开发者

Overloading c++ typecasting (functions)

开发者 https://www.devze.com 2023-03-05 20:37 出处:网络
Using C++ style typecasting开发者_如何学Gos (all 4) look exactly like some function template. e.g.

Using C++ style typecasting开发者_如何学Gos (all 4) look exactly like some function template. e.g.

template<typename TO, typename FROM>
TO dynamic_cast (FROM p);

will be used as,

dynamic_cast<Derived*>(p); // p is Base*

Why is it not allowed to overload them by language standard for custom usage ? (like we can overload the keywords like new/delete or other operators)


Why is it not allowed to overload them by language standard for custom usage?

I suppose that's because the standard committee, when introducing these, thought the semantics of all four of these casts are well defined, and applicable to all types they should be. And mostly, this is true.

The only counter example I know of is the inability to dynamic_cast between smart pointer instances:

shared_ptr<Derived> pd = dynamic_cast<shared_ptr<Derived> >(pb);

I suppose the ability to do that would have some merits.

I don't know whether this has been discussed by the volunteers that did all the work in the standards committee (and I'm too lazy to google), but if it has been discussed (and I'd think so), it's been rejected either because someone thought the disadvantages outweigh the advantages, or because nobody had found the time to make a decent proposal and shepherd it through.1


1 Don't laugh. There's actually a lot of things most agree would be nice to have, and which are only failing to materialize because nobody could be bothered to do the work of writing a decent proposal, and spending the time needed to discuss and to iteratively improve it until it can be voted on.


I think the reason is the same for you can't overload language keyword.

In fact, you have to see them as language keyword and not template function, even if the look the same. OTOH, I couldn' imagine what kind of disasters one could do by changing the meaning of this particular C++ aspect.

EDIT
I was pretty sure that someone would have come up with the question: "then why you can overload new/delete?". I think that memory allocation/deallocation customization is something that you need in certain scenarios, and the benefits of allowing you to overload them outweighs the risks. I can't see any advantage in subverting the C++ type system, IOW I fail to think a scenario where it would be useful. Do you?


Pointer conversion with dynamic_cast, reinterpret_cast and static_cast have well defined meanings and it is probably better not to allow overloading.
It would be a pain to allow users to change the meaning of const_cast.

Only object type casting remains.

struct A
{
  A() {};
  template <typename FROM>
  A(FROM&)  {
    std::cout << "Casting to A \\o/" << std::endl;
  }

  template <typename TO>
  operator TO()  {
    std::cout << "Casting from A \\o/" << std::endl;
    return TO();
  }
};

then

  int i; A a;
  A toA = static_cast<A>(i);  //  Casting to A \o/
  int fromA = static_cast<int>(a); // Casting from A \o/

Hopefully you have better use cases than mine :)


You cannot overload these operators. It is, arguably, because you cannot change the meaning of such a fundamental thing as other answers say. (Like changing the meaning for + for integers, or the * decoration to generate a pointer from a type).

Having said that, nothing prevents you from defining your own generalizations of cast functions, that take an argument and return things very related to it. Actually, I would argue the opposite point of view, you should never use the language cast operations (static/dynamic/reinterpret_cast) unless you are doing pretty low level stuff.

What you would want to do probably is to define your own casting function that most of the time behaves like the ones provided by the language but once in a while they do something more specific for your specific purposes. You have to think hard what this function really does and name it well. What kind of runtime cost you can afford, what kind of behavior on "failure" (throw execption?, return a null value?), etc.

The standard and many libraries are full of such functions. Sometimes they add a tweaked behavior over language cast, other times they do more. Some examples:

https://en.cppreference.com/w/cpp/memory/shared_ptr/pointer_cast

  • std::static_pointer_cast,
  • std::dynamic_pointer_cast,
  • std::const_pointer_cast,
  • std::reinterpret_pointer_cast

  • std::any_cast

  • std::chrono::time_point_cast

(see also https://www.boost.org/doc/libs/1_39_0/libs/smart_ptr/pointer_cast.html)

https://www.boost.org/doc/libs/1_42_0/libs/conversion/lexical_cast.htm

  • boost::lexical_cast

https://www.boost.org/doc/libs/1_63_0/libs/conversion/cast.htm

  • polymorphic_cast,
  • polymorphic_downcast,
  • polymorphic_pointer_cast
  • polymorphic_pointer_downcast

https://www.boost.org/doc/libs/1_72_0/libs/numeric/conversion/doc/html/boost_numericconversion/improved_numeric_cast__.html

  • boost::numeric_cast

https://www.boost.org/doc/libs/1_47_0/doc/html/boost_units/Quantities.html#boost_units.Quantities.Quantity_Construction_and_Conversion

  • boost::units::quanity_cast.

Sometimes they are not called cast at all :)

https://en.cppreference.com/w/cpp/utility/variant/get_if

  • std::get_if

Another example, for template code I wrote this cast function that is only invoked if the conversion can be done implicitly:

    template<class To, class From, std::enable_if_t<std::is_convertible<From, To>{}, int> =0>
    To implicit_cast(From&& f){
        return static_cast<To>(f);
    }

https://godbolt.org/z/ym8MnJ

0

精彩评论

暂无评论...
验证码 换一张
取 消