Making check_initial_confidence verify return_code bisects.

scripts/slave/recipe_modules/auto_bisect/bisector.py

Index: scripts/slave/recipe_modules/auto_bisect/bisector.py
diff --git a/scripts/slave/recipe_modules/auto_bisect/bisector.py b/scripts/slave/recipe_modules/auto_bisect/bisector.py
index b2002b08ce0931a0639d6a13743388ed0a833c87..835ceef7a14ef0b84fa7d7eb9ba666da7b531e9f 100644
--- a/scripts/slave/recipe_modules/auto_bisect/bisector.py
+++ b/scripts/slave/recipe_modules/auto_bisect/bisector.py
@@ -487,8 +487,11 @@ class Bisector(object):
     other in a statistically significant manner. False if such difference could
     not be established in the time or sample size allowed.
     """
-    if self.test_type != 'perf':
-      return True
+    if self.test_type == 'return_code':
+      # In return_code bisects, mean_value represents the overall return value
+      # of the test, i.e. 0 if all runs returned 0, 1 otherwise.

[qyearsley, 2016/03/23 20:54:40]

mean_value is a potentially confusing or misleading name.

If we were to rename it (in a separate CL), what would it be? summary_value? Or
might mean_value for return code bisects be ignored, and instead set a separate
attribute called overall_return_code (or something similar)?

[RobertoCN, 2016/03/23 21:52:09]

Created a new member of the revision class called overall_return_code.
Still need to refactor mean_value from the code that posts results back to
bisect.

+      return self.good_rev.mean_value != self.bad_rev.mean_value
+
 
     if self.bypass_stats_check:
       dummy_result = self.good_rev.values != self.bad_rev.values