When thread is no longer stopped the focus is not automatically passed to another thread

[luabud]

Hey there!

We have a problem in the Python extension when it comes to the UX of multi-process debugging.
Take the example below:

from concurrent.futures import ProcessPoolExecutor
from time import sleep

values = [2, 3, 4, 5, 6]

def cube(x): 
    res = x ** 3
    print(f"Cube of {x}:{res}")

if __name__ == "__main__":
    result = []
    with ProcessPoolExecutor(max_workers=5) as executor:
        result = executor.map(cube,values)

    for r in result:
        print(r)

If you add a breakpoint to e.g. line 7 (res = x ** 3), start debugging this file in VS Code (just make sure to select "Python File" from the drop down menu once you hit F5) and hit continue, the first subprocess that is in focus will continue, and the focus will remain in it, even though you have multiple other subprocesses paused on that breakpoint.
The request here is if there's a way the focus could automatically switch to another subprocess that is paused.

[luabud]

cc @isidorn @weinand @karthiknadig @int19h

[isidorn]

Thanks for opening the issue and I agree we should improve here.
We got the same feedback from @testforstephen here #68955

As I mentioned the problem on the VS Code side is the following:

• Once a user presses continue on a thread T1, VS Code does not know if it should pass focus to another thread. Problem being that T1 might hit a breakpoint again in 1s, and if we passed focus to T2 then it might be a bad experience for the user that he just lost focus on what he was actually debugging.

I am open for ideas on how to solve this. But we might need to add some hints to the DAP so the debug adapter could control this.

Let me know what you think. Thanks
Also fyi @gregg-miskelly since C# might be hitting this as well

[gregg-miskelly]

@isidorn The way Visual Studio handles this situation is that if the UI gets multiple stopping events, the subsequent stopping events are queued. Hitting F5 just dequeues the next stopping event, so we will switch focus to that process. I don't see any reason why this approach wouldn't work for VS Code as well. This wouldn't require any DAP changes.

[weinand]

@isidorn Gregg's suggestion sounds like an excellent idea.

[isidorn]

We can try that out. But the problem is that if Thread T1 we hit continue we will automatically switch focus to an already stopeed thread T2, even if T1 is about to stop in 100ms.

We can treat continue specially here, so that this would not happen for step over and step into.
@gregg-miskelly Visual Studio does this only for when continue is pressed? And thanks a lot for a good suggestion

[gregg-miskelly]

@isidorn No, full VS has the same behavior for all three execution commands. This can cause annoying bouncing back and forth if the process you aren't trying to step is hitting breakpoints. Fortunately if this happens you can, as a user of the debugger, fix this by disabling your breakpoints and hitting continue. Then when you step completes things will stay on the stepping process.

We have thought about implementing a "priority boost" where if the user hits step will will wait a small amount of time before dequeuing breakpoint events from other threads/processes. But the priority of this scenario has never risen high enough to implement something like this. The tricky part is to find the right delay period. You don't want to wait too long since the stepping thread may be waiting on the other process/thread (the one that is stopped). On the other hand, too short of a delay means we are back to alternating threads.

[isidorn]

@gregg-miskelly thanks for providing more details. Making the user disable all breakpoint before stepping sounds like a bad experience.

Maybe a simple 500ms delay would help. Since most of the steps would hit a consequent stop in that time.
Anyways I will try to experiment with this this milestone and once we have something you can all try it out and we can fine tune the behaviour. Thanks!

[isidorn]

I have looked into this and here are my findings:

1. The approach (queueing stop events) @gregg-miskelly suggested makes good sense and I have implemented it
2. The case that @luabud outlines is actually not a multi threaded example. Instead it is a parent session with mutiple child debug session and each child has exactly one thread. Since this is somewhat of an implementation detail I think we can use the same logic when switching focus between sessions as we do when switching focus between threads.
3. If I step over Session A, and I have come to the "end" the focus automatically switches to Session B which can be confusing to the user especially if both sessions are stopped on the same file.

Due to point 3 and to the problem of back and forth between two threads I have decided to only auto focus thread / session after 800ms delay if no stop even has come in the meantime.
I believe this timeout will improve the "back and forth" case and make it a bit clearer to the user that a focus has changed.

@luabud @gregg-miskelly @testforstephen please try it out from tomorrow and let me know what you think. In the testing that I have done this is the approach I liked the best, however I am open for suggestions on how to improve.