bluetooth: bluez: Fixed issue with missing notifications after reconnect.

device/bluetooth/bluetooth_device.cc

 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "device/bluetooth/bluetooth_device.h"
 
 #include <iterator>
 #include <memory>
 #include <string>
 
@@ skipping 452 matching lines @@
   // Connection request should only be made if there are no active
   // connections.
   DCHECK(gatt_connections_.empty());
 
   for (const auto& error_callback : create_gatt_connection_error_callbacks_)
     error_callback.Run(error);
   create_gatt_connection_success_callbacks_.clear();
   create_gatt_connection_error_callbacks_.clear();
 }
 
-void BluetoothDevice::DidDisconnectGatt() {
+void BluetoothDevice::DidDisconnectGatt(bool notifyDeviceChanged) {
+  gatt_services_.clear();
+  device_uuids_.ClearServiceUUIDs();
+  SetGattServicesDiscoveryComplete(false);

[rkc1, 2017/05/15 23:56:59]

What is the purpose of doing this? This doesn't actually notify BlueZ that
service discovery needs to be done again.

How does BlueZ know that Chrome has cleared its services cache and needs to be
notified of all services on the device again?

[ortuno, 2017/05/16 00:04:27]

This line has no effect on BlueZ since the variable that it affects is not used
for IsGattServicesDiscoveryComplete().

[rkc1, 2017/05/16 00:10:09]

Right, and since this function is being called from the BlueZ code and these
lines are added for a "BlueZ" fix, it seems a bit odd that we do this. What is
the purpose of setting that variable?

I am very confused by this fix. Not having a CL description doesn't make things
any easier.

Perja, could you please provide some context on what you are trying to do
exactly?

[ortuno, 2017/05/16 00:40:41]

I agree. It's unclear what the issue was and how this fixes it :(

There are other issue with this patch e.g. it causes mac to clear it services
twice[1]. Maybe we should revert to surface the bug again and fix it?

[1]
https://cs.chromium.org/chromium/src/device/bluetooth/bluetooth_low_energy_de...

[perja, 2017/05/16 08:23:23]

Sure, that sounds like a good approach now.

[perja, 2017/05/16 08:23:23]

It's been a while since I looked into this, but when a disconnect happened the
state in BlueZ was correct and Chrome had the old cache of services. So my fix
was to clear the caches in Chrome. But this was already done for Android so I
moved the three lines from bluetooth_device_android.cc to bluetooth_device.c

+
   // Pending calls to connect GATT are not expected, if they were then
   // DidFailToConnectGatt should have been called.
   DCHECK(create_gatt_connection_error_callbacks_.empty());
 
   // Invalidate all BluetoothGattConnection objects.
   for (BluetoothGattConnection* connection : gatt_connections_) {
     connection->InvalidateConnectionReference();
   }
   gatt_connections_.clear();
-  GetAdapter()->NotifyDeviceChanged(this);
+  if (notifyDeviceChanged)
+    GetAdapter()->NotifyDeviceChanged(this);
 }
 
 void BluetoothDevice::AddGattConnection(BluetoothGattConnection* connection) {
   auto result = gatt_connections_.insert(connection);
   DCHECK(result.second);  // Check insert happened; there was no duplicate.
 }
 
 void BluetoothDevice::RemoveGattConnection(
     BluetoothGattConnection* connection) {
   size_t erased_count = gatt_connections_.erase(connection);
@@ skipping 27 matching lines @@
   if (power < INT8_MIN) {
     return INT8_MIN;
   }
   if (power > INT8_MAX) {
     return INT8_MAX;
   }
   return static_cast<int8_t>(power);
 }
 
 }  // namespace device