[IndexedDB] Added IndexedDBCursor lifetime tracing

content/browser/indexed_db/indexed_db_cursor.cc

Index: content/browser/indexed_db/indexed_db_cursor.cc
diff --git a/content/browser/indexed_db/indexed_db_cursor.cc b/content/browser/indexed_db/indexed_db_cursor.cc
index 005d541163d54b57cf7a6ab5c3db5ade2ba5a61c..1e62d6d87586949c3fe14c4b0efded8587438029 100644
--- a/content/browser/indexed_db/indexed_db_cursor.cc
+++ b/content/browser/indexed_db/indexed_db_cursor.cc
@@ -65,9 +65,16 @@ IndexedDBCursor::IndexedDBCursor(
       transaction_(transaction),
       cursor_(std::move(cursor)),
       closed_(false),
-      ptr_factory_(this) {}
+      ptr_factory_(this) {
+  IDB_ASYNC_TRACE_BEGIN("IndexedDBCursor::lifetime", this);
+}
 
-IndexedDBCursor::~IndexedDBCursor() {}
+IndexedDBCursor::~IndexedDBCursor() {
+  if (!closed_) {

[cmumford, 2017/03/14 16:30:56]

Nit: How about Having ::Close() to a:

  if (closed_)
    return;

and then remove the check from the destructor. C++ call tail optimization should
make it equally efficient.

[dmurph, 2017/03/29 23:02:19]

Done.

+    // Call to make sure we complete our lifetime trace.
+    Close();
+  }
+}
 
 void IndexedDBCursor::Continue(std::unique_ptr<IndexedDBKey> key,
                                std::unique_ptr<IndexedDBKey> primary_key,
@@ -268,6 +275,7 @@ leveldb::Status IndexedDBCursor::PrefetchReset(int used_prefetches,
 }
 
 void IndexedDBCursor::Close() {
+  IDB_ASYNC_TRACE_END("IndexedDBCursor::lifetime", this);

[cmumford, 2017/03/14 16:30:56]

::lifetime is better called from the destructor since there's no guarantee how
long the owner may hold on to a cursor after it is closed. I realize that it's
the end of it's "effective" lifetime, but since the object still "lives" on, the
name or call site seem out of place.

[dmurph, 2017/03/29 23:02:19]

Changed to ::open

   IDB_TRACE("IndexedDBCursor::Close");
   closed_ = true;
   cursor_.reset();