Using a ton of libraries, IO, ... you cannot handle all imaginary errors. You will need some form of "catch uncaught exception" handling to still have a control flow for such a case. But if you are aware of such an error, you usually can and want to add a proper handling.

That brings up the paradoxical question of how one would go about testing the handling of an unknown error.

Having a test case for the handling of an unknown error would make it a known error and would trigger adding appropriate handling, which in turn will invalidate the test case.

How should we test the handling of an unknown error?

One thought process was to specifical implement a method which can be triggered by some internally known condition which will raise an "unknown exception". But this feels wrong on so many levels.

Any other ideas?