Loosen FirstMeaningfulPaint UMA network-idle heuristics

tools/metrics/histograms/histograms.xml

Index: tools/metrics/histograms/histograms.xml
diff --git a/tools/metrics/histograms/histograms.xml b/tools/metrics/histograms/histograms.xml
index 84630b481d6adeeb90e61be86c090eb3cb8f0185..5b7307a8a19e192f868d5697dbc1b9e35624a7db 100644
--- a/tools/metrics/histograms/histograms.xml
+++ b/tools/metrics/histograms/histograms.xml
@@ -20283,6 +20283,25 @@ http://cs/file:chrome/histograms.xml - but prefer this file for new entries.
   </summary>
 </histogram>
 
+<histogram name="FirstMeaningfulPaintDetector.FirstMeaningfulPaintOrdering"

[Ilya Sherman, 2017/03/08 09:19:46]

Would it be appropriate to group these new histograms under the
"PageLoad.Experimental.PaintTiming.*" category?

[Kunihiko Sakamoto, 2017/03/08 10:16:10]

I wanted to separate these from PageLoad.* metrics, that are aggregated and
filtered (excluding background tabs etc.) in browser process. These
FirstMeaningfulPaintDetector.* are logged from renderer process.

[Ilya Sherman, 2017/03/08 22:46:45]

What about something like
"PageLoad.Experimental.Renderer.FirstMeaningfulPaintDetector.*"?  These are
still metrics related to page load timing, correct?  The top-level category
should reflect a fairly general concept -- "FirstMeaningfulPaintDetector" seems
to me to be overly specific.

[Kunihiko Sakamoto, 2017/03/10 04:37:10]

Renamed to "PageLoad.Experimental.Renderer.FirstMeaningfulPaintDetector.*".
Thanks for the naming suggestion.

+    enum="FirstMeaningfulPaintOrdering">
+  <owner>ksakamoto@chromium.org</owner>
+  <summary>
+    Whether the two variants of First Meaningful Paint reported different value,

[Ilya Sherman, 2017/03/08 09:19:46]

nit: s/value/values

[Kunihiko Sakamoto, 2017/03/08 10:16:10]

Done.

+    and if so, which one was reported first.
+  </summary>
+</histogram>
+
+<histogram name="FirstMeaningfulPaintDetector.HadNetworkQuiet"
+    enum="NetworkQuietStatus">
+  <owner>ksakamoto@chromium.org</owner>
+  <summary>
+    Recorded when the page load reached network 0-stable (no active network
+    connection for 0.5 seconds), or network 2-stable (no more than 2 active
+    network connections for 2 seconds).

[Ilya Sherman, 2017/03/08 09:19:46]

Would it be useful to also record when a page never reached either of these
states, as a baseline?

[Kunihiko Sakamoto, 2017/03/08 10:16:10]

Unfortunately it's hard (if not impossible) to record it reliably in renderer,
since renderer can be killed any time. We record (# page loads that had 2-quiet
/ total page loads) in browser side PageLoad metrics, so not recording baseline
here would be okay.

[Ilya Sherman, 2017/03/08 22:46:45]

Okay.  In that case, could you please include a reference, in this histogram
description, to the metric that you think is an appropriate baseline?

[Kunihiko Sakamoto, 2017/03/10 04:37:10]

Done.

+  </summary>
+</histogram>
+
 <histogram base="true" name="FirstUserAction.BackgroundTime.MainIntent"
     units="minutes">
   <owner>tedchoc@chromium.org</owner>
@@ -93938,6 +93957,12 @@ http://cs/file:chrome/histograms.xml - but prefer this file for new entries.
   <int value="20" label=".log"/>
 </enum>
 
+<enum name="FirstMeaningfulPaintOrdering" type="int">
+  <int value="0" label="FMP 0-quiet &lt; FMP 2-quiet"/>
+  <int value="1" label="FMP 0-quiet &gt; FMP 2-quiet"/>
+  <int value="2" label="FMP 0-quiet = FMP 2-quiet"/>
+</enum>
+
 <enum name="FirstMeaningfulPaintSignalStatus" type="int">
   <int value="0" label="No input, network active"/>
   <int value="1" label="Had input, network active"/>
@@ -102235,6 +102260,11 @@ from previous Chrome versions.
   <int value="6" label="Channel Type Change"/>
 </enum>
 
+<enum name="NetworkQuietStatus" type="int">
+  <int value="0" label="Had network 0-quiet for 0.5 seconds"/>
+  <int value="1" label="Had network 2-quiet for 2 seconds"/>
+</enum>
+
 <enum name="NetworkSecurityType" type="int">
   <summary>
     The security types come from the connman_service_security enum in