[telemetry] Implement first version of timeline based measurement

tools/perf/metrics/timeline_interaction_record.py

Index: tools/perf/metrics/timeline_interaction_record.py
diff --git a/tools/perf/metrics/timeline_interaction_record.py b/tools/perf/metrics/timeline_interaction_record.py
new file mode 100644
index 0000000000000000000000000000000000000000..87d575de8beaa9b6d0362c35c807a9e1309b865d
--- /dev/null
+++ b/tools/perf/metrics/timeline_interaction_record.py
@@ -0,0 +1,74 @@
+# Copyright 2014 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.
+
+import re
+
+
+def IsTimelineInteractionRecord(event_name):
+  return event_name.startswith('Interaction.')
+
+
+class TimelineInteractionRecord(object):
+  """Represents an interaction that took place during a timeline recording.
+
+  As a page runs, typically a number of different (simulated) user interactions
+  take place. For instance, a user might click a button in a mail app causing a
+  popup to animate in. Then they might press another button that sends data to a
+  server and simultaneously closes the popup without an animation. These are two
+  interactions.
+
+  From the point of view of the page, each interaction might have a different
+  logical name: ClickComposeButton and SendEmail, for instance. From the point
+  of view of the benchmarking harness, the names aren't so interesting as what
+  the performance expectations are for that interaction: was it loading
+  resources from the network? was there an animation?
+
+  Determining these things is hard to do, simply by observing the state given to
+  a page from javascript. There are hints, for instance if network requests are
+  sent, or if a CSS animation is pending. But this is by no means a complete
+  story.
+
+  Instead, we expect pages to mark up the timeline what they are doing, with
+  logical names, and flags indicating the semantics of that interaction. This
+  is currently done by pushing markers into the console.time/timeEnd API: this
+  for instance can be issued in JS:
+
+     var str = 'Interaction.SendEmail/is_smooth,is_loading_resources';
+     console.time(str);
+     setTimeout(function() {
+       console.timeEnd(str);
+     }, 1000);
+
+  When run with perf.measurements.timeline_based_measurement running, this will
+  then cause a TimelineInteractionRecord to be created for this range and both
+  smoothness and network metrics to be reported for the marked up 1000ms
+  time-range.
+
+  """
+  def __init__(self, event):
+    self.start = event.start
+    self.end = event.end
+
+    m = re.match('Interaction\.(.+)\/(.+)', event.name)
+    if m:
+      self.logical_name = m.group(1)
+      if m.group(1) != '':
+        flags = m.group(2).split(',')
+      else:
+        flags = []
+    else:
+      m = re.match('Interaction\.(.+)', event.name)
+      assert m
+      self.logical_name = m.group(1)
+      flags = []
+
+    for f in flags:
+      if not f in ('is_smooth', 'is_loading_resources'):
+        raise Exception(
+          'Unrecognized flag in timeline Interaction record: %s' % f)
+    self.is_smooth = 'is_smooth' in flags
+    self.is_loading_resources = 'is_loading_resources' in flags
+
+  def GetResultNameFor(self, result_name):
+    return "%s/%s" % (self.logical_name, result_name)