Support needed to measure user and service traffic in Chrome.

chrome/browser/profiles/profile_manager.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 "chrome/browser/profiles/profile_manager.h"
 
 #include <set>
 
 #include "base/bind.h"
 #include "base/command_line.h"
@@ skipping 441 matching lines @@
       callback.Run(profile, Profile::CREATE_STATUS_INITIALIZED);
     } else {
       // Profile is either already in the process of being created, or new.
       // Add callback to the list.
       info->callbacks.push_back(callback);
     }
   }
 }
 
 bool ProfileManager::IsValidProfile(Profile* profile) {
+  // The profile_manager() returns NULL when running unit tests.
+  if (!g_browser_process->profile_manager())

[Lei Zhang, 2015/09/02 18:14:43]

Curious, why is this even needed? What happens if you leave the method as it
was?

[amohammadkhan, 2015/09/03 04:31:19]

It was causing segmentation fault in tests. It seems in tests profileManager is
not instantiated, so profiles_info_ in next line should not be reached.

[Lei Zhang, 2015/09/03 04:47:11]

I'm skeptical this is the right thing to do.
ProfileManager::ShutdownSessionServices() is the only other place that checks to
see if a ProfileManager exists, but that's because it dereferences the
ProfileManager pointer. You are not doing that here, so it's not obvious that
this is correct.

Can you be specific about what tests crashed?

[amohammadkhan, 2015/09/03 23:10:35]

ChromeNetworkDelegateTest.DataUseMeasurementUserTest was crashing. This crash
didn't happen in similar tests because they don't attach
content::ResourceRequestInfo to request but this test does. The difference is
that in this function
(https://code.google.com/p/chromium/codesearch#chromium/src/chrome/browser/net...)
other tests return but the crashing test runs BrowserThread::PostTask. 
At first I had used a negative processID and I was checking the processID to see
if its negative to return in that functions but later I changed it later to
current implementation based on checking in ProfileManager because I wasn't sure
that checking for negative processID is a good idea or not.

[Lei Zhang, 2015/09/04 01:18:06]

The right fix is to do the check in the caller, which is
NotifyEPMRequestStatus() in
chrome/browser/net/chrome_extensions_network_delegate.cc, or alternatively, to
create a ProfileManager in your test.

I don't know which is "more correct" so ask the networking code experts.

[amohammadkhan, 2015/09/04 17:25:11]

Done.

+    return false;
   for (ProfilesInfoMap::iterator iter = profiles_info_.begin();
        iter != profiles_info_.end(); ++iter) {
     if (iter->second->created) {
       Profile* candidate = iter->second->profile.get();
       if (candidate == profile ||
           (candidate->HasOffTheRecordProfile() &&
            candidate->GetOffTheRecordProfile() == profile)) {
         return true;
       }
     }
@@ skipping 1004 matching lines @@
   }
 
   FinishDeletingProfile(profile_to_delete_path, new_active_profile_path);
   if (!original_callback.is_null())
     original_callback.Run(loaded_profile, status);
 }
 
 ProfileManagerWithoutInit::ProfileManagerWithoutInit(
     const base::FilePath& user_data_dir) : ProfileManager(user_data_dir) {
 }