开发者

Template Return Type with Default Value

开发者 https://www.devze.com 2023-02-17 05:00 出处:网络
So in writing a C++ template class, I have defined a method that returns an object of the templated type, as such:

So in writing a C++ template class, I have defined a method that returns an object of the templated type, as such:

template <typename T>
class Foo
{
public:
    T GetFoo()
    {
        T value;

        //Do some stuff that might or might not set the value of 'value'

        return value;
    }
 };

int main() {

    Foo<int> foo;

    foo.GetFoo();

    return 0;
}

This gives the following warning:

prog.cpp: In member function ‘T Foo<T>::GetFoo() [with T = int]’:
prog.cpp:15: warning: ‘value’ is used uninitialized in this function

I understand why this is happening - I am returning an uninitialized int as part of GetFoo. The thing is, if I were to use Foo<SomeCla开发者_高级运维ss>, the line T value; would initialize value using the default constructor of SomeClass.

I have managed to suppress this warning by doing the following:

    T GetFoo()
    {
        T value = T();

        //Do some stuff that might or might not set the value of 'value'

        return value;
    }

This seems to work for primitive types (such as int and float) and classes, at least so long as that class has a default constructor and copy constructor. My question is - is this the accepted way of solving this problem? Are there any side effects of this I should know about?


Sounds OK, if the class has no copy constructor, you will not be able to return it anyway.


Indeed that's the standard way to solve the problem. I don't believe there should be any side effects to worry about in this case.


The line

T t;

Default constructs objects, yet declares uninitialized built-in types. There is no syntax to 'default construct' a local variable.

Therefore it's not trivial to write generic code that initializes a variable, whether built-in or class.

Wrapping the type into a member variable, however, may present a workaround that does not need copy construction (apart from the return statement):

template< typename T > struct Initializer {
   T t;
   Initializer()
   :t() // ====> default construction, works for classes _and_ built-in
   {}
};

Using this wrapper, you can build your code in a generic way:

template<typename T> T foo() {
  Initializer<T> i;
  // fill in i.t to your liking
  return i.t;
}

See a full-blown snippet at codepad.


Do you really want to return an un-initialized value from this function? Sounds like it could lead to a whole lot of crap later down the line.

Why don't you use an appropriate wrapper - if you have access to boost, consider boost::optional. This way, if you don't initialize it, it can be tested properly.


This seems to work for primitive types (such as int and float) and classes, at least so long as that class has a copy constructor

For classes, not only [public] copy-constructor, you also need public default constructor!

My question is - is this the accepted way of solving this problem?

Usually it's a good idea to provide public default constructor. STL containers use it all the time! If you don't have one, most (or maybe, all) STL containers would not work at many occasions.

See this example with private default constructor: http://ideone.com/EdPLu


Boost addresses this problem in one of its utility template classes, namely boost::value_initialized and its relatives.

0

精彩评论

暂无评论...
验证码 换一张
取 消