Record Chrome trace events in tcmalloc heap profiles

content/browser/browser_main_loop.cc

Index: content/browser/browser_main_loop.cc
diff --git a/content/browser/browser_main_loop.cc b/content/browser/browser_main_loop.cc
index 742d675d03cdf399dce5234679845faf2a7a1243..fbed41a005d7f16b6cdb16bf9665f8ffe04d3305 100644
--- a/content/browser/browser_main_loop.cc
+++ b/content/browser/browser_main_loop.cc
@@ -481,6 +481,12 @@ void BrowserMainLoop::MainMessageLoopStart() {
     memory_observer_.reset(new MemoryObserver());
     base::MessageLoop::current()->AddTaskObserver(memory_observer_.get());
   }
+
+  // TODO(jamescook): Support all tcmalloc platforms, including Windows.

[nduca, 2013/06/20 20:09:29]

How about always creating it, but have the observe do nothing if its not
supported? Then the goup about supported only goes inisde the observer...

[James Cook, 2013/06/29 00:02:42]

Done. Also consolidated all the macro craziness into a single
TRACE_MEMORY_SUPPORTED macro.

+#if defined(USE_TCMALLOC) && (defined(OS_LINUX) || defined(OS_ANDROID))
+  trace_memory_observer_.reset(new base::debug::TraceMemoryTraceLogObserver(
+      base::MessageLoop::current()->message_loop_proxy()));
+#endif

[dsinclair, 2013/06/18 15:30:15]

Does this also need to be done in places like the GPU process?

[James Cook, 2013/06/29 00:02:42]

Probably. I think I can shim this in around GpuMain, but could use some advice
on whether that's the right place or not.

 }
 
 void BrowserMainLoop::CreateThreads() {
@@ -637,6 +643,11 @@ void BrowserMainLoop::ShutdownThreadsAndCleanUp() {
   if (parts_)
     parts_->PostMainMessageLoopRun();
 
+  // TODO(jamescook): Support all tcmalloc platforms, including Windows.
+#if defined(USE_TCMALLOC) && (defined(OS_LINUX) || defined(OS_ANDROID))
+  trace_memory_observer_.reset();
+#endif
+
 #if !defined(OS_IOS)
   // Destroying the GpuProcessHostUIShims on the UI thread posts a task to
   // delete related objects on the GPU thread. This must be done before