TrackEvent: Async thread tracks don't get grouped like async process tracks in the UI

[betasheet]

From discussion around #60 (comment)

It seems that two or more (async) child tracks of a thread track don't get grouped into a single vertical track in the Perfetto UI, even when they share the same name. For child tracks of process tracks, vertical merging does happen.

I imagine this should be reproducible via the SDK as well, e.g. with something like this (untested):

  TRACE_EVENT_BEGIN("cat", "AsyncTrack", perfetto::Track::ThreadScoped(obj1));
  TRACE_EVENT_BEGIN("cat", "AsyncTrack", perfetto::Track::ThreadScoped(obj2));
  TRACE_EVENT_END("cat", perfetto::Track::ThreadScoped(obj1));
  TRACE_EVENT_END("cat", perfetto::Track::ThreadScoped(obj2));

I believe, currently, such child tracks are imported as thread tracks in trace processor (code. I wonder if the collapsing/merging logic is only enabed for process tracks, not thread tracks, in the UI? Is there a reason for this / can we also extend it to thread tracks (e.g. those which share a name, possibly where track name != thread name)? :)

[betasheet]

@chromy @ddrone

[chromy]

I have so much daja-vu about this ...I could have sworn we've fixed async slice grouping like three times already :P At this point if someone comes with:

• an example trace
• a screen shot of what it looks like
• some annotations showing how they want it to look instead
  (which I appreciate is over half the battle) I'm happy to look ...other wise no promises about when I can get to this.

[TApplencourt]

Thanks a lot for the reply!
Please find more information bellow:

Trace

packet {ng trace:        0 KB
  previous_packet_dropped: true
  trace_packet_defaults {
    timestamp_clock_id: 6
  }
}
packet {
  track_descriptor {
    uuid: 1
    thread {
      pid: 1
      tid: 1
    }
  }
}
packet {
  track_descriptor {
    uuid: 2
    name: "GPU | Thread 1"
    parent_uuid: 1
  }
}
packet {
  timestamp: 0
  trusted_packet_sequence_id: 10
  track_event {
    type: TYPE_SLICE_BEGIN
    track_uuid: 2
    name: "foo 0:40"
  }
}
packet {
  timestamp: 40
  trusted_packet_sequence_id: 10
  track_event {
    type: TYPE_SLICE_END
    track_uuid: 2
  }
}
packet {
  track_descriptor {
    uuid: 3
    name: "GPU | Thread 1"
    parent_uuid: 1
  }
}
packet {
  timestamp: 15
  trusted_packet_sequence_id: 10
  track_event {
    type: TYPE_SLICE_BEGIN
    track_uuid: 3
    name: "bar 15:60"
  }
}
packet {
  timestamp: 60
  trusted_packet_sequence_id: 10
  track_event {
    type: TYPE_SLICE_END
    track_uuid: 3
  }
}

ScreenShot

some annotations showing how they want it to look instead

The two tracks should be merged vertically.

If you remove the thread section in the parent track descriptor

<     thread {
<       pid: 1
<       tid: 1
<     }

merging occurs:

We would like a similar behavior even when packet use threads