All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=893775

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=9016f29cee8dfbd55be747942f0971936e1f8c721811229bdbf4dab6ed37c38f


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

Win 7 Nvidia GPU Perf

loading.desktop - Benchmark documentation link:
  https://bit.ly/loading-benchmarks

📍 Pinpoint job started.
https://pinpoint-dot-chromeperf.appspot.com/job/16d4b832e40000

📍 Found a significant difference after 1 commit.
https://pinpoint-dot-chromeperf.appspot.com/job/16d4b832e40000

Only flush tokenized chunks to the HTML parser for valid script types. by acomminos@fb.com
https://chromium.googlesource.com/chromium/src/+/174813503285a1ff88efe4fcb8775df6663fa371
timeToFirstContentfulPaint: 62.49 → 135.2 (+72.68)

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

Benchmark documentation link:
  https://bit.ly/loading-benchmarks

 Issue 893824  has been merged into this issue.

 Issue 891348  has been merged into this issue.

Issue 891391 has been merged into this issue.

Reposting from merged issues:

I've noticed that both The Verge and Naver (sites which regress with this revision), frequently use invalid script types in their bodies for templating:

`<script type="text/template-dynamic" style="display: none;">`

I suspect that parser throughput has gone up by skipping the yield, but the render latency has increased as well due to the lack of flush points. I'm doubtful how meaningful the earlier paints are- as the first paint prior to this revision likely occurred during the yield after parsing the no-op script tags used for templating, much of the templated DOM content would not be visible yet.

I'm going to go ahead and analyze screenshots of the first contentful paint before and after the blamed rev.

FWIW, load_avg is unaffected. Additionally, there appear to be memory usage wins on Taobao (which also uses templating script tags): https://chromeperf.appspot.com/group_report?rev=595719.

 Issue 893835  has been merged into this issue.