Hey all! Creating this bug so that we can have a central place for discussions (had some separate discussions) and I also have some questions.

So we have two shmem buffers between HTA and HTQ, say SA and SB. IIRC, there was some concerns about whether this case would be possible: HTA finished writing to SA and sent HTQ a switch IPC to switch to SA; HTQ hasn't got the switch IPC yet so still reading from SB; before HTQ actually switches to SB, HTA got a notice to Aggregate() again so we are writing to SB in HTA while HTQ is still reading from it.

I was trying to see the sequence of calls (with OOP-D): (notes at the end of each stacktrace)

[116988:117137:0206/181037.222500:ERROR:compositor_frame_sink_support.cc(271)] void viz::CompositorFrameSinkSupport::SubmitCompositorFrame(const viz::LocalSurfaceId &, viz::CompositorFrame, mojom::HitTestRegionListPtr, bool *)
=> (we have just finished storing hit-test data in HitTestManager)
[116988:117137:0206/181037.232808:ERROR:display_scheduler.cc(247)] virtual bool viz::DisplayScheduler::OnBeginFrameDerivedImpl(const viz::BeginFrameArgs &)
=> (we call "begin_frame_source_->DidFinishFrame(this)" at the end of SubmitCompositorFrame)
[116988:117137:0206/181037.234017:ERROR:display_scheduler.cc(484)] bool viz::DisplayScheduler::AttemptDrawAndSwap()
=> (this will tell HTA to start aggregating)
[116988:117137:0206/181037.235275:ERROR:hit_test_aggregator.cc(59)] void viz::HitTestAggregator::Swap()
=> (HTA finished aggregating)
[116910:116910:0206/181037.235607:ERROR:hit_test_query.cc(50)] void viz::HitTestQuery::SwitchActiveAggregatedHitTestRegionList(uint8_t)
=> (HTQ finished switching pointer to shmem)
[116988:117137:0206/181037.236550:ERROR:display_scheduler.cc(507)] void viz::DisplayScheduler::DidFinishFrame(bool)
=> (continue with these block of calls again, SubmitCompositorFrame -> ... -> DidFinishFrame)


I'm not very familiar with the timing of when clients SubmitCompositorFrame (and all the BeginFrame, OnBeginFrame, DidFinishFrame etc. notifications timing), would it be guaranteed that when Viz service receives a new SubmitCompositorFrame call, the previous IPC (switch pointer to shmem) from HTA to HTQ should have already arrived at the client side?