Add required_initial_confidence to legacy bisect.

tools/auto_bisect/bisect_perf_regression.py

Index: tools/auto_bisect/bisect_perf_regression.py
diff --git a/tools/auto_bisect/bisect_perf_regression.py b/tools/auto_bisect/bisect_perf_regression.py
index 429624a1e1ab0126d55c3e43388cd2797712e55c..5167b1d3c1a1376fd295b22119de45bfb5a5d7f3 100755
--- a/tools/auto_bisect/bisect_perf_regression.py
+++ b/tools/auto_bisect/bisect_perf_regression.py
@@ -76,9 +76,6 @@ BUILD_RESULT_SUCCEED = 0
 BUILD_RESULT_FAIL = 1
 BUILD_RESULT_SKIPPED = 2
 
-# The confidence percentage we require to consider the initial range a
-# regression based on the test results of the initial good and bad revisions.
-REGRESSION_CONFIDENCE = 80
 # How many times to repeat the test on the last known good and first known bad
 # revisions in order to assess a more accurate confidence score in the
 # regression culprit.
@@ -475,12 +472,15 @@ def _GenerateProfileIfNecessary(command_args):
   return True
 
 
-def _IsRegressionReproduced(known_good_result, known_bad_result):
+def _IsRegressionReproduced(known_good_result, known_bad_result,
+                            required_initial_confidence):
   """Checks whether the regression was reproduced based on the initial values.
 
   Args:
     known_good_result: A dict with the keys "values", "mean" and "std_err".
     known_bad_result: Same as above.
+    required_initial_confidence: Minimum confidence score for the given
+        good and bad revisions to avoid early aborting.
 
   Returns:
     True if there is a clear change between the result values for the given
@@ -492,12 +492,12 @@ def _IsRegressionReproduced(known_good_result, known_bad_result):
       return map(math_utils.Mean, values)
     return values
 
-  p_value = BisectResults.ConfidenceScore(
+  initial_confidence = BisectResults.ConfidenceScore(
       PossiblyFlatten(known_bad_result['values']),
       PossiblyFlatten(known_good_result['values']),
       accept_single_bad_or_good=True)
 
-  return p_value > REGRESSION_CONFIDENCE
+  return initial_confidence >= required_initial_confidence
 
 
 def _RegressionNotReproducedWarningMessage(
@@ -2329,7 +2329,8 @@ class BisectPerformanceMetrics(object):
       # beyond chance-induced variation.
       if not (self.opts.debug_ignore_regression_confidence or
               self._IsBisectModeReturnCode()):
-        if not _IsRegressionReproduced(known_good_value, known_bad_value):
+        if not _IsRegressionReproduced(known_good_value, known_bad_value,
+                                       self.opts.required_initial_confidence):
           # If there is no significant difference between "good" and "bad"
           # revision results, then the "bad revision" is considered "good".
           # TODO(qyearsley): Remove this if it is not necessary.
@@ -2582,6 +2583,7 @@ class BisectOptions(object):
     self.bisect_mode = bisect_utils.BISECT_MODE_MEAN
     self.improvement_direction = 0
     self.bug_id = ''
+    self.required_initial_confidence = 80.0
 
   @staticmethod
   def _AddBisectOptionsGroup(parser):
@@ -2633,6 +2635,11 @@ class BisectOptions(object):
                             'If this number is given, bisect will attempt to ' +
                             'verify that the bug is not closed before '
                             'starting.')
+    group.add_argument('--required_initial_confidence', type=float,
+                       default=80.0,
+                       help='The required confidence score for the initial '
+                            'check to see whether there is a significant '
+                            'difference between given good and bad revisions.')
 
   @staticmethod
   def _AddBuildOptionsGroup(parser):