Use devil's adb with each bisect iteration, also start/stop daemons.

scripts/slave/recipe_modules/chromium_android/api.py

Index: scripts/slave/recipe_modules/chromium_android/api.py
diff --git a/scripts/slave/recipe_modules/chromium_android/api.py b/scripts/slave/recipe_modules/chromium_android/api.py
index 3dc2fc01ff338c0a3fe6ba5a002514b624fcfa52..3b21386b782ce444be78a7eb0e75f4ed5350fd2e 100644
--- a/scripts/slave/recipe_modules/chromium_android/api.py
+++ b/scripts/slave/recipe_modules/chromium_android/api.py
@@ -318,8 +318,10 @@ class AndroidApi(recipe_api.RecipeApi):
         import sys
         sys.path.append(sys.argv[1])
         from devil import devil_env
+        from devil.android.sdk import adb_wrapper
         devil_env.config.Initialize()
         devil_env.config.PrefetchPaths(dependencies=['adb'])
+        adb_wrapper.AdbWrapper.StartServer()
         """,
         args=[devil_path])
     self.m.adb.set_adb_path(
@@ -1161,6 +1163,52 @@ class AndroidApi(recipe_api.RecipeApi):
       self.asan_device_teardown()
     self.test_report()
 
+  def android_build_wrapper(self, logcat_gs_bucket='chromium-android'):

[ghost stip (do not use), 2016/11/09 01:55:56]

this wacky construction of a contextmanager-generating function is so we can
have consistently defined logcat_gs_bucket. I think it's overkill, but jbudorick
seemed to prefer this over my initial solution in
https://codereview.chromium.org/2485763002/

[jbudorick, 2016/11/09 14:32:36]

On looking at this, a config setting probably would be cleaner, but I could live
with this.

+    @contextlib.contextmanager
+    def wrapper(bisect_obj):
+      """A context manager for use as auto_bisect's build_context_mgr.
+
+      This wraps every overall bisect run.
+      """
+      try:
+        bisect_obj.m.chromium_android.common_tests_setup_steps(
+            perf_setup=True, remove_system_webview=True)
+        bisect_obj.m.chromium.runhooks()
+
+        yield
+      finally:
+        bisect_obj.ensure_checkout()
+        bisect_obj.m.chromium_android.common_tests_final_steps(
+          logcat_gs_bucket=logcat_gs_bucket)
+    return wrapper
+
+  def android_bisect_test_wrapper(self, logcat_gs_bucket='chromium-android'):
+    @contextlib.contextmanager
+    def wrapper(bisect_obj):
+      """A context manager for use as auto_bisect's test_context_mgr.
+
+      This wraps each individual bisect test.
+      """
+      self.use_devil_adb()
+      with self.android_test_wrapper(logcat_gs_bucket)(bisect_obj):

[ghost stip (do not use), 2016/11/09 01:55:56]

this split is so android_test_wrapper can be used independently outside of
bisect and not have the repeated use_devil_adb

[jbudorick, 2016/11/09 14:32:36]

I think this belongs in the bisect modules rather than in here.

[RobertoCN, 2016/11/09 22:10:31]

+1

[ghost stip (do not use), 2016/11/09 23:05:37]

Done.

+        yield
+    return wrapper
+
+  def android_test_wrapper(self, logcat_gs_bucket='chromium-android'):
+    @contextlib.contextmanager
+    def wrapper(_bisect_obj):
+      """A context manager for running android test steps."""
+      try:
+        self.spawn_logcat_monitor()
+        self.spawn_device_monitor()
+
+        yield
+      finally:
+        self.shutdown_device_monitor()
+        self.logcat_dump(gs_bucket=logcat_gs_bucket)
+        self.stack_tool_steps()
+    return wrapper
+
   def run_bisect_script(self, extra_src='', path_to_config='', **kwargs):
     self.m.step('prepare bisect perf regression',
         [self.m.path['checkout'].join('tools',