All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=773692

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=3425338386197382c9d523d77bea509c2d8fdbfa79a58ee37f236dd14e57857c


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

chromium-rel-mac11-pro

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/168b3e65f80000

📍 Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/168b3e65f80000

[parser] Skipping inner funcs: Turn flag back on.
By marja@chromium.org · Wed Oct 04 08:23:11 2017
v8 @ fa3b73fffe232b601d2a94f129446f41c765a86e

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

The graphs indeed show a 10%-ish regression and my commit is on the range. However, the range is really large (around 800 chromium commits).

Coincidentally, I was also flipping the flag on and off a bit, which makes analyzing this bug easier.

On: 506368
Off: 505211
On: 499539

ChromiumPerf/chromium-rel-mac11-pro/system_health.common_desktop / timeToFirstMeaningfulPaint_avg / load_tools /

-> This one doesn't show a corresponding improvement when the flag was flipped off.

ChromiumPerf/chromium-rel-mac11-pro/system_health.common_desktop / timeToFirstPaint_avg / browse_accessibility_tech / browse_accessibility_tech_codesearch

-> This one shows an improvement when the flag was flipped off, but it's around 2%.

ChromiumPerf/chromium-rel-mac11-pro/system_health.common_desktop / timeToFirstContentfulPaint_avg / browse_accessibility_tech / browse_accessibility_tech_codesearch

-> Ditto

Conclusion: I think the 10% regression is being misattributed to my commit. There's probably a small codesearch regression - that might just be wontfix since it's an individual web site and according to our analysis, most web sites in our benchmarks benefit from this change.

Looking closer to the pinpoint job, it seems to indeed have bisected to my commit with good certainty. I don't understand what's going on. Why doesn't the graph show a similar regression when I flipped the flag on the last time...

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/11ee9585f80000

I started a bisect/pinpoint job for the improvement. Than we have better comparable graphs.

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/13ac1bd9f80000

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/16914df9f80000

📍 Couldn't reproduce a difference.
https://pinpoint-dot-chromeperf.appspot.com/job/11ee9585f80000

😿 Pinpoint job stopped with an error.
https://pinpoint-dot-chromeperf.appspot.com/job/13ac1bd9f80000

📍 Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/16914df9f80000

[parser] Skipping inner funcs: Turn flag back on.
By marja@chromium.org · Wed Oct 04 08:23:11 2017
v8 @ fa3b73fffe232b601d2a94f129446f41c765a86e

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

marja: any update here? Does it seem like some of these are related to your CL but not all of them?

Yes, we got a wealth of reports and they were very confusing (for some of them, the graph didn't go up / down in a logical way wrt the feature on / feature off CLs, or didn't go down with the revert as much as it went up with the CL that flipped the feature on. Also in some graphs, the regression had recovered over time, and in some graphs, my fixes helped. There were some real regressions remaining too, but we concluded the improvements trump them.

Re comment 12, code search is one of the real regressions.

... that said... the regression in the pinpoint looks a bit too steep. (Also I landed some memory fixes later...)

https://chromium-review.googlesource.com/750782 << this is the main fix

Alright, dug up the graph. It's here:

https://chromeperf.appspot.com/report?sid=35e5114cf3df09c894e871a68a480f66f8c0479a11eb5cb82ef551564ee6c727&start_rev=504536&end_rev=510854

From that it's clear that my fix fixes most of the regression.

Afaics nothing actionable here (but it might be that we've just missed something...); looked at the regressions w/ hablich@ some time ago and concluded that there are some regressions on individual pages (after the fix), most notably code search, but the improvements are more important.