I certainly agree in many cases like those you mention, particularly for unprompted popovers like notifications. For desktop professional software, on the other hand, I believe the value of swift muscle memory supercedes this. In this case, input should be buffered and applied to the state of the app after the previous input has been processed.
Open a tool window, subsequent keystrokes should be sent to that too window, even if it takes a second to show. The "new/modern" interface on my CNC is both show and doesn't properly buffer input, and its hugely painful.
EDIT: I realize you specified touch, which isn't "desktop", but my CNC control is touch based and the same applies.
> In this case, input should be buffered and applied to the state of the app after the previous input has been processed.
Yes, and this works best with keyboard-oriented interfaces (which I think is generally much better than touch screens anyways; a lot of software I write is designed for keyboard use because it has this and other benefits). However, it should only be done if the process of the UI is what is expected; if something unexpected occurs then it might be better to discard any pending input. (But, sometimes this "expected" and "unexpected" is not so clear.)
Open a tool window, subsequent keystrokes should be sent to that too window, even if it takes a second to show. The "new/modern" interface on my CNC is both show and doesn't properly buffer input, and its hugely painful.
EDIT: I realize you specified touch, which isn't "desktop", but my CNC control is touch based and the same applies.