Add tracing AutoOpenCloseEvent.

base/trace_event/persistent_async_event.h

Index: base/trace_event/persistent_async_event.h
diff --git a/base/trace_event/persistent_async_event.h b/base/trace_event/persistent_async_event.h
new file mode 100644
index 0000000000000000000000000000000000000000..59a248b37ba212de8e4283ef1f6aaf02efc1acac
--- /dev/null
+++ b/base/trace_event/persistent_async_event.h
@@ -0,0 +1,45 @@
+// Copyright 2016 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.
+
+#ifndef BASE_TRACE_EVENT_PERSISTENT_ASYNC_H_
+#define BASE_TRACE_EVENT_PERSISTENT_ASYNC_H_
+
+#include "base/macros.h"
+#include "base/time/time.h"
+#include "base/trace_event/trace_event.h"
+
+namespace base {
+namespace trace_event {
+
+// Note: The categories and event names must have application lifetime
+// (strings or literals), as documented in trace_event_common.h.

[charliea (OOO until 10-5), 2016/08/29 17:27:27]

nit: aren't "string literals" a thing? "(i.e. literals)" might be more correct
than "(strings or literals)", unless I'm misunderstanding something

[alexandermont, 2016/08/29 18:59:07]

Done

+// Thus, we must makes subclasses that have hard-coded categories

[benjhayden, 2016/08/28 05:17:20]

"must make" not "makes"

[alexandermont, 2016/08/29 18:59:06]

Done

+// and event names, rather than passing these into the PersistentAsyncEvent.
+
+class BASE_EXPORT PersistentAsyncEvent
+    : public trace_event::TraceLog::EnabledStateObserver {
+  public:

[charliea (OOO until 10-5), 2016/08/29 17:27:27]

nit: I think the whitespace is off here. Would you mind running "git cl format"
and reuploading? (Doing that before every upload is a good habit to get into.)

[alexandermont, 2016/08/29 18:59:06]

Done

+  PersistentAsyncEvent(size_t id);
+  ~PersistentAsyncEvent() override;
+
+  void Start();
+  void Stop();
+
+  virtual void RecordStartEvent();

[benjhayden, 2016/08/28 05:17:20]

The default implementations of these 3 methods don't seem useful. Would it be
better to leave them pure virtual in order to ensure that subclasses don't
forget to implement them with the right strings?

Also, shouldn't these methods be protected so that clients can only call Start()
and Stop()?

[alexandermont, 2016/08/29 18:59:06]

Done

+  virtual void RecordStartEventWithTimestamp(int64_t timestamp);
+  virtual void RecordStopEvent();
+  void OnTraceLogEnabled() override;
+  void OnTraceLogDisabled() override;
+
+  protected:
+

[charliea (OOO until 10-5), 2016/08/29 17:27:27]

I don't think we usually add a blank line before the first field. (see:
https://cs.chromium.org/chromium/src/base/trace_event/trace_log.h?q=base/trac...)

(`git cl format` might fix this)

[alexandermont, 2016/08/29 18:59:06]

Done

+  size_t id_;
+  bool active_;
+  int64_t start_time_;

[charliea (OOO until 10-5), 2016/08/29 17:27:27]

Why are we using int64_t to represent time instead of base::TimeTicks?

[alexandermont, 2016/08/29 18:59:06]

See above. TRACE_EVENT_ASYNC_BEGIN_WITH_TIMESTAMP0 needs an int64_t.

+};
+
+}

[benjhayden, 2016/08/28 05:17:21]

"Terminate namespaces with comments as shown in the given examples." per
https://google.github.io/styleguide/cppguide.html#Namespaces

[alexandermont, 2016/08/29 18:59:07]

Done

+}
+
+#endif // BASE_TRACE_EVENT_PERSISTENT_ASYNC_H_