Adventures in undefined behavior: The premature downcast

re: Adventures in undefined behavior: The premature downcast

JamesJohnston — Thu, 06 Sep 2012 19:25:45 GMT

Yeah I know.... it won't work in a function. I think that should work if it's a class member, though. Just the constructors need to initialize it in the initializer lists. :) I probably should have clarified it more.

re: Adventures in undefined behavior: The premature downcast

@JamesJohnston — Thu, 06 Sep 2012 16:05:00 GMT

Shape &shape;

will not compile. You cannot declare a reference without assigning to it, since, well, there is no such things as a null reference ;)

re: Adventures in undefined behavior: The premature downcast

alegr1 — Tue, 04 Sep 2012 21:30:04 GMT

bad_cast is thrown only is you cast into a reference type. dynamic_cast to a pointer doesn't throw.

re: Adventures in undefined behavior: The premature downcast

JamesJohnston — Tue, 04 Sep 2012 18:28:52 GMT

@POKE53280,0: Wrong; you get a null pointer if the cast is invalid, which is then noticed by the following "if" statement.

bad_cast is for references because there is no such thing as a null reference; for example:

Shape &shape;

Shape2D &shape2D = dynamic_cast<Shape2D&>(shape); // may throw bad_cast

re: Adventures in undefined behavior: The premature downcast

POKE53280,0 — Tue, 04 Sep 2012 17:24:57 GMT

A simple note to Raymond's code posted on mbrierst comment:

[Shape2D *shape = dynamic_cast<Shape2D*>(shape);

if (shape) { ... }]

I think if dynamic_cast<> fails, a std::bad_cast C++ exception is thrown, so probably the simple "if (shape)..." check is not correct.

re: Adventures in undefined behavior: The premature downcast

mbrierst — Tue, 04 Sep 2012 15:03:41 GMT

Thanks for the cheat sheet. Lesson learned.

re: Adventures in undefined behavior: The premature downcast

mbrierst — Tue, 04 Sep 2012 14:00:51 GMT

I've failed the "example even with dynamic_cast" exercise, and I don't see it posted anywhere in the comments. Does anyone have a teacher's edition of this blog with this exercise completed?

[Shape2D *shape = dynamic_cast<Shape2D*>(shape);
if (shape) { ... }
shape->Something(); /* tells the compiler that the dynamic_cast succeeded, so it can assume that shape is a Shape2D. */ -Raymond]

re: Adventures in undefined behavior: The premature downcast

Damien — Tue, 04 Sep 2012 06:37:50 GMT

@Joshua

I'd disagree. The Interrupt Vector Table is at address 0. Just because there's no way to distinguish between "pointer to address 0" and "NULL pointer" doesn't mean that the two distinct concepts should be conflated.

re: Adventures in undefined behavior: The premature downcast

Joshua — Tue, 04 Sep 2012 01:12:40 GMT

@Cesar: It got famous when it was discovered that optimization was a _bad_ idea in kernel mode.

If I were compiling for 32 bit nommu mode (enable 32 bit protected mode and set the segment selectors up for the famous unreal mode but stay in protected mode), I'd definitely disable this optimization as the interrupt vector table (which I will be accessing) is still at NULL.

re: Adventures in undefined behavior: The premature downcast

RicarDog — Mon, 03 Sep 2012 15:38:55 GMT

@JamesJohnston Regarding dynamic_cast, VS2010 still uses the same slow method of comparing strings. I would recommend against using RTTI in general if your application's performance is critical.

@Barry Kelly "Predictability is more valuable than performance." This is a case of "my scenario is better than yours". Some applications do have high performance requirements, and the C++ language was designed with them in mind. It's not an ego issue, it's just that computers don't have brains yet so they have to rely on yours.

As a side note, please NEVER cast base to derived pointers using reinterpret_cast or C-style cast such as Shape2D* shape2d = (Shape2D*)shape, even if you have guarantee that the cast is valid, because it results in undefined behavior when the type is incomplete (i.e. Shape2D is a forward-declared type) and the compiler won't complain about it. On the other hand, static_cast does generate a compile error in such cases.