Splitting templated C++ classes into .hpp/.cpp files--is it possible?

I am getting errors trying to compile a C++ template class which is split between a .hpp and .cpp file:

$ g++ -c -o main.o main.cpp  
$ g++ -c -o stack.o stack.cpp   
$ g++ -o main main.o stack.o  
main.o: In function `main':  
main.cpp:(.text+0xe): undefined reference to 'stack<int>::stack()'  
main.cpp:(.text+0x1c): undefined reference to 'stack<int>::~stack()'  
collect2: ld returned 1 exit status  
make: *** [program] Error 1  

Here is my code:

stack.hpp:

#ifndef _STACK_HPP
#define _STACK_HPP

template <typename Type>
class stack {
    public:
            stack();
            ~stack();
};
#endif

stack.cpp:

#include <iostream>
#include "stack.hpp"

template <typename Type> stack<Type>::stack() {
        std::cerr << "Hello, stack " << this << "!" << std::endl;
}

template <typename Type> stack<Type>::~stack() {
        std::cerr << "Goodbye, stack " << this << "." << std::endl;
}

main.cpp:

#include "stack.hpp"

int main() {
    stack<int> s;

    return 0;
}

ld is of course correct: the symbols aren't in stack.o.

The answer to this question does not help, as I'm already doing as it says.
This one might help, but I don't want to move every single method into the .hpp file—I shouldn't have to, should I?

Is the only reasonable solution to move everything in the .cpp file to the .hpp file, and simply include everything, rather than link in as a standalone object file? That seems awfully ugly! In that case, I might as well revert to my previous state and rename stack.cpp to stack.hpp and be done with it.

It is not possible to write the implementation of a template class in a separate cpp file and compile. All the ways to do so, if anyone claims, are workarounds to mimic the usage of separate cpp file but practically if you intend to write a template class library and distribute it with header and lib files to hide the implementation, it is simply not possible.

To know why, let us look at the compilation process. The header files are never compiled. They are only preprocessed. The preprocessed code is then clubbed with the cpp file which is actually compiled. Now if the compiler has to generate the appropriate memory layout for the object it needs to know the data type of the template class.

Actually it must be understood that template class is not a class at all but a template for a class the declaration and definition of which is generated by the compiler at compile time after getting the information of the data type from the argument. As long as the memory layout cannot be created, the instructions for the method definition cannot be generated. Remember the first argument of the class method is the 'this' operator. All class methods are converted into individual methods with name mangling and the first parameter as the object which it operates on. The 'this' argument is which actually tells about size of the object which incase of template class is unavailable for the compiler unless the user instantiates the object with a valid type argument. In this case if you put the method definitions in a separate cpp file and try to compile it the object file itself will not be generated with the class information. The compilation will not fail, it would generate the object file but it won't generate any code for the template class in the object file. This is the reason why the linker is unable to find the symbols in the object files and the build fails.

Now what is the alternative to hide important implementation details? As we all know the main objective behind separating interface from implementation is hiding implementation details in binary form. This is where you must separate the data structures and algorithms. Your template classes must represent only data structures not the algorithms. This enables you to hide more valuable implementation details in separate non-templatized class libraries, the classes inside which would work on the template classes or just use them to hold data. The template class would actually contain less code to assign, get and set data. Rest of the work would be done by the algorithm classes.

I hope this discussion would be helpful.

It is possible, as long as you know what instantiations you are going to need.

Add the following code at the end of stack.cpp and it'll work :

template class stack<int>;

All non-template methods of stack will be instantiated, and linking step will work fine.

You can do it in this way

// xyz.h
#ifndef _XYZ_
#define _XYZ_

template <typename XYZTYPE>
class XYZ {
  //Class members declaration
};

#include "xyz.cpp"
#endif

//xyz.cpp
#ifdef _XYZ_
//Class definition goes here

#endif

This has been discussed in Daniweb

Also in FAQ but using C++ export keyword.

No, it's not possible. Not without the export keyword, which for all intents and purposes doesn't really exist.

The best you can do is put your function implementations in a ".tcc" or ".tpp" file, and #include the .tcc file at the end of your .hpp file. However this is merely cosmetic; it's still the same as implementing everything in header files. This is simply the price you pay for using templates.

Only if you #include "stack.cpp at the end of stack.hpp. I'd only recommend this approach if the implementation is relatively large, and if you rename the .cpp file to another extension, as to differentiate it from regular code.

I believe there are two main reasons for trying to seperate templated code into a header and a cpp:

One is for mere elegance. We all like to write code that is wasy to read, manage and is reusable later.

Other is reduction of compilation times.

I am currently (as always) coding simulation software in conjuction with OpenCL and we like to keep code so it can be run using float (cl_float) or double (cl_double) types as needed depending on HW capability. Right now this is done using a #define REAL at the beginning of the code, but this is not very elegant. Changing desired precision requires recompiling the application. Since there are no real run-time types, we have to live with this for the time being. Luckily OpenCL kernels are compiled runtime, and a simple sizeof(REAL) allows us to alter the kernel code runtime accordingly.

The much bigger problem is that even though the application is modular, when developing auxiliary classes (such as those that pre-calculate simulation constants) also have to be templated. These classes all appear at least once on the top of the class dependency tree, as the final template class Simulation will have an instance of one of these factory classes, meaning that practically every time I make a minor change to the factory class, the entire software has to be rebuilt. This is very annoying, but I cannot seem to find a better solution.

Sometimes it is possible to have most of implementation hidden in cpp file, if you can extract common functionality foo all template parameters into non-template class (possibly type-unsafe). Then header will contain redirection calls to that class. Similar approach is used, when fighting with "template bloat" problem.

The problem is that a template doesn't generate an actual class, it's just a template telling the compiler how to generate a class. You need to generate a concrete class.

The easy and natural way is to put the methods in the header file. But there is another way.

In your .cpp file, if you have a reference to every template instantiation and method you require, the compiler will generate them there for use throughout your project.

new stack.cpp:

#include <iostream>
#include "stack.hpp"
template <typename Type> stack<Type>::stack() {
        std::cerr << "Hello, stack " << this << "!" << std::endl;
}
template <typename Type> stack<Type>::~stack() {
        std::cerr << "Goodbye, stack " << this << "." << std::endl;
}
static void DummyFunc() {
    static stack<int> stack_int;  // generates the constructor and destructor code
    // ... any other method invocations need to go here to produce the method code
}

If you know what types your stack will be used with, you can instantiate them expicitly in the cpp file, and keep all relevant code there.

It is also possible to export these across DLLs (!) but it's pretty tricky to get the syntax right (MS-specific combinations of __declspec(dllexport) and the export keyword).

We've used that in a math/geom lib that templated double/float, but had quite a lot of code. (I googled around for it at the time, don't have that code today though.)

You need to have everything in the hpp file. The problem is that the classes aren't actually created until the compiler sees that they're needed by some OTHER cpp file - so it has to have all the code available to compile the templated class at that time.

One thing that I tend to do is to try to split my templates into a generic non-templated part (which can be split between cpp/hpp) and the type-specific template part which inherits the non-templated class.

The place where you might want to do this is when you create a library and header combination, and hide the implementation to the user. Therefore, the suggested approach is to use explicit instantiation, because you know what your software is expected to deliver, and you can hide the implementations.

Some useful information is here: https://docs.microsoft.com/en-us/cpp/cpp/explicit-instantiation?view=vs-2019

For your same example: Stack.hpp

template <class T>
class Stack {

public:
    Stack();
    ~Stack();
    void Push(T val);
    T Pop();
private:
    T val;
};


template class Stack<int>;

stack.cpp

#include <iostream>
#include "Stack.hpp"
using namespace std;

template<class T>
void Stack<T>::Push(T val) {
    cout << "Pushing Value " << endl;
    this->val = val;
}

template<class T>
T Stack<T>::Pop() {
    cout << "Popping Value " << endl;
    return this->val;
}

template <class T> Stack<T>::Stack() {
    cout << "Construct Stack " << this << endl;
}

template <class T> Stack<T>::~Stack() {
    cout << "Destruct Stack " << this << endl;
}

main.cpp

#include <iostream>
using namespace std;

#include "Stack.hpp"

int main() {
    Stack<int> s;
    s.Push(10);
    cout << s.Pop() << endl;
    return 0;
}

Output:

> Construct Stack 000000AAC012F8B4
> Pushing Value
> Popping Value
> 10
> Destruct Stack 000000AAC012F8B4

I however don't entirely like this approach, because this allows the application to shoot itself in the foot, by passing incorrect datatypes to the templated class. For instance, in the main function, you can pass other types that can be implicitly converted to int like s.Push(1.2); and that is just bad in my opinion.

Because templates are compiled when required, this forces a restriction for multi-file projects: the implementation (definition) of a template class or function must be in the same file as its declaration. That means that we cannot separate the interface in a separate header file, and that we must include both interface and implementation in any file that uses the templates.

Another possibility is to do something like:

#ifndef _STACK_HPP
#define _STACK_HPP

template <typename Type>
class stack {
    public:
            stack();
            ~stack();
};

#include "stack.cpp"  // Note the include.  The inclusion
                      // of stack.h in stack.cpp must be 
                      // removed to avoid a circular include.

#endif

I dislike this suggestion as a matter of style, but it may suit you.

The 'export' keyword is the way to separate out template implementation from template declaration. This was introduced in C++ standard without an existing implementation. In due course only a couple of compilers actually implemented it. Read in depth information at Inform IT article on export

1. Remember the main reason to separate .h and .cpp files is to hide the class implementation as a separately-compiled Obj code that can be linked to the user’s code that included a .h of the class.

2. Non-template classes have all variables concretely and specifically defined in .h and .cpp files. So the compiler will have the need information about all data types used in the class before compiling/translating  generating the object/machine code Template classes have no information about the specific data type before the user of the class instantiate an object passing the required data type:

    	TClass<int> myObj;
   

3. Only after this instantiation, the complier generate the specific version of the template class to match the passed data type(s).

4. Therefore, .cpp Can NOT be compiled separately without knowing the users specific data type. So it has to stay as source code within “.h” until the user specify the required data type then, it can be generated to a specific data type then compiled

I am working with Visual studio 2010, if you would like to split your files to .h and .cpp, include your cpp header at the end of the .h file