Re: [gtkmm] OT: C++ exceptions

[Jarek Dukat <madmaxer poczta fm>]

On Sun, 2002-07-07 at 22:14, Robert J. Hansen wrote:
> JD> --- G++ 3.1 (dla 2.95 tak samo):
> JD> Foo CTOR: 1
> JD> Foo CTOR: 2
> JD> UNEXPECTED!
> JD> Aborted
> 
> JD> And according to my knowledge about C++ exceptions a correct result
> JD> should be:
> JD> Foo CTOR: 1
> JD> Foo CTOR: 2
> JD> UNEXPECTED!
> JD> Foo DTOR: 1
> JD> Ex2 caught
> 
> JD> Obviously VC ignores unexpected() handler, and G++ treats it much too
> JD> seriously :)
> JD> What do you think about it?
> 
> GCC 3.1 is correct.  According to spec, a destructor must only be
> invoked for a fully-formed object.  By throwing an exception in your
> constructor, you keep the first object from being fully formed; thus,
> no dtor gets invoked when it goes out of scope.

Yes, you're right about foo2. But foo1 hasn't been destroyed and this is
a bug, I think... It has been succesfully constructed, so it should be
cleaned up.

> Insofar as why foo(1)'s dtor isn't called... when you call foo(2), an
> exception gets thrown.  Due to the fact foo(2) isn't declared to throw
> Ex2, foo(2)'s exception doesn't get caught in the try block--it
> automagically gets handled by the unexpected handler, which in turn
> re-throws a generic exception.  The thread of execution then returns
> back to main(), which has catch() blocks for Ex1 and Ex2, but nothing
> for a generic exception.
> 
> Does this help at all?

throw; should re-throw an exception of exactly the same type that
occoured before, so I think it should be handled by catch() in main().
I'm not sure but I've read something about special behaviour of
unexpected() handler in such case. But I may be wrong. Unfortunatelly I
don't have Stroustrup handy :)

Hmmm...

-- 
struct Sig {
  string name("     J a r e k   D u k a t     ");
  string mail(" madmaxer (at) poczta (dot) fm ");
};


-----------------------------------------------------------------------
2 lata za strzelanie do podobizny Balcerowicza?
>>> http://link.interia.pl/f1607