Use a placeholder instead of the default plugin for missing plug-ins on Mac and Linux.

content/browser/plugin_service.h

Index: content/browser/plugin_service.h
diff --git a/content/browser/plugin_service.h b/content/browser/plugin_service.h
index f662b33f0a78c825b00518e76e787a1eb558ae16..d9fbc8b612e8c7b1f2b9b2405e8d6dfd951c48ea 100644
--- a/content/browser/plugin_service.h
+++ b/content/browser/plugin_service.h
@@ -121,7 +121,17 @@ class CONTENT_EXPORT PluginService
   void CancelOpenChannelToNpapiPlugin(PluginProcessHost::Client* client);
 
   // Gets the plugin in the list of plugins that matches the given url and mime
-  // type. Must be called on the FILE thread if |use_stale| is NULL.
+  // type. Returns true if the data is frome a stale plugin list, false if it
+  // is up to date. This can be called from any thread.
+  bool GetPluginInfoArray(const GURL& url,

[jam, 2011/09/30 17:01:21]

is it really necessary to add a wrapper function around PluginList instead of
having callers call it directly?

[Bernhard Bauer, 2011/09/30 17:22:21]

That's at least Robert's plan for the future. I copied this method from
http://codereview.chromium.org/8071013/.

[jam, 2011/09/30 17:40:19]

I see. it's a bit unfortunate to have these wrappers, but ok.

+                          const std::string& mime_type,
+                          bool allow_wildcard,
+                          std::vector<webkit::WebPluginInfo>* info,
+                          std::vector<std::string>* actual_mime_types);
+
+  // Gets plugin info for an individual plugin and filters the plugins using
+  // the |context| and renderer IDs. This will report whether the data is stale
+  // via |is_stale| and returns whether or not the plugin can be found.
   bool GetPluginInfo(int render_process_id,
                      int render_view_id,
                      const content::ResourceContext& context,
@@ -138,13 +148,18 @@ class CONTENT_EXPORT PluginService
   void RefreshPluginList();
 
   // Asynchronously loads plugins if necessary and then calls back to the
-  // provided function on the calling MessageLoop on completion.
+  // provided function on the calling MessageLoop with the list of plugins
+  // on completion.
   void GetPlugins(const GetPluginsCallback& callback);
 
   // Asynchronously loads the list of plugin groups if necessary and then calls
   // back to the provided function on the calling MessageLoop on completion.
   void GetPluginGroups(const GetPluginGroupsCallback& callback);
 
+  // Asynchronously loads plugins if necessary and then calls back to the
+  // provided function on the calling MessageLoop on completion.
+  void EnsurePluginsLoaded(const base::Closure& callback);

[jam, 2011/09/30 17:01:21]

this function doesn't seem necessary since it can already be done with existing
methods (the fewer methods there are, the better!). if someone just wants to
know that plugins are loaded, they can call GetPlugins and ignore the result.

[Bernhard Bauer, 2011/09/30 17:22:21]

I find it a bit awkward having the list of plug-ins as an additional parameter
and then ignoring it. ChromeRenderMessageFilter::GetPluginInfo already has
enough parameters as it is (and I was actually hitting the base::Callback limits
with it!).

[jam, 2011/09/30 17:40:19]

we already do this (i.e. BufferedRersourceHandler)
we ignore parameters all the time in various methods, so that's nothing new.

For a base class like PluginService that's used by a lot of code, having the
fewest number of methods is useful, because it makes it easy for developers not
familiar with it to quickly scan all the public methods. in this case, i think
the cost of having a CRMF method take an unused parameter is low vs adding
another function to PS

[Bernhard Bauer, 2011/10/05 13:34:33]

ok, done.

+
   // Tells all the renderer processes to throw away their cache of the plugin
   // list, and optionally also reload all the pages with plugins.
   // NOTE: can only be called on the UI thread.