[ChromeDriver] Implementing PerfLoggingPrefs for perf log

chrome/test/chromedriver/chrome/devtools_client_impl.cc

Index: chrome/test/chromedriver/chrome/devtools_client_impl.cc
diff --git a/chrome/test/chromedriver/chrome/devtools_client_impl.cc b/chrome/test/chromedriver/chrome/devtools_client_impl.cc
index 8bb0875c6939d32e790faecb4fc450398e750161..dd9bcbaa033101a5d00678495019b302bf9dfef3 100644
--- a/chrome/test/chromedriver/chrome/devtools_client_impl.cc
+++ b/chrome/test/chromedriver/chrome/devtools_client_impl.cc
@@ -474,9 +474,14 @@ bool ParseInspectorMessage(
   } else if (message_dict->GetInteger("id", &id)) {
     base::DictionaryValue* unscoped_error = NULL;
     base::DictionaryValue* unscoped_result = NULL;
+    // As per Chromium issue 392577, DevTools does not necessarily return a
+    // "result" dictionary for every valid response. In particular,
+    // Tracing.start and Tracing.end command responses do not contain one.
+    // So, if neither "error" nor "result" keys are present, just provide
+    // a blank result dictionary.
     if (!message_dict->GetDictionary("error", &unscoped_error) &&
         !message_dict->GetDictionary("result", &unscoped_result))
-      return false;
+      unscoped_result = new base::DictionaryValue();
 
     *type = kCommandResponseMessageType;
     command_response->id = id;