Change Chromoting logger to be setup in plugin's NP_Initialize.

remoting/host/plugin/host_log_handler.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 "remoting/host/plugin/host_log_handler.h"
+
+#include "remoting/base/util.h"
+#include "remoting/host/plugin/host_script_object.h"
+
+namespace remoting {
+
+// Records whether or not we have a scriptable object registered for logging.
+// This is set inside the lock, but is read (in LogToUI) outside of a lock so
+// that we don't needlessly slow down the system when we log.
+static bool g_has_logging_scriptable_object = false;
+
+// The lock that protects the logging globals.
+static base::Lock g_logging_lock;
+
+// The scriptable object that will display the log information to the user.
+static HostNPScriptObject* g_logging_scriptable_object = NULL;
+
+// The previously registered LogMessageHandler. If not NULL, we call this after
+// we're doing processing the log message.
+static logging::LogMessageHandlerFunction g_logging_old_handler = NULL;
+
+// Set to true when we register our global log handler so that we don't try
+// to register it twice.
+static bool g_has_registered_log_handler = false;
+
+// static
+void HostLogHandler::RegisterLogMessageHandler() {
+  base::AutoLock lock(g_logging_lock);
+
+  if (!g_has_registered_log_handler) {
+    LOG(INFO) << "Already have global log handler";

[Wez, 2011/09/01 01:21:23]

It doesn't seem necessary to log something here, since this is a normal state of
affairs, and the "Registering" message will already have been logged.

[garykac, 2011/09/01 01:30:55]

Done.

+    return;
+  }
+
+  LOG(INFO) << "Registering global log handler";
+
+  // Record previous handler so we can call it in a chain.
+  g_logging_old_handler = logging::GetLogMessageHandler();
+
+  // Set up log message handler.
+  // This is not thread-safe so we need it within our lock.
+  // Note that this will not log anything until a scriptable object instance
+  // has been created to handle the log message display.
+  logging::SetLogMessageHandler(&LogToUI);
+}
+
+// static
+void HostLogHandler::RegisterLoggingScriptObject(
+    HostNPScriptObject* script_object) {
+  base::AutoLock lock(g_logging_lock);
+
+  LOG(INFO) << "Registering log handler scriptable object";
+
+  // Register this script object as the one that will handle all logging calls
+  // and display them to the user.
+  // If multiple plugins are run, then the last one registered will handle all
+  // logging for all instances.
+  g_logging_scriptable_object = script_object;
+  g_has_logging_scriptable_object = true;
+}
+
+// static
+void HostLogHandler::UnregisterLoggingScriptObject(
+    HostNPScriptObject* script_object) {
+  base::AutoLock lock(g_logging_lock);
+
+  // Ignore unless we're the currently registered script object.
+  if (script_object != g_logging_scriptable_object)
+    return;
+
+  // Unregister this script object for logging.
+  g_has_logging_scriptable_object = false;
+  g_logging_scriptable_object = NULL;
+
+  LOG(INFO) << "Unregistering log handler scriptable object";
+}
+
+// static
+bool HostLogHandler::LogToUI(int severity, const char* file, int line,
+                             size_t message_start,
+                             const std::string& str) {
+  // Note: We're reading |g_has_logging_scriptable_object| outside of a lock.
+  // This lockless read is done so that we don't needlessly slow down global
+  // logging with a lock for each log message.
+  //
+  // This lockless read is safe because:
+  //
+  // Misreading a false value (when it should be true) means that we'll simply
+  // skip processing a few log messages.
+  //
+  // Misreading a true value (when it should be false) means that we'll take
+  // the lock and check |g_logging_scriptable_object| unnecessarily. This is not
+  // problematic because we always set |g_logging_scriptable_object| inside a
+  // lock.
+  //
+  // Misreading an old cached value is also not problematic for the same
+  // reasons: a mis-read either skips a log message or causes us to take a lock
+  // unnecessarily.
+  if (g_has_logging_scriptable_object) {
+    base::AutoLock lock(g_logging_lock);
+
+    if (g_logging_scriptable_object) {
+      std::string message = remoting::GetTimestampString();
+      message += (str.c_str() + message_start);
+      g_logging_scriptable_object->PostLogDebugInfo(message);
+    }
+  }
+
+  // Call the next log handler in the chain.
+  if (g_logging_old_handler)
+    return (g_logging_old_handler)(severity, file, line, message_start, str);
+
+  return false;
+}
+
+}  // namespace remoting