The example you give is a statement, not an expression. There is no value returned, and the return-type is void, which can't be constructed. Only imperative languages, that can ignore the lack of a value from an expression (by simply moving on to the next statement), can deal with void returning functions. In Haskell this causes the end-of-the-program because it works solely with expressions, and an expression must have a value.
It's been a good 30 years since I've written any C, and I know its type-system is compromised at best, but I'm fairly sure you can't do something like this?
void x = new void();
Or, using your foo example:
void x = foo();
If you can then void is inhabited (and incorrectly named). I realise you could create void* and other elaborate means of claiming you have an inhabited void, but really can you legally construct one, not hack any underlying value to pretend to have one - there is a difference. If you can construct one and assign it to a variable, then it's inhabited.
Railing against Haskell when its implementation actually works correctly is a very strange position to take. The clue really is in the name empty-set = void, singleton-set = unit.
