I wouldn't try to change CanActivateWindow() to deal with this, rather make the tests hide the windows.

Re sky: Thanks! 
Shall we change TestWindowDelegate to handle this? In Window destructor, it will call WindowDelegate::OnWindowDestroying() first and then notify its observers. Maybe we should the window first in TestWindowDelegate::OnWindowDestroying()? 
(Actually I tried to do this in TestWindowDelegate::OnWindowDestroying(), but it broke a lot of unittests which I could not figure out why. So instead to modify TestWindowDelegate, I created a customized subclass overriding TestWindowDelegate to handle my test cases as a workaround. But I don't know if it's an issue that we should fix it in a general place)

What specific tests are failing? I'm inclined to make the test explicitly hide the window first. I suspect it isn't necessary to do this in TestWindowDelegate, rather from the test itself.

Currently there is no test failing because of this. But with my CL https://chromium-review.googlesource.com/c/chromium/src/+/734809 in place, several unittests in split_view_controller_unittest.cc and window_selector_unittest.cc are failing. 

The reason is in SplitViewController::OnWindowDestroying(window), it's possible that it will start the overview mode. MRU window list is calculated here to decide which window should be put in overview window grid. The to-be-destroyed window should not put in the MRU window list otherwise it will cause crash when the overview window grid tries to access the window. 

As you suggested, I created a customized TestWindowDelegate and override OnWindowDestroying() to hide the window first to fix these tests. If it's unnecessary to (or should not) do this in TestWindowDelegate, I'll close this issue then.

Close this.