Support saving browser metrics to disk and reading them during next run.

components/metrics/metrics_service.cc

Index: components/metrics/metrics_service.cc
diff --git a/components/metrics/metrics_service.cc b/components/metrics/metrics_service.cc
index 7e31ddb93156bf301ef333bd5cbe71bb7acc9691..45265e394425639c7269f6385d9bf76c73febe40 100644
--- a/components/metrics/metrics_service.cc
+++ b/components/metrics/metrics_service.cc
@@ -1155,11 +1155,20 @@ void MetricsService::RecordCurrentHistograms() {
 
 void MetricsService::RecordCurrentStabilityHistograms() {
   DCHECK(log_manager_.current_log());
+  histogram_snapshot_manager_.StartDeltas();
   // "true" indicates that StatisticsRecorder should include histograms in
   // persistent storage.
-  histogram_snapshot_manager_.PrepareDeltas(
-      base::StatisticsRecorder::begin(true), base::StatisticsRecorder::end(),
-      base::Histogram::kNoFlags, base::Histogram::kUmaStabilityHistogramFlag);
+  for (auto it = base::StatisticsRecorder::begin(true),
+            end = base::StatisticsRecorder::end();
+       it != end; ++it) {

[Ilya Sherman, 2016/05/05 22:22:52]

nit: Why not just write "it != base::StatisticsRecorder::end()"?

[bcwhite, 2016/05/06 16:59:27]

Ah.  Because end() actually has to create a new iterator (which involves
acquiring a lock and creating another end() iterator for the embedded std::map)
so it is best not to do that with each iteration through the loop.

I could change the SR to create the end() iterator during initialization, cache
it, and ever-after just return the cached version.
http://stackoverflow.com/questions/1432216/is-end-required-to-be-constant-in-...

But since the SR keeps all its member variables as static that can be pushed and
popped as more SRs are created (only in tests), the cached end() would require
the same.

So I just cached it in the loop.  :-)

+    if (((*it)->flags() & base::Histogram::kUmaStabilityHistogramFlag) ==
+        base::Histogram::kUmaStabilityHistogramFlag) {

[Ilya Sherman, 2016/05/05 22:22:52]

Optional nit: I personally prefer the style used on line 1148, which omits the
extra == check.

[bcwhite, 2016/05/06 16:59:27]

Me too, but it bites!  This particular kUma...Flag is NOT a single flag but a
merge of two flags.  Thus, omitting the == means matching anything that has at
least one of the bits set which, in this case, matches any
kUmaTargetedHistogramFlag (or: pretty much everything).

Ask me how I discovered this.  Go on...  I dare ya!

I'll add a comment.

+      histogram_snapshot_manager_.PrepareDelta(*it);
+    }
+  }
+  for (MetricsProvider* provider : metrics_providers_)
+    provider->RecordInitialHistogramSnapshots(&histogram_snapshot_manager_);
+  histogram_snapshot_manager_.FinishDeltas();
 }
 
 void MetricsService::LogCleanShutdown() {