Add experiment to disable loading of background tabs during session restore.

chrome/browser/sessions/session_restore_delegate.cc

Index: chrome/browser/sessions/session_restore_delegate.cc
diff --git a/chrome/browser/sessions/session_restore_delegate.cc b/chrome/browser/sessions/session_restore_delegate.cc
index a24167ae1f24bfbbbee5a6f895ec54570c6097ca..18adde4f2d58bbae6e4ac0d6e857978b5e573649 100644
--- a/chrome/browser/sessions/session_restore_delegate.cc
+++ b/chrome/browser/sessions/session_restore_delegate.cc
@@ -4,6 +4,7 @@
 
 #include "chrome/browser/sessions/session_restore_delegate.h"
 
+#include "base/metrics/field_trial.h"
 #include "chrome/browser/sessions/session_restore_stats_collector.h"
 #include "chrome/browser/sessions/tab_loader.h"
 
@@ -11,6 +12,19 @@
 void SessionRestoreDelegate::RestoreTabs(
     const std::vector<RestoredTab>& tabs,
     const base::TimeTicks& restore_started) {
-  SessionRestoreStatsCollector::TrackTabs(tabs, restore_started);
-  TabLoader::RestoreTabs(tabs, restore_started);
+  base::FieldTrial* trial =
+      base::FieldTrialList::Find("SessionRestoreBackgroundLoading");

[sky, 2015/04/03 20:59:09]

Can you elaborate on the value of running a field trial? It's my understand
field trials are useful when we want to collect data for two paths and compare
them. The set of metrics that are going to be reported for this field trial are
going to be very different (the number of tabs is drastically different for one,
which is baked into some of the names). I have no doubt the values will look
better. But what we are lacking is quantifying whether the don't load all is
better. How are you going to quantify that?

At a minimum you need to resolve the naming. Having the name mean the set of
tabs that is active, vs total is misleading.

IMO the UX folks should make the call here. Either they feel it's ok to load on
demand, and we just throw the switch for everyone, or they don't think it's
worth it and we don't bother. Has anyone spoken with UX about this? I'm happy to
move the discussion there if it hasn't.

[Georges Khalil, 2015/04/03 21:51:33]

Discussed offline.

+  // If there is no trial, enable experiment to get coverage on the perf
+  // waterfall.
+  if (trial && trial->group_name() != "Disabled") {
+    SessionRestoreStatsCollector::TrackTabs(tabs, restore_started);
+    TabLoader::RestoreTabs(tabs, restore_started);
+  } else {
+    // If tab loading is disabled, only track active tabs.
+    std::vector<RestoredTab> active_tabs;
+    for (const auto& restored_tab : tabs)

[sky, 2015/04/03 20:59:09]

nit: braces here.

[Georges Khalil, 2015/04/03 21:51:33]

Acknowledged.

+      if (restored_tab.is_active)
+        active_tabs.push_back(restored_tab);
+    SessionRestoreStatsCollector::TrackTabs(active_tabs, restore_started);
+  }
 }