Move wifi_data_provider_chromeos -> content

content/browser/geolocation/device_data_provider.h

 // 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.
 
 // A device data provider provides data from the device that is used by a
 // NetworkLocationProvider to obtain a position fix. This data may be either
 // cell radio data or wifi data. For a given type of data, we use a singleton
 // instance of the device data provider, which is used by multiple
 // NetworkLocationProvider objects.
 //
@@ skipping 107 matching lines @@
     DCHECK(CalledOnClientThread());
     return listeners_.erase(listener) == 1;
   }
 
   bool has_listeners() const {
     DCHECK(CalledOnClientThread());
     return !listeners_.empty();
   }
 
  protected:
-  virtual ~DeviceDataProviderImplBase() {
-    DCHECK(CalledOnClientThread());

[John Knottenbelt, 2013/01/14 14:12:31]

According to jam@, there are some tests that hold a reference to the provider
causing the last reference to be lost on the UI thread instead of the
geolocation thread.

It seems a shame to remove the guarantee that this class will always be
destroyed on the same thread its used on. Can you take a look and see if there
are any obvious fixes?

[stevenjb, 2013/01/14 18:02:38]

jam@, any thoughts? I just copied this, but I'm really not familiar wiht how
it's used. I could put the DCHECK back and see what fails...

That said, a DCHECK like this in the destructor is kind of unusual, and this
does derive from base::RefCountedThreadSafe, which implies to me that destroying
this on another thread should be valid?

[jam, 2013/01/14 18:57:21]

this check is just invalid because as you say the destruction of this object can
happen on different threads because of the ref counted. normally if we want to
add a dcheck like this, a traits on the object would ensure that it destructs on
a sepecific message loop. that can be done, but it seems out of scope of this
change.

the issue is not with a specific test that holds a reference. it's just that we
never used the cros implementation in tests for a long time, and that one seems
to post tasks differently from other tests which changes the ref counting

-  }
+  virtual ~DeviceDataProviderImplBase() {}
 
   // Calls DeviceDataUpdateAvailable() on all registered listeners.
   typedef std::set<ListenerInterface*> ListenersSet;
   void NotifyListeners() {
     // Always make the notify callback via a posted task, so we can unwind
     // callstack here and make callback without causing client re-entrancy.
     client_loop_->PostTask(FROM_HERE, base::Bind(
         &DeviceDataProviderImplBase<DataType>::NotifyListenersInClientLoop,
         this));
   }
@@ skipping 159 matching lines @@
   scoped_refptr<DeviceDataProviderImplBase<DataType> > impl_;
 
   DISALLOW_COPY_AND_ASSIGN(DeviceDataProvider);
 };
 
 typedef DeviceDataProvider<WifiData> WifiDataProvider;
 
 }  // namespace content
 
 #endif  // CONTENT_BROWSER_GEOLOCATION_DEVICE_DATA_PROVIDER_H_