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

base/metrics/histogram_unittest.cc

Index: base/metrics/histogram_unittest.cc
diff --git a/base/metrics/histogram_unittest.cc b/base/metrics/histogram_unittest.cc
index 58b72480018591f72679a6e1ff036727f25ada2c..668ac1ba7e83ad47bfd107beb97e407bff57d1a8 100644
--- a/base/metrics/histogram_unittest.cc
+++ b/base/metrics/histogram_unittest.cc
@@ -147,7 +147,7 @@ TEST_P(HistogramTest, NameMatchTest) {
   EXPECT_EQ(2, samples->GetCount(10));
 }
 
-// Check that delta calculations work correct.
+// Check that delta calculations work correctly.
 TEST_P(HistogramTest, DeltaTest) {
   HistogramBase* histogram =
       Histogram::FactoryGet("DeltaHistogram", 1, 64, 8,
@@ -176,6 +176,32 @@ TEST_P(HistogramTest, DeltaTest) {
   EXPECT_EQ(0, samples->TotalCount());
 }
 
+// Check that final-delta calculations work correctly.
+TEST_P(HistogramTest, FinalDeltaTest) {
+  HistogramBase* histogram =
+      Histogram::FactoryGet("FinalDeltaHistogram", 1, 64, 8,
+                            HistogramBase::kNoFlags);
+  histogram->Add(1);
+  histogram->Add(10);
+  histogram->Add(50);
+
+  std::unique_ptr<HistogramSamples> samples = histogram->SnapshotDelta();
+  EXPECT_EQ(3, samples->TotalCount());
+  EXPECT_EQ(1, samples->GetCount(1));
+  EXPECT_EQ(1, samples->GetCount(10));
+  EXPECT_EQ(1, samples->GetCount(50));
+  EXPECT_EQ(samples->TotalCount(), samples->redundant_count());
+
+  histogram->Add(2);
+  histogram->Add(50);
+
+  samples = histogram->SnapshotFinalDelta();
+  EXPECT_EQ(2, samples->TotalCount());
+  EXPECT_EQ(1, samples->GetCount(2));
+  EXPECT_EQ(1, samples->GetCount(50));
+  EXPECT_EQ(samples->TotalCount(), samples->redundant_count());
+}

[Ilya Sherman, 2016/05/05 07:22:37]

Please test both the case where SnapshotFinalDelta() is called after a call to
SnapshotDelta(), and where it is called without any prior calls to
SanpshotDelta().  That is, please add an additional TEST_P for the other case.

[bcwhite, 2016/05/05 16:01:17]

What is the benefit of that?  There is no difference in the code paths between
the two.  In one case, the "previously logged" samples are zero and in the other
it is not, but that is only data.

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

Without looking at the implementation, it's not hard to imagine having different
code paths for the case where there is and where there isn't a prior call ot
SnapshotDelta().  I prefer to write tests defensively, since the implementation
can change.

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

I agree which is why the second stage ignored an existing sample (1) added a new
sample (2).  There isn't any code difference between them but it expands the
scope of the test to effectively handle entries that previously had a Delta
operation done on them but now have no change, and vice-versa.

+
 TEST_P(HistogramTest, ExponentialRangesTest) {
   // Check that we got a nice exponential when there was enough room.
   BucketRanges ranges(9);