search

catapult-project / catapult

Deprecated Catapult GitHub. Please instead use http://crbug.com "Speed>Benchmarks" component for bugs and https://chromium.googlesource.com/catapult for downloading and editing source code..
https://chromium.googlesource.com/catapult
BSD 3-Clause "New" or "Revised" License
1.93k stars 562 forks source link

There shouldn't be a PostTask flow event from ThreadProxy::ScheduledActionDrawAndSwap to MessageLoop::RunTask that handles next input event #1045

Closed catapult-bot closed 8 years ago

catapult-bot commented 9 years ago

Issue by yuhao Tuesday Jun 23, 2015 at 19:11 GMT Originally opened as https://github.com/google/trace-viewer/issues/1045

In the following trace: https://github.com/natduca/measurmt-traces/blob/master/mobile/googlemaps_ipc_newbinding.html

There is an outgoing PostTask flow event from ThreadProxy::ScheduledActionDrawAndSwap starting at 4,568.656 ms to a MessageLoop::RunTask event starting at 4,578.056 ms. As per discussion with Brian Anderson, this doesn't seem likely to happen on the C++ side. Is this a bug of the visualization? There are a couple of other occurrences in this trace.

My guess is that the arrow is pointing to a MessageLoop::RunTask that's not traced on the C++ side.

@natduca @dj2

catapult-bot commented 9 years ago

Comment by dj2 Tuesday Jun 23, 2015 at 19:13 GMT

All MessageLoop::RunTask's are traced on the c++ side. Can you look at the actual file and see how the id's match up? Is it just that it was bound to the wrong place?

catapult-bot commented 9 years ago

Comment by yuhao Tuesday Jun 23, 2015 at 19:18 GMT

Seems like we always bind PostTask finish phase to the next slice. Is this how we implemented on the C++ side? In this case, the ts of the PostTask flow event is the same at the ts of MessageLoop::RunTask, and so we bind the flow event to it.

catapult-bot commented 9 years ago

Comment by natduca Monday Jun 29, 2015 at 15:51 GMT

Is this a different issue than the rounding issue? I'm getting a little lost in the different bugs :)

catapult-bot commented 9 years ago

Comment by yuhao Monday Jun 29, 2015 at 17:25 GMT

It doesn't look like a rounding issue. By looking at the raw trace file, it doesn't make sense to even have that flow event there. That's why I created a new issue for this.

catapult-bot commented 9 years ago

Comment by natduca Monday Jun 29, 2015 at 17:45 GMT

Maybe we should switch messageloop binding point to parent?

catapult-bot commented 9 years ago

Comment by yuhao Saturday Jul 18, 2015 at 06:29 GMT

We will revisit this after #1079 is resolved. This might be solved by the new flow event API, or there is some behavior of Chrome we don't yet understand.

benshayden commented 8 years ago

This looks like it was resolved by the new flow event API. Please re-open if not.