[boost] Re: Smart Pointers - (many) thanks and (some) questions

From: Darin Adler (darin_at_[hidden])
Date: 1999-12-27 02:46:06

> Well, this begs the question: why not add an operator bool() method to
> shared_ptr and scoped_ptr?

Unfortunately, operator bool() doesn't work too well for this kind of thing,
due to the integer promotion rules. Here's an example of the problem:

    void f(boost::scoped_ptr<int> p)
    {
        int x = p; // Ooops! Converted a pointer to a 0 or a 1!
    }

To work around this, some classes in the C++ standard instead provide
conversion to void*, which work inside if statements, but don't have the bad
effect above. The void* conversion is still a bit dangerous and can be used
by accident when assigning to a void* for example. There are
more-sophisticated versions of the void* conversion trick (like "const
volatile incomplete*"), but I don't know of one that's perfect.

I consider this an unfortunate mistake in the design of bool. The implicit
conversion from bool to int isn't so valuable. But I think it's too late to
complain about this.

> If the goal is to "mimic a built-in pointer", then an operator bool()
> conversion would make sense.  It seems that using get() should be
> avoided whenever possible, as it could be dangerous to let people play
> with the _real_ pointer as long as reference-counted version exists.

Something like the void* trick, or a null() or is_null() member function
would be an improvement. I don't like:

    if (p.get())

myself. Of course, I have a bigger hang-up about null pointers, as readers
of this list may already know. I just can't stand the fact that:

    C* x;

means that x points to an object of class C and you can use all the members
declared in class C, unless it's 0, in which case you can't use *any* of the
members declared in class C! It's just a gaping hole in the type system.

    -- Darin