Record InstantExtended pref setting on startup with an UMA histogram.

chrome/browser/ui/search/search.cc

Index: chrome/browser/ui/search/search.cc
===================================================================
--- chrome/browser/ui/search/search.cc	(revision 185538)
+++ chrome/browser/ui/search/search.cc	(working copy)
@@ -279,6 +279,14 @@
              template_url);
 }
 
+void RecordInstantExtendedPrefValue(bool pref_value) {
+  static bool recorded = false;

[sreeram, 2013/03/01 16:42:07]

This bool is static across the whole browser process, not per profile. So, this
histogram will only be recorded for the first profile that creates a window.
That's not what we want, I guess?

[Alexei Svitkine (slow), 2013/03/01 17:57:56]

I agree that's kind of weird. From the bug:

"It's much better to have an
opt-in/opt-out/eligible-for-field-trial-to-assign-status histogram recorded
regularly (either on startup or every so often or upon some user action).  Then
we can easily tell what percent of the population opted out.  Recording a user
action only tell X people opted out this day.  That can zero if no one new is
opting out, but we can still have a huge population that disappeared.  If we
don't happen to look at the user action counts on the day they disappeared then
we'd be entirely oblivious to this negative reaction."

Perhaps we should take this discussion there?

(Note: The above quote is also kind of mixing up the about:flags setting with
this one.)

+  if (!recorded) {
+    UMA_HISTOGRAM_BOOLEAN("InstantExtended.PrefValue", pref_value);
+    recorded = true;
+  }
+}
+
 void EnableInstantExtendedAPIForTesting() {
   CommandLine* cl = CommandLine::ForCurrentProcess();
   cl->AppendSwitch(switches::kEnableInstantExtendedAPI);