Because of the byte string handling.
A couple examples:

1. ad.load_snippet(name='spt', package='com.google.android.mobly.snippet.example1')
   WARNING:root:[AndroidDevice|84B7N15A28012457] a bytes-like object is required, not 'str'

2. ad.stop_services()
   Traceback (most recent call last):
   File "", line 1, in
   File "/Users/angli/Developer/mobly/mobly/controllers/android_device.py", line 417, in stop_services
   self._terminate_jsonrpc_client(client)
   File "/Users/angli/Developer/mobly/mobly/controllers/android_device.py", line 576, in _terminate_jsonrpc_client
   client.stop_app()
   File "/Users/angli/Developer/mobly/mobly/controllers/android_device_lib/snippet_client.py", line 68, in stop_app
   if 'OK (0 tests)' not in out:
   TypeError: a bytes-like object is required, not 'str'

Reported in #87
https://travis-ci.org/google/mobly/jobs/197104959

=================================== FAILURES ===================================
____________________ UtilsTest.test_start_standing_subproc _____________________
self = <tests.mobly.utils_test.UtilsTest testMethod=test_start_standing_subproc>
def test_start_standing_subproc(self):
with self.assertRaisesRegexp(utils.Error,
"Process .* has terminated"):
> utils.start_standing_subprocess("sleep 0", check_health_delay=0.1)
E AssertionError: Error not raised
tests/mobly/utils_test.py:32: AssertionErro

Today's generated test trigger function share the same code path as a regular test, which makes accounting, reporting, and logging difficult. It also makes it hard to support selecting specific generated test cases via --test_cases.

It is time to redesign and implement the parameterized test mechanism properly.

Execute adb -s <serial> reconnect at the end of AndroidDevice controller object cleanup.

It has several benefits:

1. Kills any lingering adb processes/cmds that the test may have started and does not clean up properly.
2. Releases all forwarded adb ports.
3. Recovers unresponsive devices in some cases. Some devices may stop responding to all adb commands due to adb calls made by tests, leaving the device in an unusable state. Refreshing adb connection fixes the problem.

Overall this seems to be a good approach since it can clears any baggage a test may have left with adb and makes the execution cleaner.
The command does not affect other devices connected to the same host.

Refactor and open source the attenuator controller code.

In the Results array of the summary, the number of entries and whether there is an Extra error message currently differs depending on where the test failed.

1. We should have a result for every test case requested (explicitly or implicitly)
2. We should use extr error to surface the point of failure (e.g., setup_class, setup_test, abort, teardown_test, etc.)

If the object is obtained by directly calling android_device.get_all_instances instead of android_device.create, the device object will try to stop SL4A even if SL4A is not started in the first place.

>>> from mobly.controllers import android_device
>>> ad = android_device.get_all_instances()[0]
>>> ad.stop_services()
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/usr/local/google/home/angli/Developer/mobly/mobly/controllers/android_device.py", line 417, in stop_services
    self._terminate_jsonrpc_client(client)
  File "/usr/local/google/home/angli/Developer/mobly/mobly/controllers/android_device.py", line 574, in _terminate_jsonrpc_client
    client.closeSl4aSession()
  File "/usr/local/google/home/angli/Developer/mobly/mobly/controllers/android_device_lib/jsonrpc_client_base.py", line 292, in rpc_call
    return self._rpc(name, *args)
  File "/usr/local/google/home/angli/Developer/mobly/mobly/controllers/android_device_lib/jsonrpc_client_base.py", line 281, in _rpc
    raise ProtocolError(ProtocolError.NO_RESPONSE_FROM_SERVER)
mobly.controllers.android_device_lib.jsonrpc_client_base.ProtocolError: No response from server.

When running the sample test, it raises ProtocolError, and SL4A server is started

self._start_jsonrpc_client(self.sl4a, host_port, device_port)
File "/Library/Python/2.7/site-packages/mobly/controllers/android_device.py", line 598, in _start_jsonrpc_client
client.connect(port=host_port)
File "/Library/Python/2.7/site-packages/mobly/controllers/android_device_lib/jsonrpc_client_base.py", line 214, in connect
raise ProtocolError(ProtocolError.NO_RESPONSE_FROM_HANDSHAKE)
ProtocolError: No response from handshake.
[SampleTestBed] 01-03 18:45:57.005 ERROR No response from handshake.
[SampleTestBed] 01-03 18:45:57.005 INFO [Test Case] setup_class FAIL
[SampleTestBed] 01-03 18:45:57.006 INFO Summary for test run SampleTestBed@01-03-2017_18-45-52-772: Error 0, Executed 1, Failed 1, Passed 0, Requested 1, Skipped 0
frankbp:Downloads xielieping$ adb devices
List of devices attached
P4M0214A08001775 device

Move EventDispatcher handling into the sl4a client so consolidate sl4a client code.

After this, ed can be accessed via self.sl4a.ed as opposed to self.ed.

In various places in the code there are hardcoded sleeps, eg the 10s sleep in monsoon.
Blind sleeps add flakiness and slow down test execution; we need to remove these.

Something similar to the Android site would be great. ;)

Right now the logger prefix for Android device controller can not be changed.
Sometimes it is useful to be able to change the prefix to be more fitting for the context of the test.
E.g. Instead of [AndroidDevice|<serial>], [AndroidDevice|caller] could be more useful in a call test.

Operations like toggling Wi-Fi/BT, connecting to a Wi-Fi network etc.

This way users can have an easy example to start with, as opposed to having to deal with the shenanigans of SL4A.

To do so, we'll need to create a new repo to host the code for the snippet implementation first.

Corresponding to google/mobly-snippet-lib#24

Reference:
https://greenify.uservoice.com/knowledgebase/articles/749142-how-to-grant-permissions-required-by-some-features

There is an edge case where ad.is_rootable incorrectly returns 'True', in which case we issue adb root call, causing some NBU devices to stop responding to adb commands.

The edge case is: adb server is not running prior to starting the test.
In this situation, the following steps occur:

1. Starting the test
2. AndroidDevice controller obj issues adb call to determine build type
3. adb server gets started, printing extra strings to stdout
4. is_rootable tries to match the string 'user' to the stdout, which evaluates to 'False' due to the extra output.
5. The user build device is deemed 'rootable'.

Sample output from such a case:

# Beginning of the test
$ adb -s <serial> shell getprop ro.build.type
# Results in the following stdout:
# * daemon not running. starting it now on port 5037 *
# * daemon started successfully *
# user^M

ad.adb.getprop is the first adb call made during the lifecycle of an AndroidDevice obj, so it's critical to handle this check correctly.
Other adb-based check may also fail if adb server somehow gets restarted and stdout is polluted. As a less urgent item, we should consider making checks based on adb's stdout more robust.

Currently, getting the stack trace of the failure requires reading the details txt

Right now we have to manually add the device serial every time we throw an exception from inside AndroidDevice, which is tedious and annoying.
We should have a better way to handle this, perhaps an inner exception class.

Is it possible to enable WIFI with out using adb command(adb shell settings get global wifi_on)?

I mean possibility for accessing internal API like wifiManager.setWifiEnabled(true) in Java, Considering fact that adb is not reliable on higher versions like API 24 real devices.

Would be more helpful if we have sample example how to handle network connections :)

Right now AndroidDevice.reboot does not restore any snippet.
It should reload the snippet loaded before reboot and make their aliases available.

Devices made by GiONEE does not have ro.build.product in their getprop outputs, causing ad.model to fail.
For these devices, we need to check ro.product.model for model info.

Let's also evaluate the feasibility of replacing ro.build.product with ro.product.model for good.

In jsonrpc_client_base:

    try:
        self._conn = socket.create_connection(('127.0.0.1', self.host_port),
                                              _SOCKET_TIMEOUT)
        self._conn.settimeout(_SOCKET_TIMEOUT)
    except (socket.timeout, socket.error, IOError):
        logging.exception("Failed to create socket connection!")
        raise

Would result in the trace being printed twice.

If possible, would be nice to have an ad shell taking care of all the RPC service launches.
$ mobly_android_shell -s --service_names sl4a snippet_a

Now that we have launched Mobly Bundled Snippet, we can get rid of the sl4a-based exampled, so we don't have to do this weird dance of where to get sl4a.

Current monsoon controller code is fairly old.
We need to improve the code with best practices of today and remove unused APIs like execute_sequence_and_measure.

Slow loading snippets fail because _is_app_running() returns True even when nothing is on the other side of the socket.
This leads to race conditions between the snippet app really starting and the connection attempt.

>>> sock = socket.socket(socket.AF_INET, socket.SOCK_STREAM)
>>> print self._port
8822
>>> result = sock.connect_ex(('127.0.0.1', self._port))
>>> print result
0
And yet...
$ nc 127.0.0.1 8822
$

Calling ad.reboot causes error:
[testbedB] 12-27 16:00:18.368 INFO [Test Case] test_hello
[testbedB] 12-27 16:00:18.392 ERROR Process 6561 has terminated. ret: 0, stderr: , stdout:
Traceback (most recent call last):
File "/Users/angli/Developer/mobly/mobly/base_test.py", line 307, in exec_one_testcase
self._setup_test(test_name)
File "/Users/angli/Developer/mobly/mobly/base_test.py", line 157, in _setup_test
self.setup_test()
File "/Users/angli/Desktop/SampleTest.py", line 31, in setup_test
self.ad.reboot()
File "/Users/angli/Developer/mobly/mobly/controllers/android_device.py", line 808, in reboot
self.stop_services()
File "/Users/angli/Developer/mobly/mobly/controllers/android_device.py", line 420, in stop_services
self.stop_adb_logcat()
File "/Users/angli/Developer/mobly/mobly/controllers/android_device.py", line 694, in stop_adb_logcat
utils.stop_standing_subprocess(self._adb_logcat_process)
File "/Users/angli/Developer/mobly/mobly/utils.py", line 353, in stop_standing_subprocess
_assert_subprocess_running(proc)
File "/Users/angli/Developer/mobly/mobly/utils.py", line 304, in _assert_subprocess_running
" stdout: %s" % (proc.pid, ret, err, out))
Error: Process 6561 has terminated. ret: 0, stderr: , stdout:

This test only works in PST timezone, which is not fit for continuous integration.
It needs to be fixed to be timezone agnostic.

Because the convoluted timeout scheme of JsonRpcClientBase.connect.

Right now people have to explicitly say they don't need sl4a in config file, which is odd since most users don't use sl4a.

Let's flip the assumption and make sl4a an optional service that people can request in setup if needed.

The current summary log has limited debug information and is in json.

We need to consider what other useful information we can include in the summary log.
Like information on the invocation status, and device monitoring info.

Also consider using a different format so it's possible to stream this file instead of only writing it out in the end.
This will make the summary log more crash proof and takes less memory.

Right now https://google.github.io/mobly/ has an almost empty index page.
Also should probably pick a more vibrant color...

This causes a stacktrace to be printed every time the user exists sl4a_client.

^D
Traceback (most recent call last):
File "/usr/local/bin/sl4a_shell.py", line 8, in
execfile(file)
File "/Users/angli/Developer/mobly/tools/sl4a_shell.py", line 66, in
Sl4aShell().main(args.serial)
File "/Users/angli/Developer/mobly/mobly/controllers/android_device_lib/jsonrpc_shell_base.py", line 86, in main
self.start_console()
File "/Users/angli/Developer/mobly/mobly/controllers/android_device_lib/jsonrpc_shell_base.py", line 78, in start_console
self._ad.stop_services()
File "/Users/angli/Developer/mobly/mobly/controllers/android_device.py", line 415, in stop_services
self._terminate_sl4a()
File "/Users/angli/Developer/mobly/mobly/controllers/android_device.py", line 721, in _terminate_sl4a
self._terminate_jsonrpc_client(self.sl4a)
File "/Users/angli/Developer/mobly/mobly/controllers/android_device.py", line 574, in _terminate_jsonrpc_client
client.closeSl4aSession()
File "/Users/angli/Developer/mobly/mobly/controllers/android_device_lib/jsonrpc_client_base.py", line 277, in rpc_call
return self._rpc(name, *args)
File "/Users/angli/Developer/mobly/mobly/controllers/android_device_lib/jsonrpc_client_base.py", line 266, in _rpc
raise ProtocolError(ProtocolError.NO_RESPONSE_FROM_SERVER)
mobly.controllers.android_device_lib.jsonrpc_client_base.ProtocolError: No response from server.

We should replace Example 5 with a simpler and more reliable case demonstrating mulit-device interactions.

There are some unit tests that depend on other tests, e.g. test_runner_test depends on android_device_test.

This makes it hard to integrate with bazel, where:

• tests cannot depend on other tests
• all dependencies are explicit, which does not play well with auto-discovery of tests.

Common functionality should be abstracted to util library in the test package (similar to mock_controller).

In the new invocation model, the import_test_modules function and its associated code paths are not needed any more and should be removed.

The current config schema could not satisfy our needs any more.
E.g. Testbed config has controller info and params mixed together, which makes it hard to figure out what resources can be shared in a test bed.

We need to redefine the config so it can carry our feature roadmap.

Some preliminary requirements:

1. Separate controller config and test params should be separate in testbed config.
2. Able to differentiate inherent property of a controller (a phone has bt chip), and property that is assigned (a phone is the caller)
3. Easy to mark whether a controller is sharable, and how sharable it is (can be used in X tests in parallel)
4. Maybe a new format to make it more readable and easier to construct complex configuration.
5. Consider exposing more information of the MH invocation to test cases.

The event dispatcher teardown has some issues:

        if event_name == "EventDispatcherShutdown":
            self._sl4a.closeSl4aSession()

This throws an error, but the error is squashed because it's done inside a Future whose result is never checked.

self.poller.set_result("Done")

This is a private API for Executor implementations and should not be called externally.

 # The polling thread is guaranteed to finish after a max of 60 seconds,
 # so we don't wait here.
 self.executor.shutdown(wait=False)

Don't quite understand the comment here. Doesn't this mean the polling loop might still try to handle events after the server is torn down? And, couldn't this keep the python interpreter open for up to 60 seconds after the test ends?

    if self.sl4a:
        self.sl4a.closeSl4aSession()
        self.sl4a.close()
        self.sl4a = None
    if self.ed:
        self.ed.clean_up()
        self.ed = None

It's awkward and surprising that ed is torn down after its host sl4a session. Actually I'm surprised this works. Won't any sl4a commands clean_up tries to issue fail?

Running the unit tests gives different results without code changes:

$ python setup.py test
running test
running egg_info
writing requirements to mobly.egg-info/requires.txt
writing mobly.egg-info/PKG-INFO
writing top-level names to mobly.egg-info/top_level.txt
writing dependency_links to mobly.egg-info/dependency_links.txt
reading manifest file 'mobly.egg-info/SOURCES.txt'
writing manifest file 'mobly.egg-info/SOURCES.txt'
running build_ext
==================================================================================================================== test session starts =====================================================================================================================
platform linux2 -- Python 2.7.6, pytest-3.0.5, py-1.4.31, pluggy-0.4.0
rootdir: /usr/local/google/home/adorokhine/work/mobly.git, inifile: 
collected 114 items 

tests/mobly_android_device_test.py ................
tests/mobly_asserts_test.py .
tests/mobly_base_class_test.py .....................................................
tests/mobly_logger_test.py .
tests/mobly_records_test.py ................
tests/mobly_sl4a_client_test.py ...........
tests/mobly_test_runner_test.py ...........
tests/mobly_utils_test.py ..F

========================================================================================================================== FAILURES ==========================================================================================================================
_______________________________________________________________________________________________________ MoblyUtilsTest.test_is_port_available_positive _______________________________________________________________________________________________________

self = <mobly_utils_test.MoblyUtilsTest testMethod=test_is_port_available_positive>

    def test_is_port_available_positive(self):
        # Unfortunately, we cannot do this test reliably for SOCK_STREAM
        # because the kernel allow this socket to linger about for some
        # small amount of time.  We're not using SO_REUSEADDR because we
        # are working around behavior on darwin where binding to localhost
        # on some port and then binding again to the wildcard address
        # with SO_REUSEADDR seems to be allowed.
        test_s = socket.socket(socket.AF_INET, socket.SOCK_DGRAM)
        test_s.bind(('localhost', 0))
        port = test_s.getsockname()[1]
        test_s.close()
>       self.assertTrue(utils.is_port_available(port))
E       AssertionError: None is not true

tests/mobly_utils_test.py:51: AssertionError
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! Interrupted: stopping after 1 failures !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
============================================================================================================ 1 failed, 111 passed in 0.48 seconds ============================================================================================================

Ran again:

$ python setup.py test
running test
running egg_info
writing requirements to mobly.egg-info/requires.txt
writing mobly.egg-info/PKG-INFO
writing top-level names to mobly.egg-info/top_level.txt
writing dependency_links to mobly.egg-info/dependency_links.txt
reading manifest file 'mobly.egg-info/SOURCES.txt'
writing manifest file 'mobly.egg-info/SOURCES.txt'
running build_ext
==================================================================================================================== test session starts =====================================================================================================================
platform linux2 -- Python 2.7.6, pytest-3.0.5, py-1.4.31, pluggy-0.4.0
rootdir: /usr/local/google/home/adorokhine/work/mobly.git, inifile: 
collected 114 items 

tests/mobly_android_device_test.py ................
tests/mobly_asserts_test.py .
tests/mobly_base_class_test.py .....................................................
tests/mobly_logger_test.py .
tests/mobly_records_test.py ................
tests/mobly_sl4a_client_test.py ...........
tests/mobly_test_runner_test.py ...........
tests/mobly_utils_test.py .....

================================================================================================================= 114 passed in 0.56 seconds =================================================================================================================

Refactor/rewrite internal controller for callbox and publish it to open source repo.

There have been multiple requests to run regular AndroidInstrumentation tests with mobly.
A sample use cases would be running tests on multiple devices and aggregate results across them.
Let's create utils to make such tasks easy.

'load_snippet' sets a snippet client reference attribute on the AndroidDevice object.
We should remove the attribute after snippet is stopped.

For large testbeds, the time it takes to load snippets and sl4a is quite significant. Would be very nice to be able to easily do this setup in parallel.

I realize this code is part of the user test class, but perhaps we can make a utility method to perform some action for every device in parallel?

Seems like most devices don't have the logpersist service. Atleast not my Nexus 6P with latest Android 7.0.1. Any thoughts on how I can make this work?

$python hello.py -c config.json
[SampleTestBed] 12-11 13:27:46.351 INFO ==========> HelloWorldTest <==========
[SampleTestBed] 12-11 13:27:46.625 ERROR [AndroidDevice|DEVICE_ID_REDACTED] Failed to start adb logcat!
Traceback (most recent call last):
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device.py", line 384, in start_services
    self.start_adb_logcat()
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device.py", line 580, in start_adb_logcat
    self.adb.shell("logpersist.stop --clear")
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device_lib/adb.py", line 137, in adb_call
    return self._exec_adb_cmd(clean_name, arg_str)
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device_lib/adb.py", line 108, in _exec_adb_cmd
    return self._exec_cmd(' '.join((self.adb_str, name, arg_str)))
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device_lib/adb.py", line 105, in _exec_cmd
    raise AdbError(cmd=cmd, stdout=out, stderr=err, ret_code=ret)
AdbError: Error executing adb cmd 'adb -s DEVICE_ID_REDACTED shell logpersist.stop --clear'. ret: 127, stdout: , stderr: /system/bin/sh: logpersist.stop: not found

[SampleTestBed] 12-11 13:27:46.626 ERROR [AndroidDevice|DEVICE_ID_REDACTED] Failed to start some services, abort!
Traceback (most recent call last):
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device.py", line 129, in _start_services_on_ads
    ad.start_services(skip_sl4a=getattr(ad, KEY_SKIP_SL4A, False))
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device.py", line 384, in start_services
    self.start_adb_logcat()
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device.py", line 580, in start_adb_logcat
    self.adb.shell("logpersist.stop --clear")
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device_lib/adb.py", line 137, in adb_call
    return self._exec_adb_cmd(clean_name, arg_str)
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device_lib/adb.py", line 108, in _exec_adb_cmd
    return self._exec_cmd(' '.join((self.adb_str, name, arg_str)))
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device_lib/adb.py", line 105, in _exec_cmd
    raise AdbError(cmd=cmd, stdout=out, stderr=err, ret_code=ret)
AdbError: Error executing adb cmd 'adb -s DEVICE_ID_REDACTED shell logpersist.stop --clear'. ret: 127, stdout: , stderr: /system/bin/sh: logpersist.stop: not found

[SampleTestBed] 12-11 13:27:46.735 ERROR Failed to initialize objects for controller AndroidDevice, abort!
Traceback (most recent call last):
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/test_runner.py", line 376, in register_controller
    objects = create(controller_config)
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device.py", line 82, in create
    _start_services_on_ads(ads)
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device.py", line 129, in _start_services_on_ads
    ad.start_services(skip_sl4a=getattr(ad, KEY_SKIP_SL4A, False))
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device.py", line 384, in start_services
    self.start_adb_logcat()
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device.py", line 580, in start_adb_logcat
    self.adb.shell("logpersist.stop --clear")
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device_lib/adb.py", line 137, in adb_call
    return self._exec_adb_cmd(clean_name, arg_str)
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device_lib/adb.py", line 108, in _exec_adb_cmd
    return self._exec_cmd(' '.join((self.adb_str, name, arg_str)))
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device_lib/adb.py", line 105, in _exec_cmd
    raise AdbError(cmd=cmd, stdout=out, stderr=err, ret_code=ret)
AdbError: Error executing adb cmd 'adb -s DEVICE_ID_REDACTED shell logpersist.stop --clear'. ret: 127, stdout: , stderr: /system/bin/sh: logpersist.stop: not found

[SampleTestBed] 12-11 13:27:46.736 ERROR Failed to setup HelloWorldTest.
Traceback (most recent call last):
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/base_test.py", line 504, in run
    self._setup_class()
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/base_test.py", line 128, in _setup_class
    self.setup_class()
  File "hello.py", line 11, in setup_class
    self.ads = self.register_controller(android_device)
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/test_runner.py", line 376, in register_controller
    objects = create(controller_config)
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device.py", line 82, in create
    _start_services_on_ads(ads)
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device.py", line 129, in _start_services_on_ads
    ad.start_services(skip_sl4a=getattr(ad, KEY_SKIP_SL4A, False))
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device.py", line 384, in start_services
    self.start_adb_logcat()
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device.py", line 580, in start_adb_logcat
    self.adb.shell("logpersist.stop --clear")
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device_lib/adb.py", line 137, in adb_call
    return self._exec_adb_cmd(clean_name, arg_str)
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device_lib/adb.py", line 108, in _exec_adb_cmd
    return self._exec_cmd(' '.join((self.adb_str, name, arg_str)))
  File "build/bdist.macosx-10.11-x86_64/egg/mobly/controllers/android_device_lib/adb.py", line 105, in _exec_cmd
    raise AdbError(cmd=cmd, stdout=out, stderr=err, ret_code=ret)
AdbError: Error executing adb cmd 'adb -s DEVICE_ID_REDACTED shell logpersist.stop --clear'. ret: 127, stdout: , stderr: /system/bin/sh: logpersist.stop: not found

[SampleTestBed] 12-11 13:27:46.736 ERROR Error executing adb cmd 'adb -s DEVICE_ID_REDACTED shell logpersist.stop --clear'. ret: 127, stdout: , stderr: /system/bin/sh: logpersist.stop: not found

[SampleTestBed] 12-11 13:27:46.736 INFO [Test Case] setup_class FAIL
[SampleTestBed] 12-11 13:27:46.737 INFO Summary for test run SampleTestBed@12-11-2016_13-27-46-350: Error 0, Executed 1, Failed 1, Passed 0, Requested 1, Skipped 0

If we make updates to snippet lib and introduce new internal Rpcs or other protocol changes, the client needs to be able to figure out which version of the lib is used and behave accordingly.

If a app is installed but not instrumented properly, SnippetClient._is_app_installed() will throw an error saying the app is not installed.
The error message should correctly reflect the situation.

We should probably throw an error if attempts to register a different package under an existing alias is made?

ad.load_snippet('snippet', 'my.package')
ad.load_snippet('snippet', 'my.other.package') # Raise error