Desktop fast shutdown experiment

chrome/browser/chrome_browser_field_trials_desktop.cc

Index: chrome/browser/chrome_browser_field_trials_desktop.cc
diff --git a/chrome/browser/chrome_browser_field_trials_desktop.cc b/chrome/browser/chrome_browser_field_trials_desktop.cc
index 1901825d0d8eba14feef045856b9da736cbaf9ce..1a6ec57f8b0b8cb71d61d65f5b7fb88f809aac92 100644
--- a/chrome/browser/chrome_browser_field_trials_desktop.cc
+++ b/chrome/browser/chrome_browser_field_trials_desktop.cc
@@ -14,6 +14,7 @@
 #include "base/metrics/field_trial.h"
 #include "base/metrics/histogram_macros.h"
 #include "base/path_service.h"
+#include "chrome/browser/features.h"
 #include "chrome/browser/prerender/prerender_field_trial.h"
 #include "chrome/common/chrome_paths.h"
 #include "chrome/common/chrome_switches.h"
@@ -112,6 +113,9 @@ void SetupDesktopFieldTrials(const base::CommandLine& parsed_command_line) {
 #if defined(OS_WIN)
   SetupStabilityDebugging();
 #endif  // defined(OS_WIN)
+  // Activate the experiment as early as possible to increase its visibility
+  // (e.g. the likelihood of its presence in the serialized system profile).
+  base::FeatureList::IsEnabled(kDesktopFastShutdown);

[rkaplow, 2016/11/30 21:15:36]

I'm not entirely clear why this is needed - is the intention here for this to
tag the initial stability proto? Is that because you're concerned there will be
a crash  after the isEnabled call?

[manzagop (departed), 2016/12/01 23:29:13]

Yes! Because we're specifically interested in the stability of this experiment,
I want to make sure the experiment gets registered as early as possible so
there's more chances of the crash key callback having run and the system profile
proto serialization having occurred. Otherwise you could imagine shutdown being
initiated, chrome entering the experiment but then being killed before it gets
to write the experiment to the system profile on disk.

I realize this is not how it usually goes. You could argue that a chrome that
crashes mid-run never triggered the experiment and that we don't want to count
it. In this case, we think it might have a large impact on overall stability, so
I think this is the right call.

Actually, there's a second reason . We need to have entered the experiment so
the rendez-vous metrics (https://codereview.chromium.org/2399233002/) are
assigned to the experiment. 
https://cs.chromium.org/chromium/src/chrome/browser/chrome_browser_main.cc?sq...

 }
 
 }  // namespace chrome