Take extensions out of Profile into a profile-keyed service, ExtensionSystem.

chrome/browser/profiles/off_the_record_profile_impl_unittest.cc

Index: chrome/browser/profiles/off_the_record_profile_impl_unittest.cc
diff --git a/chrome/browser/profiles/off_the_record_profile_impl_unittest.cc b/chrome/browser/profiles/off_the_record_profile_impl_unittest.cc
index 18d82ec0b7b2e22a2a515469e2520e59a361cd98..21114bd42e10f6b1e2698bd315563e6b3a9bf2df 100644
--- a/chrome/browser/profiles/off_the_record_profile_impl_unittest.cc
+++ b/chrome/browser/profiles/off_the_record_profile_impl_unittest.cc
@@ -7,6 +7,7 @@
 #include "chrome/browser/net/ssl_config_service_manager.h"
 #include "chrome/browser/prefs/browser_prefs.h"
 #include "chrome/browser/prefs/scoped_user_pref_update.h"
+#include "chrome/browser/profiles/profile_dependency_manager.h"
 #include "chrome/common/pref_names.h"
 #include "chrome/test/base/browser_with_test_window_test.h"
 #include "chrome/test/base/testing_browser_process.h"
@@ -146,6 +147,9 @@ TEST_F(OffTheRecordProfileImplTest, GetHostZoomMap) {
   scoped_ptr<Profile> child_profile(
       new OffTheRecordProfileImpl(parent_profile.get()));
 
+  ProfileDependencyManager::GetInstance()->CreateProfileServices(
+      child_profile.get(), false);

[Elliot Glaysher, 2012/02/09 20:10:24]

Shouldn't this be a call to OTRProfileImpl::Init()?

[Yoyo Zhou, 2012/02/16 21:49:20]

It certainly seems like it should be, but I recall that the test crashes when I
do that. I can try that again.

(FYI, this test fails a CHECK in Debug without this change because this OTR
profile is getting created abnormally and doesn't have Init called on it, so
services are destroyed in its destructor without having being created.)

[Yoyo Zhou, 2012/02/22 03:02:05]

Here is the crash with explicitly calling Init (I think it's the same if we use
parent_profile->CreateOffTheRecordProfile):


[32530:32530:0217/181204:7629766454379:FATAL:browser_thread_impl.cc(344)] Check
failed: thread. 
Backtrace:
	base::debug::StackTrace::StackTrace() [0x3dbd1f6]
	logging::LogMessage::~LogMessage() [0x3de4c30]
	content::BrowserThread::UnsafeGetMessageLoopForThread() [0x305ba56]
	ProxyServiceFactory::CreateProxyConfigService() [0x19ec59c]
	ProfileIOData::InitializeOnUIThread() [0x1b1ac5a]
	OffTheRecordProfileIOData::Handle::LazyInitialize() [0x1af4f6b]
	OffTheRecordProfileIOData::Handle::GetChromeURLDataManagerBackendGetter()
[0x1af485f]
	OffTheRecordProfileImpl::GetChromeURLDataManager() [0x1af34ab]
	OffTheRecordProfileImpl::Init() [0x1af1b54]
	OffTheRecordProfileImplTest_GetHostZoomMap_Test::TestBody() [0xc4c5ac]

[Miranda Callahan, 2012/02/27 21:59:05]

This looks like one of those common threading errors in tests. :-(  What if you
explicitly create and start up a UI thread, as is done in
template_url_service_test:

http://code.google.com/codesearch#OAMlx_jo-ck/src/chrome/browser/search_engin...

http://code.google.com/codesearch#OAMlx_jo-ck/src/chrome/browser/search_engin...

[Yoyo Zhou, 2012/02/28 02:03:34]

Alas, BrowserWithTestWindowTest (which this test inherits from) already does
this. Do you know if there's a way to make sure this test thinks it's running on
the UI thread?

+
   // Prepare child host zoom map.
   HostZoomMap* child_zoom_map = child_profile->GetHostZoomMap();
   ASSERT_TRUE(child_zoom_map);
@@ -162,12 +166,12 @@ TEST_F(OffTheRecordProfileImplTest, GetHostZoomMap) {
 
   EXPECT_NE(parent_zoom_map->GetZoomLevel(host),
             child_zoom_map->GetZoomLevel(host)) <<
-                "Child change must not propaget to parent.";
+                "Child change must not propagate to parent.";
 
   parent_zoom_map->SetZoomLevel(host, zoom_level_40);
   ASSERT_EQ(parent_zoom_map->GetZoomLevel(host), zoom_level_40);
 
   EXPECT_EQ(parent_zoom_map->GetZoomLevel(host),
             child_zoom_map->GetZoomLevel(host)) <<
-                "Parent change should propaget to child.";
+                "Parent change should propagate to child.";
 }