[chromedriver] Load the automation extension as a component extension.

chrome/test/chromedriver/chrome_launcher.cc

Index: chrome/test/chromedriver/chrome_launcher.cc
diff --git a/chrome/test/chromedriver/chrome_launcher.cc b/chrome/test/chromedriver/chrome_launcher.cc
index 36e2a1c345f646b7ae04c4a5e659b3ae25af9811..76bf9a8e83b566c57800284457037f895a47fbdd 100644
--- a/chrome/test/chromedriver/chrome_launcher.cc
+++ b/chrome/test/chromedriver/chrome_launcher.cc
@@ -520,12 +520,8 @@ Status ProcessExtensions(const std::vector<std::string>& extensions,
     Status status = UnpackAutomationExtension(temp_dir, &automation_extension);
     if (status.IsError())
       return status;
-    if (command->HasSwitch("disable-extensions")) {
-      command->AppendSwitchNative("load-component-extension",
-                                  automation_extension.value());
-    } else {
-      extension_paths.push_back(automation_extension.value());
-    }
+    command->AppendSwitchNative("load-component-extension",

[chrisgao (Use stgao instead), 2013/08/28 17:01:58]

If the users pass in an argument "--load-component-extension=path", it fails
like the current case in the bug.
I remember there was a crash if use "load-component-extension" without
"disable-extensions", although not 100% sure which version of chrome it was.

How about doing it this way?
If "load-component-extension" is passed in, we update its value instead of
adding a switch with the same name.

+                                automation_extension.value());
   }
 
   if (extension_paths.size()) {