Friday, August 22, 2008

The language that never ends ...

It has been said (alas, a reference escapes me at this time) that nobody can know all of C++.

Well, today I've just had another proof of it (not that I ever thought I knew all the language). Question: what's wrong with the following code?

  bool fn()
    void* p = 0;

    return p;

Answer: absolutely nothing. Yes, that's right, you can convert an "rvalue of arithmetic, enumeration, pointer, or pointer-to-member type" to an rvalue of type bool(C++ Standard: 4.12.)

Who'd a thought it?!

I found upon this because I'm changing the way some of the type conversions work in VOLE as part of the new functionality of version 0.5. As well as adding specialisations of vole::com_return_traits<> for float and double, I'm adding comprehensive (i.e. a cartesian product of from-type, to-type and coercion level) unit-tests for the builtins' specialisations. I thought that com_return_traits<bool> was too permissive, in allowing any of the basic integral types to be converted when the coercion level is naturalPromotion. Turns out I was wrong!

No comments: