After a bit of trial and error, I figured out that the windows test fail due to the usage of pytest-xdist. I did not investigate, why exactly that breaks it, though. Instead I chose a programatic apporach: was_called_by does a try-except and on exceptions, it logs the error and returns False.

This may give false warnings in rare cases where ApplicationBuilder was used correctly but the _was_called_by raises an exception, but I reckon that these cases are even more rare than symlinks.

The benefit of this change is that was_called_by now can never cause the code to fail - IMO this is an upside, as PTB figuring out if it should raise a warning should not be a cause for the program not running.

Good work! It looks good to me.