Port messages sent by WebIDBFactoryImpl to Mojo.

content/child/indexed_db/indexed_db_database_callbacks_impl.h

Index: content/child/indexed_db/indexed_db_database_callbacks_impl.h
diff --git a/content/child/indexed_db/indexed_db_database_callbacks_impl.h b/content/child/indexed_db/indexed_db_database_callbacks_impl.h
new file mode 100644
index 0000000000000000000000000000000000000000..1e431e7df200c06ece10610538b462536d6014b4
--- /dev/null
+++ b/content/child/indexed_db/indexed_db_database_callbacks_impl.h
@@ -0,0 +1,44 @@
+// Copyright 2016 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.
+
+#ifndef CONTENT_CHILD_INDEXED_DB_INDEXED_DB_DATABASE_CALLBACKS_IMPL_H_
+#define CONTENT_CHILD_INDEXED_DB_INDEXED_DB_DATABASE_CALLBACKS_IMPL_H_
+
+#include "content/common/indexed_db/indexed_db.mojom.h"
+#include "mojo/public/cpp/bindings/associated_binding.h"
+
+namespace blink {
+class WebIDBDatabaseCallbacks;
+}
+
+namespace content {
+
+class IndexedDBDatabaseCallbacksImpl
+    : public indexed_db::mojom::DatabaseCallbacks {
+ public:
+  explicit IndexedDBDatabaseCallbacksImpl(
+      blink::WebIDBDatabaseCallbacks* callbacks);
+  ~IndexedDBDatabaseCallbacksImpl() override;
+
+  void Bind(indexed_db::mojom::DatabaseCallbacksAssociatedPtrInfo* ptr_info,

[Ken Rockot(use gerrit already), 2016/09/28 20:06:44]

Any reason why you chose to have this take a PtrInfo* instead of a Request? I
generally discourage this because it's a less flexible interface -- you can
always create a new pipe to get a Request and Ptr/PtrInfo, but if you already
have a Request, you can't use an method that takes a Ptr*/PtrInfo*.

In fact I'd really like to delete all the Binding/AssociatedBinding methods
which take a Ptr*/PtrInfo*. But if there are legitimate reasons to keep them
which I'm overlooking (IMHO less typing does not trump having a minimal yet
robust interface), I'd like to hear about them!

[Reilly Grant (use Gerrit), 2016/09/29 06:44:50]

It looks like the Ptr -> Request flow works when you want to pass the request
(mojo::GetProxy calls CreateAssociatedInterface with WILL_PASS_REQUEST) but if
you want to pass the Ptr you actually have to pass a PtrInfo and there's no
method to convert a PtrInfo to a Request (without calling MakeAssociatedRequest
which is not typesafe because you have to pass an untyped
ScopedInterfaceEndpointHandle). AssociatedBinding::Bind calls
CreateAssociatedInterface with WILL_PASS_PTR so that's what I used.

[Ken Rockot(use gerrit already), 2016/09/29 22:37:50]

I see, so it's out of convenience because AssociatedBinding already essentially
does GetProxy but with WILL_PASS_PTR instead.

My point still stands, a method which takes a request is strictly more flexible
than one which forces you to use a request it generates internally, but I
understand you using this given the APIs we have available. So, not your problem
and this is fine as-is for now.

[Reilly Grant (use Gerrit), 2016/09/30 07:24:20]

I agree, and liked your other suggestion so I've changed this to take a request.

+            mojo::AssociatedGroup* associated_group);
+  void OnConnectionError();
+
+  // indexed_db::mojom::DatabaseCallbacks implementation
+  void ForcedClose() override;
+  void VersionChange(int64_t old_version, int64_t new_version) override;
+  void Abort(int64_t transaction_id,
+             int32_t code,
+             const base::string16& message) override;
+  void Complete(int64_t transaction_id) override;
+
+ private:
+  scoped_refptr<base::SingleThreadTaskRunner> callback_runner_;
+  blink::WebIDBDatabaseCallbacks* callbacks_;
+  mojo::AssociatedBinding<indexed_db::mojom::DatabaseCallbacks> binding_;
+};
+
+}  // namespace content
+
+#endif  // CONTENT_CHILD_INDEXED_DB_INDEXED_DB_DATABASE_CALLBACKS_IMPL_H_