Avoid hanging by using non-blocking pipe

[paulromano]

Some of have observed that WATTS will sometimes hang when writing output (#49). This PR should fix this by using a non-blocking pipe. However, this solution (and in fact our prior code) only works on Unix-based platforms. On Windows, we now revert to using normal subprocess.run instead of our special version. This means that, unfortunately, if you're on Windows, using the show_stdout and show_stderr arguments won't work (they rely on tee_stdout and tee_stderr which in turn implicitly rely on our implementation of run).

I don't have WATTS installed anywhere on a Windows environment so I'm hoping someone can check whether this actually works there (@samgdotson?).

[samgdotson]

Thanks for taking the time to look into this!
watts appears successful on Windows with these updates. At the moment there are issues within ABCE itself that are preventing a complete run, but the current issue is resolved. If the non-blocking pipe solution still works on UNIX (can this be confirmed, perhaps @nstauff?), then I would approve (however, I do not have write access so further approval is still necessary).

Also of concern is that the nonblocking-pipe solution created a significant slowdown when run on WSL2. Is there a time difference between the original code and this PR?

[paulromano]

@samgdotson Thanks for confirming that it works on Windows. Does the slowdown you're seeing on WSL2 happen regardless of whether show_stdout is set to true or not?

[paulromano]

@samgdotson To follow up on your question at your meeting today: on Linux, I do not find any observable slowdown from using the output capturing here compared to just running an application directly.

[nstauff]

@paulromano @samgdotson - I tested the PyARC plugin on my MAC and it works fine. Thanks!

[samgdotson]

@nstauff could you merge this PR? I do not have write access.