Add process lifetime annotations to stack samples.

components/metrics/call_stack_profile_metrics_provider.cc

 // Copyright 2015 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 "components/metrics/call_stack_profile_metrics_provider.h"
 
 #include <stddef.h>
 #include <stdint.h>
 
 #include <algorithm>
@@ skipping 15 matching lines @@
 #include "base/threading/thread_task_runner_handle.h"
 #include "base/time/time.h"
 #include "components/metrics/proto/chrome_user_metrics_extension.pb.h"
 
 using base::StackSamplingProfiler;
 
 namespace metrics {
 
 namespace {
 
+const ProcessPhase

[Alexei Svitkine (slow), 2016/11/07 23:22:07]

Please document these mappings with a comment above.

I guess this is to map specific bits to protobuf constants. Which seems fine,
but should be made more clear.

[bcwhite, 2016/11/08 01:02:36]

Done.

+    kProtoPhases[CallStackProfileMetricsProvider::PHASES_MAX_VALUE] = {
+        ProcessPhase::MAIN_LOOP_START,
+        ProcessPhase::MAIN_NAVIGATION_START,
+        ProcessPhase::FIRST_NONEMPTY_PAINT,
+        ProcessPhase::FIRST_CONTENTFUL_PAINT,
+
+        ProcessPhase::SHUTDOWN_START,
+};
+
+const ProcessActivity
+    kProtoActivities[CallStackProfileMetricsProvider::ACTIVITIES_MAX_VALUE] = {
+        ProcessActivity::COLLECTING_METRICS,
+        ProcessActivity::WRITING_PROFILES_STATE,
+};
+
 // ProfilesState --------------------------------------------------------------
 
 // A set of profiles and the CallStackProfileMetricsProvider state associated
 // with them.
 struct ProfilesState {
   ProfilesState(const CallStackProfileParams& params,
                 base::StackSamplingProfiler::CallStackProfiles profiles,
                 base::TimeTicks start_timestamp);
   ProfilesState(ProfilesState&&);
   ProfilesState& operator=(ProfilesState&&);
@@ skipping 176 matching lines @@
   memcpy(&name_bytes[0], &basename[0], basename_length_in_bytes);
   return base::HashMetricName(name_bytes);
 }
 
 // Transcode |sample| into |proto_sample|, using base addresses in |modules| to
 // compute module instruction pointer offsets.
 void CopySampleToProto(
     const StackSamplingProfiler::Sample& sample,
     const std::vector<StackSamplingProfiler::Module>& modules,
     CallStackProfile::Sample* proto_sample) {
-  for (const StackSamplingProfiler::Frame& frame : sample) {
+  for (const StackSamplingProfiler::Frame& frame : sample.frames) {
     CallStackProfile::Entry* entry = proto_sample->add_entry();
     // A frame may not have a valid module. If so, we can't compute the
     // instruction pointer offset, and we don't want to send bare pointers, so
     // leave call_stack_entry empty.
     if (frame.module_index == StackSamplingProfiler::Frame::kUnknownModuleIndex)
       continue;
     int64_t module_offset =
         reinterpret_cast<const char*>(frame.instruction_pointer) -
         reinterpret_cast<const char*>(modules[frame.module_index].base_address);
     DCHECK_GE(module_offset, 0);
     entry->set_address(static_cast<uint64_t>(module_offset));
     entry->set_module_id_index(frame.module_index);
   }
 }
 
+// Transcode Sample annotations into protobuf fields.

[Alexei Svitkine (slow), 2016/11/07 23:22:07]

Maybe expand the comment for the motivation - that internally in Chrome we use
bit fields to set these since there shouldn't be more than 32 at any time, but
the protobuf uses repeated fields for future proof-ness and because the set of
32 may evolve over time.

[bcwhite, 2016/11/08 01:02:36]

Done.

+void CopyAnnotationsToProto(uint32_t new_phases,
+                            uint32_t activities_begun,
+                            uint32_t activities_ended,
+                            CallStackProfile::Sample* sample_proto) {
+  for (size_t bit = 0; new_phases != 0 && bit < sizeof(new_phases) * 8;
+       ++bit, new_phases >>= 1) {

[Alexei Svitkine (slow), 2016/11/07 23:22:07]

Instead of new_phases >>= 1 logic, I'd actually just unset the bit in the if - I
think that would be a bit clearer.

(Interestingly, x86 has instructions to find the next set bit which are exposed
as e.g. __builtin_clz on GCC. However, I didn't find a base:: abstraction for
these, so I think current approach is fine.)

[bcwhite, 2016/11/08 01:02:36]

That will be two extra instructions (complement and mask) per loop iteration. 
You sure?

  for (size_t bit = 0; new_phases != 0 && bit < sizeof(new_phases) * 8;
       ++bit) {
    const uint32_t flag = 1U << bit;
    if (new_phases & flag) {
      if (bit >= arraysize(kProtoPhases)) {
	NOTREACHED();
	continue;
      }
      sample_proto->add_process_phase(kProtoPhases[bit]);
      new_phases &= ~flag;
    }
  }

[Alexei Svitkine (slow), 2016/11/14 18:26:39]

It will only do those instructions when the bit is set, which should be true for
not too many bits. So I do prefer that version - since it's much clearer to me.

[bcwhite, 2016/11/16 18:27:44]

I think it's perfectly clear this way.  But done.

+    if (new_phases & 1) {
+      if (bit >= arraysize(kProtoPhases)) {
+        NOTREACHED();
+        continue;
+      }
+      sample_proto->add_process_phase(kProtoPhases[bit]);
+    }
+  }
+
+  for (size_t bit = 0;
+       activities_begun != 0 && bit < sizeof(activities_begun) * 8;
+       ++bit, activities_begun >>= 1) {
+    if (activities_begun & 1) {
+      if (bit >= arraysize(kProtoActivities)) {
+        NOTREACHED();
+        continue;
+      }
+      sample_proto->add_activity_begun(kProtoActivities[bit]);
+    }
+  }
+
+  for (size_t bit = 0;
+       activities_ended != 0 && bit < sizeof(activities_ended) * 8;
+       ++bit, activities_ended >>= 1) {
+    if (activities_ended & 1) {
+      if (bit >= arraysize(kProtoActivities)) {
+        NOTREACHED();
+        continue;
+      }
+      sample_proto->add_activity_ended(kProtoActivities[bit]);
+    }
+  }
+}
+
 // Transcode |profile| into |proto_profile|.
 void CopyProfileToProto(
     const StackSamplingProfiler::CallStackProfile& profile,
     CallStackProfileParams::SampleOrderingSpec ordering_spec,
     CallStackProfile* proto_profile) {
   if (profile.samples.empty())
     return;
 
   if (ordering_spec == CallStackProfileParams::PRESERVE_ORDER) {
     // Collapse only consecutive repeated samples together.
     CallStackProfile::Sample* current_sample_proto = nullptr;
+    uint32_t phases = 0;
+    uint32_t activities = 0;
     for (auto it = profile.samples.begin(); it != profile.samples.end(); ++it) {
       if (!current_sample_proto || *it != *(it - 1)) {
         current_sample_proto = proto_profile->add_sample();
         CopySampleToProto(*it, profile.modules, current_sample_proto);
         current_sample_proto->set_count(1);
+        CopyAnnotationsToProto(
+            it->process_phases & ~phases,          // Phases are "set only".
+            it->current_activities & ~activities,  // Activities can begin
+            activities & ~it->current_activities,  // and end.
+            current_sample_proto);
+        phases = it->process_phases;
+        activities = it->current_activities;
       } else {
         current_sample_proto->set_count(current_sample_proto->count() + 1);
       }
     }
   } else {
     // Collapse all repeated samples together.
     std::map<StackSamplingProfiler::Sample, int> sample_index;
     for (auto it = profile.samples.begin(); it != profile.samples.end(); ++it) {
       auto location = sample_index.find(*it);
       if (location == sample_index.end()) {
         CallStackProfile::Sample* sample_proto = proto_profile->add_sample();
         CopySampleToProto(*it, profile.modules, sample_proto);
         sample_proto->set_count(1);
+        CopyAnnotationsToProto(it->process_phases, it->current_activities, 0,
+                               sample_proto);
         sample_index.insert(
             std::make_pair(
                 *it, static_cast<int>(proto_profile->sample().size()) - 1));
       } else {
         CallStackProfile::Sample* sample_proto =
             proto_profile->mutable_sample()->Mutable(location->second);
         sample_proto->set_count(sample_proto->count() + 1);
       }
     }
   }
@@ skipping 162 matching lines @@
 
 // static
 bool CallStackProfileMetricsProvider::IsReportingEnabledByFieldTrial() {
   const std::string group_name = base::FieldTrialList::FindFullName(
       CallStackProfileMetricsProvider::kFieldTrialName);
   return group_name ==
       CallStackProfileMetricsProvider::kReportProfilesGroupName;
 }
 
 }  // namespace metrics