IndexedDB: Prevent terminated worker from hanging connections

Source/modules/indexeddb/IDBDatabase.cpp

Index: Source/modules/indexeddb/IDBDatabase.cpp
diff --git a/Source/modules/indexeddb/IDBDatabase.cpp b/Source/modules/indexeddb/IDBDatabase.cpp
index 8106fdc1cfbc22979f067cc5a57ffcf8ec7d7dd2..2e7adb4fb7b236e7e8acf8e40fa4f91eba2e66eb 100644
--- a/Source/modules/indexeddb/IDBDatabase.cpp
+++ b/Source/modules/indexeddb/IDBDatabase.cpp
@@ -331,8 +331,10 @@ void IDBDatabase::closeConnection()
     ASSERT(m_closePending);
     ASSERT(m_transactions.isEmpty());
 
-    m_backend->close();
-    m_backend.clear();
+    if (m_backend) {
+        m_backend->close();
+        m_backend.clear();
+    }
 
     if (m_contextStopped || !executionContext())
         return;
@@ -402,10 +404,16 @@ bool IDBDatabase::hasPendingActivity() const
 
 void IDBDatabase::stop()
 {
-    // Stop fires at a deterministic time, so we need to call close in it.
-    close();
-
     m_contextStopped = true;
+
+    // Normally, connections are not closed until all transactions are
+    // finished, which requires a round-trip to the back end. In a worker
+    // context, no more messages will be received once the context is torn
+    // down. So in the case of context stop, just close immediately.

[alecflett, 2014/01/15 23:40:23]

The comment could be generalized here - I think something along the lines of "We
know the context is stopping, so we can proactively close the backend without
waiting for an onabort/onclose event from the backend."

meaning, there's no other "Normally" behavior now, right? since
IDBDatabase::stop is actually closing the backend unconditionally...

[jsbell, 2014/01/15 23:56:56]

By "Normally" I meant "on explicit close() or GC", i.e. not context shutdown.
But I see what you mean, the comment is implying knowledge of the history of the
code. I'll fix.

+    if (m_backend) {
+        m_backend->close();
+        m_backend.clear();
+    }
 }
 
 const AtomicString& IDBDatabase::interfaceName() const