Add PluginDataRemover.

chrome/browser/plugin_service.cc

Index: chrome/browser/plugin_service.cc
diff --git a/chrome/browser/plugin_service.cc b/chrome/browser/plugin_service.cc
index cc6408fb02b186fbfb0dc887160b4df7e0d98236..86bed6239f3368d16e7110fe7ec27977c67941b2 100644
--- a/chrome/browser/plugin_service.cc
+++ b/chrome/browser/plugin_service.cc
@@ -226,26 +226,28 @@ PluginProcessHost* PluginService::FindOrStartPluginProcess(
     const FilePath& plugin_path) {
   DCHECK(BrowserThread::CurrentlyOn(BrowserThread::IO));
 
-  PluginProcessHost *plugin_host = FindPluginProcess(plugin_path);
+  if (plugin_path.value().empty())

[jam, 2010/11/12 20:44:58]

curious, any reason to change this logic?  it seems redundant to have this check
since GetPluginInfoByPath will return null.  if the dcheck/notreached can fire
in normal circumstances, then perhaps no need to have it.  if it does fire, when
does that happen?

+    return NULL;
+
+  PluginProcessHost* plugin_host = FindPluginProcess(plugin_path);
   if (plugin_host)
     return plugin_host;
 
   WebPluginInfo info;
   if (!NPAPI::PluginList::Singleton()->GetPluginInfoByPath(
-          plugin_path, &info)) {
-    DCHECK(false);
+      plugin_path, &info)) {
+    NOTREACHED();
     return NULL;
   }
 
   // This plugin isn't loaded by any plugin process, so create a new process.
-  plugin_host = new PluginProcessHost();
-  if (!plugin_host->Init(info, ui_locale_)) {
-    DCHECK(false);  // Init is not expected to fail
-    delete plugin_host;
+  scoped_ptr<PluginProcessHost> new_host(new PluginProcessHost());
+  if (!new_host->Init(info, ui_locale_)) {
+    NOTREACHED();  // Init is not expected to fail
     return NULL;
   }
 
-  return plugin_host;
+  return new_host.release();
 }
 
 void PluginService::OpenChannelToPlugin(