Raise exception when devices fail to list tests.

build/android/pylib/local/device/local_device_gtest_run.py

Index: build/android/pylib/local/device/local_device_gtest_run.py
diff --git a/build/android/pylib/local/device/local_device_gtest_run.py b/build/android/pylib/local/device/local_device_gtest_run.py
index cb8dc471521d7c7e58fee0918195f2dfa7d2784c..17f182ad63fa0e43c0c397f118bf210b83cfc0f7 100644
--- a/build/android/pylib/local/device/local_device_gtest_run.py
+++ b/build/android/pylib/local/device/local_device_gtest_run.py
@@ -319,9 +319,9 @@ class LocalDeviceGtestRun(local_device_test_run.LocalDeviceTestRun):
 
     # Query all devices in case one fails.
     test_lists = self._env.parallel_devices.pMap(list_tests).pGet(None)
-
     # If all devices failed to list tests, raise an exception.
-    if all([tl is None for tl in test_lists]):
+    # Check that tl is not None and is not empty.
+    if all([(tl is None or not tl) for tl in test_lists]):

[BigBossZhiling, 2016/04/25 20:07:15]

But in this case, if a suite indeed has an empty list of test cases, we will
raise an exception, instead of passing 0 test cases. I think that is fine
because we will never have an empty test suite, or what's the point of running
an empty test suite? I guess it's ok to raise an exception when the suite has 0
test case inside it.

[jbudorick, 2016/04/25 20:11:38]

Yeah, I think I'm ok with this in general. However, this should be

  if all(not t1 for t1 in test_lists):

       raise device_errors.CommandFailedError(
           'Failed to list tests on any device')
     return list(sorted(set().union(*[set(tl) for tl in test_lists if tl])))