Minor IndexedDB refactorings.

third_party/WebKit/Source/modules/indexeddb/IDBObjectStore.h

Index: third_party/WebKit/Source/modules/indexeddb/IDBObjectStore.h
diff --git a/third_party/WebKit/Source/modules/indexeddb/IDBObjectStore.h b/third_party/WebKit/Source/modules/indexeddb/IDBObjectStore.h
index 9349f10380237c95024c5baf0788d9d093d8318c..0bc4ab1ca37fb3cead3e7efe1a0b934ef014b85e 100644
--- a/third_party/WebKit/Source/modules/indexeddb/IDBObjectStore.h
+++ b/third_party/WebKit/Source/modules/indexeddb/IDBObjectStore.h
@@ -59,14 +59,17 @@ public:
     ~IDBObjectStore() { }
     DECLARE_TRACE();
 
+    const IDBObjectStoreMetadata& metadata() const { return m_metadata; }
+    const IDBKeyPath& idbKeyPath() const { return metadata().keyPath; }

[cmumford, 2016/09/19 23:26:48]

IDBObjectStore *is* an "idb" thing, so idbKeyPath is slightly redundant. How
about just "keyPath"?

[pwnall, 2016/09/20 09:11:14]

There's a keyPath() accessor in the WebIDL interface, which does a bit more (on
line 69). I'm slightly worried about having a method with one overload
implementing a method in the spec and the other overload used internally. If you
think there wouldn't be much confusion, I can make the switch.

[cmumford, 2016/09/23 22:21:11]

I'm fine with that - thx for pointing it out.

+
     // Implement the IDBObjectStore IDL
-    int64_t id() const { return m_metadata.id; }
-    const String& name() const { return m_metadata.name; }
+    int64_t id() const { return metadata().id; }
+    const String& name() const { return metadata().name; }
     void setName(const String& name, ExceptionState&);
     ScriptValue keyPath(ScriptState*) const;
     DOMStringList* indexNames() const;
     IDBTransaction* transaction() const { return m_transaction.get(); }
-    bool autoIncrement() const { return m_metadata.autoIncrement; }
+    bool autoIncrement() const { return metadata().autoIncrement; }

[cmumford, 2016/09/19 23:26:48]

What is the rationale for switching m_metadata to metadata()?

[pwnall, 2016/09/20 09:11:14]

The idea is that we can switch the metadata representation (e.g., to a
RefPtr<IDBObjectStoreMetadata>, which would share the underlying structure with
the database metadata) without causing as much code churn.

[cmumford, 2016/09/23 22:21:11]

So you'd be switching from a before that is one of:

  m_metadata.autoIncrement;
  metadata().autoIncrement;

to one of:

  m_metadata->autoIncrement;
  metadata()->autoIncrement;

To me it just seems just a bit more complicated, but I don't have strong
feelings on the subject. How about getting dmurph's opinion and pick one way and
go with it.

[pwnall, 2016/09/23 22:33:04]

Sorry I wasn't clear earlier. metadata() would always return const&, so the
calls to it wouldn't need to change.

 
     IDBRequest* openCursor(ScriptState*, const ScriptValue& range, const String& direction, ExceptionState&);
     IDBRequest* openKeyCursor(ScriptState*, const ScriptValue& range, const String& direction, ExceptionState&);
@@ -96,16 +99,13 @@ public:
     // Used internally and by InspectorIndexedDBAgent:
     IDBRequest* openCursor(ScriptState*, IDBKeyRange*, WebIDBCursorDirection, WebIDBTaskType = WebIDBTaskTypeNormal);
 
-    void markDeleted() { m_deleted = true; }
+    void markDeleted();
     bool isDeleted() const { return m_deleted; }
     void abort();
     void transactionFinished();
 
-    const IDBObjectStoreMetadata& metadata() const { return m_metadata; }
     void setMetadata(const IDBObjectStoreMetadata& metadata) { m_metadata = metadata; }
 
-    typedef HeapVector<Member<IDBKey>> IndexKeys;
-
     // Used by IDBIndex::setName:
     bool containsIndex(const String& name) const
     {
@@ -127,7 +127,14 @@ private:
     Member<IDBTransaction> m_transaction;
     bool m_deleted = false;
 
-    typedef HeapHashMap<String, Member<IDBIndex>> IDBIndexMap;
+    // Caches the IDBIndex instances returned by the index() method.
+    // The spec requires that an object store's index() returns the same
+    // IDBIndex instance for a specific index, so this cache is necessary
+    // for correctness.
+    //
+    // index() throws for completed/aborted transactions, so this is not used
+    // after a transaction is finished, and can be cleared.
+    using IDBIndexMap = HeapHashMap<String, Member<IDBIndex>>;
     IDBIndexMap m_indexMap;
 
     // Used to mark indexes created in an aborted upgrade transaction as