All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=868517

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=81287334d7a705de5d1265fbe8caa412cbbafaa4834621360e1927eedc67c0e8


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

Android Nexus5X WebView Perf
Win 7 Nvidia GPU Perf

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/13f7432da40000

📍 Found significant differences after each of 2 commits.
https://pinpoint-dot-chromeperf.appspot.com/job/13f7432da40000

win: Attempt to fix compile in branded builds. by thakis@chromium.org
https://chromium.googlesource.com/chromium/src/+/1a0c7ebc9a25ca0e51cadcfb6912e9011c0109e5
9216 → 9365 (+149.2)

[css-grid] New list-based data structure for the Grid by svillar@igalia.com
https://chromium.googlesource.com/chromium/src/+/b9351e20a47f21d28527b55d49e2191f0a08845f
9279 → 2060 (-7219)

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

My commit is a compile fix. If the perf bots do official builds then they couldn't build for a while before that fix, but then they should attribute the regression to all changes that landed since the last successful compile, not just to that.

This was expected. Inserting a large item is the worst case scenario for the new list-based grid data structure. We precisely landed this new test in order to track this regression. It should not affect real life use cases though.

If it is expected and acceptable feel free to close as wontfix.  If there is still work being done to improve the worst-case scenario then please reference that bug as a blocking bug for this one.

Closing then as WONTFIX.

It was already expected to have this regression, because it's an artificial test testing the worst case scenario for the new list based data structure for grids. The previous vector-based implementation was pretty fast inserting new items but very greedy memory-wise. The new one has great memory savings for huge sparse grids but it's slower on this rare use cases. That's a trade off as usual between memory and performance.

Knowing that, we decided to land the perf test before landing the new data structure to keep track of the regression.

Issue 868972 has been merged into this issue.