Forward declarations and shared_ptr - c++

I'm trying to refactor my code so that I use forward declarations instead of including lots of headers. I'm new to this and have a question regarding boost::shared_ptr.
Say I have the following interface:
#ifndef I_STARTER_H_
#define I_STARTER_H_
#include <boost/shared_ptr.hpp>
class IStarter
{
public:
virtual ~IStarter() {};
virtual operator()() = 0;
};
typedef boost::shared_ptr<IStarter> IStarterPtr;
#endif
I then have a function in another class which takes an IStarterPtr object as argument, say:
virtual void addStarter(IStarterPtr starter)
{
_starter = starter;
}
...
IStarterPtr _starter;
how do I forward declare IStarterPtr without including IStarter.h?
I'm using C++98 if that is of relevance.

Shared pointers work with forward declared types as long as you dont call * or -> on them so it should work to simply write :-
class IStarter;
typedef boost::shared_ptr<IStarter> IStarterPtr;
You need to include <boost/shared_ptr.hpp> of course

Though it would add a header file, you could put that in a separate header file :
#include <boost/shared_ptr.hpp>
class IStarter;
typedef boost::shared_ptr<IStarter> IStarterPtr;
and then include it both in IStarter.h and in your other header, avoiding code duplication (though it's quite small in this case).
There might be better solutions though.

You can't forward declare typedefs in C++98 so what I usually do in this case is pull out the typedefs I need an put them into a types.h file, or something similar. That way the common type code is still separated from the definition of the class itself.

There is a way but you need to include the boost header in your file :
#include <boost/shared_ptr.hpp>
class IStarter;
typedef boost::shared_ptr<IStarter> IStarterPtr;
// ...
virtual void addStarter(IStarterPtr starter)
{
_starter = starter;
}
// ...
IStarterPtr _starter;

Related

Can I provide an incomplete header for a C++ class to hide the implementation details?

I would like to split a class implementation into three parts, to avoid that users need to deal with the implementation details, e.g., the libaries that I use to implement the functionality:
impl.cpp
#include <api.h>
#include <impl.h>
Class::Class() {
init();
}
Class::init() {
myData = SomeLibrary::Type(42);
}
Class::doSomething() {
myData.doSomething();
}
impl.h
#include <somelibrary.h>
class Class {
public:
Class();
init();
doSomething();
private:
SomeLibary::Type myData;
}
api.h
class Class {
Class();
doSomething();
}
The problem is, that I am not allowed to redefine headers for the class definition. This does not work when I define Class() and doSomething() only in api.h, either.
A possible option is to define api.h and do not use it in the project at all, but install it (and do not install impl.h).
The obvious drawback is, that I need to make sure, that the common methods in api.h and impl.h always have the same signature, otherwise programs using the library will get linker errors, that I cannot predict when compiling the library.
But would this approach work at all, or will I get other problems (e.g. wrong pointers to class members or similar issues), because the obj file does not match the header?
The short answer is "No!"
The reason: any/all 'client' projects that need to use your Class class have to have the full declaration of that class, in order that the compiler can properly determine such things as offsets for member variables.
The use of private members is fine - client programs won't be able to change them - as is your current implementation, where only the briefest outlines of member functions are provided in the header, with all actual definitions in your (private) source file.
A possible way around this is to declare a pointer to a nested class in Class, where this nested class is simply declared in the shared header: class NestedClass and then you can do what you like with that nested class pointer in your implementation. You would generally make the nested class pointer a private member; also, as its definition is not given in the shared header, any attempt by a 'client' project to access that class (other than as a pointer) will be a compiler error.
Here's a possible code breakdown (maybe not error-free, yet, as it's a quick type-up):
// impl.h
struct MyInternal; // An 'opaque' structure - the definition is For Your Eyes Only
class Class {
public:
Class();
init();
doSomething();
private:
MyInternal* hidden; // CLient never needs to access this! Compiler error if attempted.
}
// impl.cpp
#include <api.h>
#include <impl.h>
struct MyInternal {
SomeLibrary::Type myData;
};
Class::Class() {
init();
}
Class::init() {
hidden = new MyInternal; // MUCH BETTER TO USE unique_ptr, or some other STL.
hidden->myData = SomeLibrary::Type(42);
}
Class::doSomething() {
hidden->myData.doSomething();
}
NOTE: As I hinted in a code comment, it would be better code to use std::unique_ptr<MyInternal> hidden. However, this would require you to give explicit definitions in your Class for the destructor, assignment operator and others (move operator? copy constructor?), as these will need access to the full definition of the MyInternal struct.
The private implementation (PIMPL) idiom can help you out here. It will probably result in 2 header and 2 source files instead of 2 and 1. Have a silly example I haven't actually tried to compile:
api.h
#pragma once
#include <memory>
struct foo_impl;
struct foo {
int do_something(int argument);
private:
std::unique_ptr<foo_impl> impl;
}
api.c
#include "api.h"
#include "impl.h"
int foo::do_something(int a) { return impl->do_something(); }
impl.h
#pragma once
#include <iostream>
struct foo_impl {
foo_impl();
~foo_impl();
int do_something(int);
int initialize_b();
private:
int b;
};
impl.c
#include <iostream>
foo_impl::foo_impl() : b(initialize_b()} { }
foo_impl::~foo_impl() = default;
int foo_impl::do_something(int a) { return a+b++; }
int foo_impl::initialize_b() { ... }
foo_impl can have whatever methods it needs, as foo's header (the API) is all the user will see. All the compiler needs to compile foo is the knowledge that there is a pointer as a data member so it can size foo correctly.

Using multiple structs through header files C++

Apologies if you have seen this question before however it has yet to be answered, essentially in my code I have two structs, defined in separate headers and used globally throughout the project. I simply wish to use both structs (which again, are defined in two separate headers) in other cpp files than just the ones that the header file belongs to.
Here is some sample code which I have tested:
class1.h
#include "class2.h"
#include <vector>
#include <string>
struct trans1{
string name;
};
class class1 {
private:
vector <trans2> t2;
public:
class1();
};
class2.h
#include "class1.h"
#include <vector>
#include <string>
struct trans2{
string type;
};
class class2{
private:
vector <trans1> t1;
public:
class2();
};
errorlog:
In file included from class1.h:3:0,
from class1.cpp:1:
class2.h:21:13: error: 'trans1' was not declared in this scope
vector <trans1> t1;
^
class2.h:21:19: error: template argument 1 is invalid
vector <trans1> t1;
^
class2.h:21:19: error: template argument 2 is invalid
I understand that this is ridiculous code in a real world application however this is the simplest way I could demonstrate.
It is worth noting that if I simply comment out the declaration of vector t1 or t2 under 'private:' the code compiles without fail. It is just the fact I am using a second struct.
Any help anyone? Thanks.
Simply forward-declare the classes that will be used. Put all implementation code into a cpp file, not inline in the header.
Make the vector private. This way no file that includes the header can force code generation against an incomplete class.
you can try to forward declare trans1 in class2.h and trans2 in class1.h like this:
class2.h :
// includes
struct trans1;
// rest of your code
the same thing (but with trans2) in class1.h
Don't forget to add Include guards in your code!
edit: and yes, you need to change your vectors to store pointers, otherwise it won't link
You need to put the "trans" structs in their own header file(s) and include them in your class header files.
You could forward declare them, but this would require changing your vector to use pointers. (In that case I would recommend std::vector<std::unique_ptr<trans>>). This could be appropriate if your structs are big and complex.
The main advantage of the forward-declaration approach is to reduce compile times. However if the structs are really so simple as in your example, I wouldn't bother with the extra overhead of using pointers here.
If You were to do this in single .cpp file, the solution would be trivial:
struct trans1 { ... };
struct trans2 { ... };
class class1 { ... };
class class2 { .... };
Now you just need to rearrange the code to get this result in every translation unit. (the order of classes/structs in the file is important)

C++ circular include

I can't solve this circular dependency problem; always getting this error:
"invalid use of incomplete type struct GemsGame"
I don't know why the compiler doesn't know the declaration of GemsGame even if I included gemsgame.h
Both classes depend on each other (GemsGame store a vector of GemElements, and GemElements need to access this same vector)
Here is partial code of GEMELEMENT.H:
#ifndef GEMELEMENT_H_INCLUDED
#define GEMELEMENT_H_INCLUDED
#include "GemsGame.h"
class GemsGame;
class GemElement {
private:
GemsGame* _gemsGame;
public:
GemElement{
_gemsGame = application.getCurrentGame();
_gemsGame->getGemsVector();
}
};
#endif // GEMELEMENT_H_INCLUDED
...and of GEMSGAME.H:
#ifndef GEMSGAME_H_INCLUDED
#define GEMSGAME_H_INCLUDED
#include "GemElement.h"
class GemsGame {
private:
vector< vector<GemElement*> > _gemsVector;
public:
GemsGame() {
...
}
vector< vector<GemElement*> > getGemsVector() {
return _gemsVector;
}
}
#endif // GEMSGAME_H_INCLUDED
Remove the #include directives, you already have the classes forward declared.
If your class A needs, in its definition, to know something about the particulars of class B, then you need to include class B's header. If class A only needs to know that class B exists, such as when class A only holds a pointer to class B instances, then it's enough to forward-declare, and in that case an #include is not needed.
If you deference the pointer and the function is inline you will need the full type. If you create a cpp file for the implementation you can avoid the circular dependecy (since neither of the class will need to include each others .h in their headers)
Something like this:
your header:
#ifndef GEMELEMENT_H_INCLUDED
#define GEMELEMENT_H_INCLUDED
class GemsGame;
class GemElement {
private:
GemsGame* _gemsGame;
public:
GemElement();
};
#endif // GEMELEMENT_H_INCLUDED
your cpp:
#include "GenGame.h"
GenElement::GenElement()
{
_gemsGame = application.getCurrentGame();
_gemsGame->getGemsVector();
}
Two ways out:
Keep the dependent classes in the same H-file
Turn dependency into abstract interfaces: GemElement implementing IGemElement and expecting for IGemsGame, and GemsGame implementing IGemsGame and containing a vector of IGemElement pointers.
Look at the top answer of this topic: When can I use a forward declaration?
He really explains everything you need to know about forward declarations and what you can and cannot do with classes that you forward declare.
It looks like you are using a forward declaration of a class and then trying to declare it as a member of a different class. This fails because using a forward declaration makes it an incomplete type.

how to use not yet defined datatypes in a header?

I am programming on linux using g++ and I often encounter the problem that I need to use a class or data type in a header file which I define later, either at a later point in the header or in another header file.
For instance look at this header file:
class example
{
mydatatype blabla;
};
struct mydatatype
{
int blablainteger;
char blablachar;
};
This will give error because mydatatype is used before its defined
so usually I change it like this:
struct mydatatype; // <-- class prototype
class example
{
mydatatype *blabla; // <-- now a pointer to the data type
// I will allocate the data during runtime with the new operator
};
struct mydatatype
{
int blablainteger;
char blablachar;
};
Now it works. I could often just put the definition above, or include the header which is needed, but I don't want to include headers in a header or juggle with the definition order, it always gets messy.
The solution I showed usually works, but now I have encountered a new phenomenon. This time the datatype is not a class but a typedef, I cant use prototypes for a typedef and I don't want to use the actual datatype which the typedef incorporates.. it's messy too.
Is there any solution to this?
Firstly, the solution you've thought of (prototype and pointer), is unneeded, and slower than just implementing it without the pointer.
The "proper" solution for this, would be creating seperate headers for each type, and then include them in your other header. That way it will always be defined! You can even make them so that they include eachother.
However, if you've ever opened a .h file provided by g++, you've most likely seen this at the start of the header:
#ifndef SOMETHING_H
#define SOMETHING_H
// Code
#endif /* SOMETHING_H */
This is to solve the issue of types redefining themselves.
If they weren't there, and you included the header file multiple times, the types would be redefined, and an error would be thrown. This makes it so that the types are always present, but never included twice.
I hope that helps!
Place each class/type in it's own header file, and then include the relevant header file in other headers where you need it. Use an inclusion guard in each header e.g.:
// SomeHeaderFile.h
#ifndef SOME_HEADER_FILE_H
#define SOME_HEADER_FILE_H
// code
#endif
I disagree that this is messy - it allows you have an organised structure to you project, it allows each class to operate independently of others and without worrying about order, and it's a good idea to place each class in it's own file anyway.
You could just define the class inside the other class like
template<class T>
class vertex {
private:
class edge {
public:
vertex<T> *to;
double weight;
edge() {
weight = INFINITY;
to = NULL;
};
} *paths;
T data;
unsigned nof_paths;
public:
vertex(T val) {
data = val;
paths = NULL;
nof_paths = 0;
}
void addPathTo(vertex<T>*&);
edge* getAllPaths() {
return paths;
};
};
Obviously this works for small classes... if your class is ENORMOUS you'll be better using separate header files like the other guys said.

c++ headers keeping sane

The biggest problem I seem to run into when coding in c++ is the fact that you must declare a class before you can reference it. Say I have two header file like this...
Header1.h
#include "Header2.h"
#include <deque>
#include <string>
#include <iostream>
using namespace std;
class HelloPackage;
class Hello
{
public:
string Message;
HelloPackage * Package;
Hello():Message("")
{
}
Hello(string message, HelloPackage * pack)
{
Message = message;
Package = pack;
}
void Execute()
{
cout << Message << endl;
//HelloPackage->NothingReally doesn't exist.
//this is the issue essentially
Package->NothingReally(8);
}
};
Header2.h
#include "Header1.h"
#include <deque>
#include <string>
#include <iostream>
using namespace std;
class HelloPackage
{
public:
deque<Hello> Hellos;
HelloPackage()
{
Hellos = deque<Hello>();
}
int AddHello(string Message)
{
Hellos.push_back(Hello(Message,this));
}
void ExecuteAll()
{
for each(Hello h in Hellos)
h.Execute();
}
int NothingReally(int y)
{
int a = 0;
a += 1;
return a + y;
}
}
What I'm wondering is, is there any elegant solution for dealing with these issues? In say c#, and java, you're not restricted by this "linear" compiling.
Use header include guards, either "#ifndef / #define / #endif", or "#pragma once"
Put your code in a .cpp, not inline in the header
???
Profit
The reason this will work for you is because you can then use forward declarations of the class you want to reference without including the file if you so wish.
You are missing include guards
why define methods in the header?
Besides these problems with your code, to answer your question : normal way is to forward declare classes - not to include headers in headers (unless you have to).
If you follow a few basic rules, it is not awkward at all. But in comparison to e.g. Java or C#, you have to follow these rules by yourself, the compiler and/or language spec does not enforce it.
Other answers already noted that, but I will recap here so you have it in one place:
Use include guards. They make sure that your header (and thus your class definition) is only included once.
Normally, you will want to separate the declaration and implementation of your methods. This makes the header files more reusable and will reduce compilation time, because the header requires normally fewer #includes than the CPP (i.e. implementation) file.
In the header, use forward declarations instead of includes. This is possible only if you just use the name of the respective type, but don't need to know any "internals". The reason for this is that the forward declaration just tells the compiler that a certain name exists, but not what it contains.
This is a forward declaration of class Bar:
class Bar;
class Foo {
void foooh(Bar * b);
};
Here, the compiler will know that there is a Bar somewhere, but it does not know what members it has.
Use "using namespace xyz" only in CPP files, not in headers.
Allright, here comes your example code, modified to meet these rules. I only show the Hello class, the HelloPackage is to be separated into header and CPP file accordingly.
Hello.h (was Header1.h in your example)
#include <string>
class HelloPackage;
class Hello
{
public:
Hello();
Hello(std::string message, HelloPackage * pack);
void Execute();
private:
string Message;
HelloPackage * Package;
};
Hello.cpp
#include "Hello.h"
#include "HelloPackage.h"
using namespace std;
Hello::Hello() : Message("")
{}
Hello::Hello(string message, HelloPackage * pack)
{
Message = message;
Package = pack;
}
void Hello::Execute()
{
cout << Message << endl;
// Now accessing NothingReally works!
Package->NothingReally(8);
}
One question that may arise is why is the include for string is needed. Couldn't you just forward declare the string class, too?
The difference is that you use the string as embedded member, you don't use a pointer to string. This is ok, but it forces you to use #include, because the compiler must know how much space a string instance needs inside your Hello class.