[Linux] Call g_thread_init() in UtilityThreadImpl::OnLoadPlugins().

content/utility/utility_thread_impl.cc

 // Copyright (c) 2011 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #include "content/utility/utility_thread_impl.h"
 
 #include <stddef.h>
 
+#if defined(TOOLKIT_USES_GTK)
+#include <gtk/gtk.h>
+#endif
+
 #include "base/file_path.h"
 #include "base/memory/scoped_vector.h"
 #include "content/common/child_process.h"
 #include "content/common/child_process_messages.h"
 #include "content/common/indexed_db_key.h"
 #include "content/common/utility_messages.h"
 #include "content/common/webkitplatformsupport_impl.h"
 #include "content/public/utility/content_utility_client.h"
 #include "third_party/WebKit/Source/WebKit/chromium/public/WebIDBKey.h"
 #include "third_party/WebKit/Source/WebKit/chromium/public/WebKit.h"
@@ skipping 101 matching lines @@
 void UtilityThreadImpl::OnBatchModeFinished() {
   ChildProcess::current()->ReleaseProcess();
 }
 
 #if defined(OS_POSIX)
 void UtilityThreadImpl::OnLoadPlugins(
     const std::vector<FilePath>& plugin_paths) {
   webkit::npapi::PluginList* plugin_list =
       webkit::npapi::PluginList::Singleton();
 
+  // On Linux, some plugins expect the browser to have loaded glib/gtk. Do that
+  // before attempting to call into the plugin.
+#if defined(TOOLKIT_USES_GTK)
+  g_thread_init(NULL);

[piman, 2011/12/02 18:13:39]

It seems that we should do that once and for all in the constructor, rather than
every time we get here.
Also, any reason not to do the full GTK initialization like in the plugin
process?

[Robert Sesek, 2011/12/02 18:37:44]

Why? The only IPC message that requires this is LoadPlugins, and the process is
destroyed at the end of this method.
Because nothing is actually being put onscreen at this stage, I don't think GTK
needs to be fully initialized. I can do it if you think it's necessary, though.

[piman, 2011/12/02 18:45:21]

Today. If you want to leave things here, then please add an assert that it's
only called exactly once.
I can imagine plugins wanting to query X extensions or whatever, and relying on
gtk to be fully initialized. Plugins are scary, very often broken in subtle
ways. Best is to provide them with a friendly environment.

[Robert Sesek, 2011/12/02 19:19:42]

Moving to ctor > adding an assert. Done.
OK. Done.

+#endif
+
   for (size_t i = 0; i < plugin_paths.size(); ++i) {
     ScopedVector<webkit::npapi::PluginGroup> plugin_groups;
     plugin_list->LoadPlugin(plugin_paths[i], &plugin_groups);
 
     if (plugin_groups.empty()) {
       Send(new UtilityHostMsg_LoadPluginFailed(i, plugin_paths[i]));
       continue;
     }
 
     const webkit::npapi::PluginGroup* group = plugin_groups[0];
     DCHECK_EQ(group->web_plugin_infos().size(), 1u);
 
     Send(new UtilityHostMsg_LoadedPlugin(i, group->web_plugin_infos().front()));
   }
 
   ReleaseProcessIfNeeded();
 }
 #endif