> It's difficult to really test such errors.

When I can't control the called function, I usually step through the 
code in a debugger and change the result variable in question to the 
appropriate value to see if the code handles failed function calls 
correctly. 

In the three Win32 API cases, it's clear that the failure path was not 
tested.

And I can't see how the ferr if statement was tested with a failure 
value either.