[Findit] Support gtest on Android.

appengine/findit/waterfall/trigger_base_swarming_task_pipeline.py

Index: appengine/findit/waterfall/trigger_base_swarming_task_pipeline.py
diff --git a/appengine/findit/waterfall/trigger_base_swarming_task_pipeline.py b/appengine/findit/waterfall/trigger_base_swarming_task_pipeline.py
index f8dbfe1b3558814e2b2c3893ca0145f22d6e8639..f290f020dc02f903f30240b96a4381d4c41ac7d3 100644
--- a/appengine/findit/waterfall/trigger_base_swarming_task_pipeline.py
+++ b/appengine/findit/waterfall/trigger_base_swarming_task_pipeline.py
@@ -29,7 +29,6 @@ class TriggerBaseSwarmingTaskPipeline(BasePipeline):  # pragma: no cover.
                                     builder_name, build_number, step_name,
                                     tests, iterations):
     """Returns a SwarmingTaskRequest instance to run the given tests only."""
-
     # Make a copy of the referred request and drop or overwrite some fields.
     new_request = copy.deepcopy(ref_request)
     new_request.name = self._GetSwarmingTaskName(ref_task_id)
@@ -47,8 +46,21 @@ class TriggerBaseSwarmingTaskPipeline(BasePipeline):  # pragma: no cover.
             not a.startswith('--test-launcher-filter-file'))
     ]
     new_request.extra_args.append('--gtest_filter=%s' % ':'.join(tests))
+
+    # On Android, this is only supported for gtest, but not for other test

[chanli, 2016/11/29 23:11:49]

Nit: 'this' is a little ambiguous here, maybe change it to 'gtest_repeat' flag
explicitly?

[stgao, 2016/11/30 00:26:27]

Good catch. Done.

+    # types. E.g. instrumentation tests currently support it via --test-repeat.
+    # Here we blindly treat all tests on Android as gtest, and let other test
+    # types fail out, because it is hard to distinguish them programmatically.
+    #
+    # https://crbug.com/669632 tracks the effort to unify the command switches
+    # of the Android test runner that are used here.
     new_request.extra_args.append('--gtest_repeat=%s' % iterations)
-    new_request.extra_args.append('--test-launcher-retry-limit=0')
+
+    ref_os = swarming_util.GetTagValue(ref_request.tags, 'os') or ''
+    if ref_os.lower() == 'android':  # Workaround. pragma: no cover.
+      new_request.extra_args.append('--num_retries=0')
+    else:
+      new_request.extra_args.append('--test-launcher-retry-limit=0')
 
     # Also rerun disabled tests. Scenario: the test was disabled before Findit
     # runs any analysis. One possible case:
@@ -62,6 +74,9 @@ class TriggerBaseSwarmingTaskPipeline(BasePipeline):  # pragma: no cover.
     #   4. Findit picks the latest Waterfall build of the matching configuration
     #      for the CQ build in which the flaky test is found.
     #   5. In the picked Waterfall build, the test is already disabled.
+    #
+    # Note: test runner on Android doesn't support this flag yet even though it
+    # exists.

[chanli, 2016/11/29 23:11:49]

Question: If this flag is not supported on Android, will an error happen or this
flag will be simply ignored?

[stgao, 2016/11/30 00:26:27]

Ignored. Comments updated.

     new_request.extra_args.append('--gtest_also_run_disabled_tests')
 
     # Remove the env setting for sharding.