Index: build/android/pylib/test_result.py |
diff --git a/build/android/pylib/test_result.py b/build/android/pylib/test_result.py |
index d6074f92b8f1acafde567050fd98d6e0735b77c0..2eb1c66f08107dd4207a4d71e9431f638a1f1676 100644 |
--- a/build/android/pylib/test_result.py |
+++ b/build/android/pylib/test_result.py |
@@ -177,7 +177,8 @@ class TestResults(object): |
# Summarize in the test output. |
summary = ['Summary:\n'] |
- summary += ['TESTS_TO_RUN=%d\n' % (len(tests_to_run))] |
+ if tests_to_run: |
Isaac (away)
2012/11/07 19:32:37
I think we can remove these checks now that we are
nilesh
2012/11/07 19:35:27
Actual list of tests is not passed from run_instru
|
+ summary += ['TESTS_TO_RUN=%d\n' % (len(tests_to_run))] |
num_tests_ran = (len(self.ok) + len(self.failed) + |
len(self.crashed) + len(self.unknown)) |
tests_passed = [t.name for t in self.ok] |
@@ -189,7 +190,7 @@ class TestResults(object): |
'FAILED=%d %s\n' % (len(tests_failed), tests_failed), |
'CRASHED=%d %s\n' % (len(tests_crashed), tests_crashed), |
'UNKNOWN=%d %s\n' % (len(tests_unknown), tests_unknown)] |
- if num_tests_ran != len(tests_to_run): |
+ if tests_to_run and num_tests_ran != len(tests_to_run): |
# Add the list of tests we failed to run. |
tests_failed_to_run = list(set(tests_to_run) - set(tests_passed) - |
set(tests_failed) - set(tests_crashed) - |