Chrome: 72, 73
OS: Windows, Linux

*************************************************

1. unpack the attached test.zip
2. run the following:
   chrome test.html test.html
3. look at the page load duration displayed in the message box

Expected: ~3.5 sec in active tab, 0.5-1sec in background tab
Observed: ~5.5 sec in active tab, ~2.5sec in background tab

*************************************************

test.html is an HTML/CSS-only version of ECMA spec page:
https://www.ecma-international.org/ecma-262/9.0/index.html

*************************************************

Bisected to: 611314 (good) - 611325 (bad)
https://chromium.googlesource.com/chromium/src/+log/d956af27..edf93e9e?pretty=fuller

Suspecting based on chrome://tracing analysis (below):
r611317 = 17f95119ef504da028a2ad2170e44ec0a62305e8 = crrev.com/c/1351366 by chrishtr@chromium.org
"Force a new token chunk after custom element start tags."
Landed in 72.0.3624.0

*************************************************

Two traces (good/bad) are attached.
Some observations in the bad builds:

* 100 times more "slices" in the parsing stage (188k vs 1.8k)
* ThreadControllerImpl::RunTask and MessageLoop::RunTask are the top "Self time" consumers
  at 2 seconds total, which is the absolute value of the observed slowdown
* zooming in reveals HTMLDocumentParser::EnqueueTokenizedChunk in each slice

 

Deleted: test.7z 486 KB

test.7z 486 KB Download

Deleted: trace_r611314.json.gz 2.4 MB

trace_r611314.json.gz 2.4 MB Download

Deleted: trace_r611329.json.gz 3.6 MB

trace_r611329.json.gz 3.6 MB Download