Add tracing AutoOpenCloseEvent.

base/trace_event/persistent_async_event.cc

Index: base/trace_event/persistent_async_event.cc
diff --git a/base/trace_event/persistent_async_event.cc b/base/trace_event/persistent_async_event.cc
new file mode 100644
index 0000000000000000000000000000000000000000..0dba10a1d183d6156bec1e17e6e3f87f4650abd6
--- /dev/null
+++ b/base/trace_event/persistent_async_event.cc
@@ -0,0 +1,51 @@
+// 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.
+
+#include "base/trace_event/persistent_async_event.h"
+
+#include "base/macros.h"
+#include "base/time/time.h"
+#include "base/trace_event/trace_event.h"
+
+namespace base {
+
+namespace trace_event {
+
+PersistentAsyncEvent::PersistentAsyncEvent(
+  const char* category, const char* event_name): active_(false),

[benjhayden, 2016/09/14 01:03:33]

active_(false) should go on the next line.

[alexandermont, 2016/09/14 20:14:56]

Done

+  id_(static_cast<void*>(this)),

[benjhayden, 2016/09/14 01:03:33]

I thought you were going to take the VFC* in the ctor as id.
I suppose using the PAE* as the id would work ok, but in that case, it doesn't
need to be a separate field. You can say static_cast<void*>(this) directly in
the TRACE_EVENT_* macros to save a word of memory and a couple lines of code.

[Primiano Tucci (use gerrit), 2016/09/14 09:38:53]

RIght. as this code is right now, this id_ field seems useless (unless you make
it overridable in the ctor).

[alexandermont, 2016/09/14 20:14:56]

Done

+  category_(category),
+  event_name_(event_name) {
+  base::trace_event::TraceLog::GetInstance()->AddEnabledStateObserver(this);
+}
+
+PersistentAsyncEvent::~PersistentAsyncEvent() {
+  base::trace_event::TraceLog::GetInstance()->RemoveEnabledStateObserver(this);

[Primiano Tucci (use gerrit), 2016/09/14 09:38:53]

You cannot use the standard EnabledStateObserver for non-long-lived objects.
More context in crbug.com/610629 .
Switch to  AsyncEnabledStateObserver, which will require a weakptr here.

[alexandermont, 2016/09/14 20:14:56]

Done

+}
+
+void PersistentAsyncEvent::Begin() {
+  TRACE_EVENT_COPY_ASYNC_BEGIN0(category_, event_name_, id_);

[Primiano Tucci (use gerrit), 2016/09/14 09:38:53]

Why _COPY_? Aren't the argument supposed to be long lived strings? (Add a
comment there)

[alexandermont, 2016/09/14 20:14:56]

Done

+  active_ = true;
+  start_time_ = base::TimeTicks::Now();
+}
+
+void PersistentAsyncEvent::End() {
+  TRACE_EVENT_COPY_ASYNC_END0(category_, event_name_, id_);
+  active_ = false;
+}
+
+void PersistentAsyncEvent::OnTraceLogEnabled() {
+  if (active_)
+    TRACE_EVENT_COPY_ASYNC_BEGIN_WITH_TIMESTAMP0(
+        category_, event_name_, id_, start_time_.ToInternalValue());
+}
+
+void PersistentAsyncEvent::OnTraceLogDisabled() {
+  // TODO(alexandermont): How do we call RecordStopEvent before log disabled?

[benjhayden, 2016/09/14 01:03:33]

Wait for this CL to land then use OnBeforeTraceLogDisabled.
https://codereview.chromium.org/2327333002

+  if (active_)
+    TRACE_EVENT_COPY_ASYNC_END0(category_, event_name_, id_);

[Primiano Tucci (use gerrit), 2016/09/14 09:38:53]

This one doesn't work, it's too late to add events here.
Aren't async event auto-closed by the importer?
https://codereview.chromium.org/2327333002

I am hoping that we don't need that CL. Another callback from tracing is going
to be another footgun.

[alexandermont, 2016/09/14 20:14:56]

Will hold off until we decide whether we are in fact using that CL.

+}
+
+}  // namespace trace_event
+}  // namespace base