[Sync] Refactor sync datatype error handling.

chrome/browser/sync/glue/model_associator.h

Index: chrome/browser/sync/glue/model_associator.h
diff --git a/chrome/browser/sync/glue/model_associator.h b/chrome/browser/sync/glue/model_associator.h
index eaa2a45e3021c159c3e39514b6e73fd937914310..b49215ea4447462d7e74b235fdad9e624c53cc65 100644
--- a/chrome/browser/sync/glue/model_associator.h
+++ b/chrome/browser/sync/glue/model_associator.h
@@ -9,6 +9,8 @@
 #include "base/basictypes.h"
 #include "chrome/browser/sync/syncable/model_type.h"
 
+class SyncError;
+
 namespace sync_api {
 class BaseNode;
 }
@@ -27,10 +29,10 @@ class AssociatorInterface {
   // should be identical and corresponding. Returns true on
   // success. On failure of this step, we should abort the sync
   // operation and report an error to the user.
-  virtual bool AssociateModels() = 0;
+  virtual bool AssociateModels(SyncError* error) = 0;

[akalin, 2011/07/25 22:43:13]

Change these to return SyncErrors too?  (and subclasses)

[Nicolas Zea, 2011/07/25 23:08:19]

Since this is already an established (and now deprecated API) that relies on
[Dis]AssociateModels returning true/false on success/error, and because we're
planning on switching everything over to the new API anyways, I'd prefer to do
the less intrusive change for now (just adding an extra parameter), if that's
ok. I've added a TODO for once we have less datatypes reliant on this approach.

[akalin, 2011/07/25 23:49:32]

Yeah, for practicality's sake I guess we can leave it like this.  Although I
have a feeling the old-style APIs are gonna stick around for a while. :|

 
   // Clears all the associations between the chrome and sync models.
-  virtual bool DisassociateModels() = 0;
+  virtual bool DisassociateModels(SyncError* error) = 0;
 
   // The has_nodes out parameter is set to true if the sync model has
   // nodes other than the permanent tagged nodes.  The method may