Make sync's change processors sequence-affine.

components/sync/driver/generic_change_processor.h

Index: components/sync/driver/generic_change_processor.h
diff --git a/components/sync/driver/generic_change_processor.h b/components/sync/driver/generic_change_processor.h
index 96f580e34679faca46d1c818afb3cc9876787cc1..f8e5c530266b143afa557ea4780e54922f332386 100644
--- a/components/sync/driver/generic_change_processor.h
+++ b/components/sync/driver/generic_change_processor.h
@@ -14,7 +14,7 @@
 #include "base/compiler_specific.h"
 #include "base/macros.h"
 #include "base/memory/weak_ptr.h"
-#include "base/threading/non_thread_safe.h"
+#include "base/sequence_checker.h"
 #include "components/sync/model/attachments/attachment_service.h"
 #include "components/sync/model/attachments/attachment_service_proxy.h"
 #include "components/sync/model/attachments/attachment_store.h"
@@ -38,17 +38,16 @@ class Entry;
 typedef std::vector<SyncData> SyncDataList;
 
 // Datatype agnostic change processor. One instance of GenericChangeProcessor
-// is created for each datatype and lives on the datatype's thread. It then
+// is created for each datatype and lives on the datatype's sequence. It then
 // handles all interaction with the sync api, both translating pushes from the
 // local service into transactions and receiving changes from the sync model,
 // which then get converted into SyncChange's and sent to the local service.
 //
 // As a rule, the GenericChangeProcessor is not thread safe, and should only
-// be used on the same thread in which it was created.
+// be used on the same sequence in which it was created.
 class GenericChangeProcessor : public ChangeProcessor,
                                public SyncChangeProcessor,
-                               public AttachmentService::Delegate,
-                               public base::NonThreadSafe {

[stanisc, 2016/12/21 18:48:09]

I think it would be nice to have a new class for sequence checking that is a
replacement for NonThreadSafe. That would be more convenient (less code, easier
to read) than using sequence_checkher_ member.

[gab, 2016/12/21 18:52:17]

Note that composition is prefered to inheritance by style guide, even for this,
as was discussed here:
https://groups.google.com/a/chromium.org/d/topic/chromium-dev/TivgMWBuTKI/dis...

Also, FWIW, sequences provide thread-safety. So maybe we should just update
NonThreadSafe to be sequence-affine instead of thread-affine (requires survey of
codebase to make sure no one uses this in thread-affine scenarios...).

+                               public AttachmentService::Delegate {
  public:
   // Create a change processor for |type| and connect it to the syncer.
   // |attachment_store| can be null which means that datatype will not use sync
@@ -147,6 +146,8 @@ class GenericChangeProcessor : public ChangeProcessor,
   void NotifyLocalChangeObservers(const syncable::Entry* current_entry,
                                   const SyncChange& change);
 
+  base::SequenceChecker sequence_checker_;
+
   const ModelType type_;
 
   // The SyncableService this change processor will forward changes on to.