But that misses the point: it is NOT reasonable behavior for a null return value for a pre-constructor to cause the constructor itself to return null (unless that's documented behavior?!?). Consider this:

I completely agree. I don't think anybody would argue the current behaviour is right; it's quite wrong and as you demonstrated it can have follow up effects that lead to hard to trace issues.

the question I was attempting to address is "what should happen instead?". In the past some people have suggested that this situation (non-returning functions) should be made impossible (at the parser stage) and I don't think we can do that without demanding a very strict style of programmers.

I admit it was unclear that I was partially addressing that suggestion instead of the bug you reported and that this was unclear. As far as I can see there is nothing here on which we disagree.

I would, BTW, like to suggest that "fans of kassen" should have plenty of member functions that return stuff, cups of coffee and dark beers would be especially good :Ź).

Cheers,
Kas.