Refactor bisect-perf-regression_test.py and add smoke test.

tools/bisect-perf-regression.py

Index: tools/bisect-perf-regression.py
diff --git a/tools/bisect-perf-regression.py b/tools/bisect-perf-regression.py
index 682a45fba1c604f64f14a4c01e2149cfab62c765..bbdb6422e27290d585ab35b6d155c01cafffcf44 100755
--- a/tools/bisect-perf-regression.py
+++ b/tools/bisect-perf-regression.py
@@ -286,6 +286,8 @@ def ConfidenceScore(good_results_lists, bad_results_lists):
   # Flatten the lists of results lists.
   sample1 = sum(good_results_lists, [])
   sample2 = sum(bad_results_lists, [])
+  if not sample1 or not sample2:
+    return 0.0

[qyearsley, 2014/07/31 22:34:09]

Previously ConfidenceScore was throwing an error if called with an empty list.

 
   # The p-value is approximately the probability of obtaining the given set
   # of good and bad values just by chance.
@@ -3054,6 +3056,12 @@ class BisectPerformanceMetrics(object):
     last_broken_revision = None
     last_broken_revision_index = -1
 
+    culprit_revisions = []
+    other_regressions = []
+    regression_size = 0.0
+    regression_std_err = 0.0
+    confidence = 0.0

[qyearsley, 2014/07/31 22:34:09]

Previously it was possible that these were referenced before being defined
below.

+
     for i in xrange(len(revision_data_sorted)):
       k, v = revision_data_sorted[i]
       if v['passed'] == 1:
@@ -3233,7 +3241,7 @@ class BisectPerformanceMetrics(object):
                                   results_dict['confidence'])
     _PrintStepTime(revision_data_sorted)
     self._PrintReproSteps()
-    self._PrintThankYou()
+    _PrintThankYou()

[qyearsley, 2014/07/31 22:34:09]

The new test testDryRun was able to catch this :-)

     if self.opts.output_buildbot_annotations:
       bisect_utils.OutputAnnotationStepClosed()