568f970
No values → 1136

Understanding performance regressions:
  http://g.co/ChromePerformanceRegressions

Benchmark documentation link:
  None

My change is not supposed to have changed any code in any existing binary at all, so it would be very surprising if it regressed performance. It's only adding new build targets that are not used for anything yet..

Wait - "No values" just suggests it was broken, which isn't surprising because my change broke the build, which is why it was reverted. This seems like pinpoint shouldn't be alerting for it?

I think this means something else landed while trichrome in place and caused the regression. Here's the changelog: https://chromium.googlesource.com/chromium/src/+log/bef92b9197b49def215aced79961f50e3b83544e~1..f505cde61a75f566b4edd2043015a93e5568f970

Nothing really jumps out :\

Based on the trace the delay happens before the browser process starts executing, so this could be some change in the test harness too.

+jianli

Possibly turning on "offline indicator" for tests caused more work to happen early: https://chromium.googlesource.com/chromium/src/+/3f1cecfb0bbefa53008b956e21b82c7fa03c025b%5E%21/#F1

Finch data for dev channel seems to suggest it's not an issue: https://uma.googleplex.com/p/chrome/variations?sid=43a25c00be971c14dd8364e734f7e74c

How do we rule out changes in test harness?

I am going to submit a patch to turn if off in ChromeActivityTestRule soon.

https://chromium-review.googlesource.com/c/chromium/src/+/1252362

Sami, there's a bunch of different perf bugs that all happened at the same time, between trichrome landing and being reverted: issue 890833,  issue 890880 ,  issue 890881 ,  issue 890890 ,  issue 891234 ,  issue 891256 
They're variously assigned to thomasanderson or me (android sheriffs at the time, plus it was my change)
I'm not sure what to actually do about this; did the change in #8 here address some/all of these?

Would you be able to collect these together and dupe them/assign them as appropriate, or give me some hints on how to figure out what happened? :)

So looks like the startup regression has gone away so we can close this: https://chromeperf.appspot.com/report?sid=d70f414145f7f470265fda9f4ddf9df2362fd002febaaaaa28de3658409b898a&start_rev=593508&end_rev=600884

I'll add comments to the other bugs.

Thanks Sami!