tree faee84f2f20e7f20aefd4f56fd48e0952694ae86
parent b0e640c4e04160a1be8182fa2bed99e8f9927ad0
author Jack He <siyuanh@google.com> 1517014580 -0800
committer Jack He <siyuanh@google.com> 1517363983 -0800

HFP: Only remove state machine when device is unbonded and disconnected

SUMMARY:
* Remove a state machine only when a device is both unbonded and
  disconnected
* Add HeadsetServiceAndStateMachineTest to verify this behavior
* Reject API connect() attempt when device does not have headset UUIDs

DETAILS:
* A state machine is no longer useful if the associated device is no
  longer bonded. However, sometimes the device may get disconnected
  after it is unbonded. Therefore, we should only remove a state machine
  when the device is both unbonded and disconnected.
* Also, we should only allow an HeadsetService.connect() API call when
  there is at least one headset UUID present.
* HeadsetServiceAndStateMachineTest is added to integerate both
  HeadsetService and HeadsetStateMachine to complete a set of
  semi-integration tests. This new set of tests allows us to verify
  whether a connect() API call will eventually trigger a connectHfp() to
  the native interface, utilizing all logic in the middle.
* However, we do need to assume certain timing condition in such
  semi-integration test. 250ms is set as a limit for any state machine
  based message handling, which should be enough for most modern day
  Android devices.
* We have to use AdapterService instead of BluetoothDevice to get bond
  state, device name, and so on since we want to be able to mock these
  methods and BluetoothDevice class is final.
* Thus, this CL also adds a new parameter to HeadsetStateMachine during
  construction so that it has a reference to AdapterService suppplied
  through its constructor. This is to make sure that entire HFP stack
  uses the same reference of AdapterService obtained by HeadsetService.

Bug: 72529611
Test: disconnect and then quickly unpair device
Change-Id: I49ec70d60e257ffd4484e536bdb66d6da7b3b377