[memory_infra] Fix trace rerun options in memory benchmarks

tools/perf/benchmarks/memory_infra.py

Index: tools/perf/benchmarks/memory_infra.py
diff --git a/tools/perf/benchmarks/memory_infra.py b/tools/perf/benchmarks/memory_infra.py
index e7934e6f6c76dbc449b47d66ee51e22c3226541f..81f3b3f7097c905a3f69fb1570d8b9b763a18cfa 100644
--- a/tools/perf/benchmarks/memory_infra.py
+++ b/tools/perf/benchmarks/memory_infra.py
@@ -40,11 +40,10 @@ class _MemoryInfra(perf_benchmark.PerfBenchmark):
     return tbm_options
 
   @classmethod
-  def HasBenchmarkTraceRerunDebugOption(cls):
+  def HasTraceRerunDebugOption(cls):

[perezju, 2016/03/10 13:45:47]

The bug was caused by this method having the wrong name.

     return True
 
   def SetupBenchmarkDefaultTraceRerunOptions(self, tbm_options):
-    # TODO(perezju): Call this function for TBM benchmarks: crbug.com/593678
     tbm_options.SetLegacyTimelineBasedMetrics((
         memory_timeline.MemoryTimelineMetric(),
     ))
@@ -138,11 +137,14 @@ class MemoryLongRunningIdleGmailTBM(_MemoryInfra):
     for category in v8_categories:
       category_filter.AddIncludedCategory(category)
     options = timeline_based_measurement.Options(category_filter)
-    options.SetLegacyTimelineBasedMetrics([
-        v8_gc_latency.V8GCLatency(),
-        memory_timeline.MemoryTimelineMetric()])
     return options
 
+  def SetupBenchmarkDefaultTraceRerunOptions(self, tbm_options):
+    tbm_options.SetLegacyTimelineBasedMetrics((
+        v8_gc_latency.V8GCLatency(),
+        memory_timeline.MemoryTimelineMetric(),
+    ))

[nednguyen, 2016/03/10 16:33:32]

We also need to set the categories properly, I think?

[perezju, 2016/03/14 11:55:03]

So, you mean all of the code to set category filters from
CreateTimelineBasedMeasurementOptions should be moved into
SetupBenchmarkDefaultTraceRerunOptions?

I understand the idea is that when passing --rerun-with-debug-trace (when
SetupBenchmarkDefaultTraceRerunOptions is *not* called) we don't exclude as much
categories/metrics, etc. But I'm not quite sure as to what should be passed in
either case to ensure memory infra is still always enabled.

cc'ing Primiano to help us figure this one out.

[nednguyen, 2016/03/14 15:10:49]

Ah, I just look at the code. It looks like you don't need to anything else,
since this method is for adding more options to the non-debug one: 
https://github.com/catapult-project/catapult/blob/master/telemetry/telemetry/...

So I think you can leave this function as-is, or override it only if you want
more metrics/categories enabled.

+
   @classmethod
   def Name(cls):
     return 'memory.long_running_idle_gmail_tbm'