All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=895392

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=35d9656d0c6f1136e73584b7df46989665e3bc5d42f6aca48b757e04798f55e9


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

Android Nexus6 WebView Perf

blink_perf.bindings - Benchmark documentation link:
  https://bit.ly/blink-perf-benchmarks

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/12af884ee40000

📍 Found significant differences after each of 9 commits.
https://pinpoint-dot-chromeperf.appspot.com/job/12af884ee40000

Makes AudioSendStream signal that it's part of allocation. by srte@webrtc.org
https://webrtc.googlesource.com/src/+/b686396ec6b6d9e90acc9f6c65eb2cf6a52fe1b8
node-list-access: 196.3 → 209.1 (+12.77)

Revert bucket-boundary checks for branch. (part 2) by bcwhite@chromium.org
https://chromium.googlesource.com/chromium/src/+/fc3a02269e5d241d2b17503f0abe8c12d3375a33
node-list-access: 217.9 → 210.9 (-6.997)

Collect V8 and WebAssembly statistics in separate histograms by mythria@chromium.org
https://chromium.googlesource.com/chromium/src/+/699d26931464b82eb6b0b9846bf7ab1112414f6f
node-list-access: 209.6 → 215.1 (+5.566)

Remove URLRequestContextGetter field from DownloadUrlParameters. by mmenke@chromium.org
https://chromium.googlesource.com/chromium/src/+/6d93c3cdf6de2be69b1d7dc6fc863890d270be71
node-list-access: 208.4 → 178.6 (-29.81)

Remove CR_DEFINE_STATIC_LOCAL from /components. by avi@chromium.org
https://chromium.googlesource.com/chromium/src/+/ac53c5c532ea0584e86790b49bbe39a770bbbc56
node-list-access: 179.9 → 171.8 (-8.109)

[heap] Add flags for predictable GC schedule by mlippautz@chromium.org
https://chromium.googlesource.com/v8/v8/+/a9f74f22a03f4f9e5014d0f49964839702d65a09
node-list-access: 171 → 178.4 (+7.384)

[heap] Add helper for visitation using slot snapshot by mlippautz@chromium.org
https://chromium.googlesource.com/v8/v8/+/15dd4cafc67dc7db357f359f7106f0fc159d9727
node-list-access: 178.4 → 173.5 (-4.927)

[cleanup] Split the mutable roots list into immovable and movable by ishell@chromium.org
https://chromium.googlesource.com/v8/v8/+/33ebe358a5255f080bebe5577f275ae3e54a6ca8
node-list-access: 173.5 → 162.9 (-10.61)

[async] Improve async function handling. by bmeurer@chromium.org
https://chromium.googlesource.com/v8/v8/+/0038e5f05fd71fada7851c01a7db104cfe352089
node-list-access: 163.8 → 176.4 (+12.54)

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

Benchmark documentation link:
  https://bit.ly/blink-perf-benchmarks

My CL was download stuff only, which seems unlikely to affect this test, and just removed a (refcounted) argument to some methods, adding it to a few others, so doubt this is related to my CL.

My CL was about async functions only, which is not used in the test case.

My cl just collects stats for V8/Wasm caches in separate buckets. So it doesn't impact blink bindings. Also it is behind a flag and is not enabled by default.

Given the experiences of bmeurer@ and mythria@ who've changed code that literally cannot have affected perf, yet have a perf claim that bmeurer@ changed "163.8 → 176.4 (+12.54)" and mythria@ changed "209.6 → 215.1 (+5.566)", I tend to disbelieve these perf numbers.

FWIW, my change affected construct-once objects, so any perf loss (which I would argue doesn't exist as it eliminated an allocation) is spread over all the uses.

Whatever the cause was, the graph already recovered.  So, I close this issue.

Oops, I misunderstood the graph.

I took a closer look at other bots, and concluded that this is not a real regression.  Nexus 6 is less powerful and fructuation looks bigger than other platforms, I think.