OOP

========================================================================

Members

========================================================================

Regular Members

const member functions
- it can only call other const member functions
- e.g
```
const T& operator[] const(size_type i) {return array[i];}
```
  - 第一个const 就是返回一个const T& object，
  - 第二个const 是在执行过程中对objectread only access. 如果没有第一个const，他并不会在compile 中出现问题，但是这会在使用中出现问题。若无第二个const，便无法保证他object不会被改变。那么compile的时候就会出现问题！
  - mutable 可能是一个很好的工具，他可以让这个指针本身会变，即使你用const template function.如果你有const member functions 的话，有的member的子variable可能不让你 access。这个时候你应该加上mutable。
  - 或者用const_cast
- const member variable: must be initialized in ctor.
deleted functions generally are declared as public, to generate better error messages.
something& cannot be a class member? yes,but you have to declare and initialize them at the same time.
- about std::function as a member variable: Just use std::function, & or const std::function& are error prone

Static Member functions, variables

Access Specifiers:

static members can be accessed without any objects
non-static functions can access static members. no need for class name!
static member function:
- 用作整个class 的通用function，这个function 只和class 相关，和object 无关。他的目的仅仅是access static member variable，无法access 其他variables
- 他是 class::static_func, 所以他可以避免overloading，进而可以拿class:static_func来作为template argument

static member variable: static members must be defined outside of the class, in the same namescope using ::

class Foo{
    public: 
        // This is declaration, see below for definition
        static int i; 
        // can declare static func inside class
        static void foo_static(){
            cout<<__FUNCTION__<<endl;
        }
        // calls static, if in the same class, no need to add class name
        void foo(){
            foo_static();
            Foo::foo_static();
        }
};

class Child:public Foo{
    public:
        void foo(){
            // even child class can call the base class static func
            foo_static();
        }
}; 

// This is how to initialize static member func, this is definition
// If in hpp, then every instance will get the same value. Else, should be in the source file.
int Foo::i(1); 

int main()
{
    Foo f;
    f.foo(); 
    cout<<Foo::i<<endl; 
    Foo::foo_static();

    Child c; 
    c.foo();
    return 0;
}

========================================================================

Accessibility

========================================================================

Basic

Who can access what?

friend functions: can access private members
- 【习惯】friend function 不需要Public 或private。所以你可以把所有friend function 放到class definition 最上面，接近public位置。
```
 class foo{
 	friend std::istream& operator>>(std::istream&, Str&)
 	public:
 	...
 }
```
  - 这个friend function 和member function 是同等privilege
  - 有时，你不用friend function:
    - Can I add std::istream & operator>>(std::istream & is, Transform2D & tf); as a friend function to the Transform2D? This way I can modify private member Matrix's values? A: you don't need this
Friend Class - can access friend funcs too
- e.g, student_handle 列为core 的friend class，这样all member functions in student_handle will have access to core's private, protected, and public members (program verified).
```
class Core{
        friend class Student_Handle; 
    }
```
- friend class 无法传递：即parent 的friend 不是child friend (所以data member 你都获得不了，不过你可以根据parent 的virtual function 来接触child class); friend 的friend 不可以传递哦。
- Friend class is not mutual.

Inner class: enclosed class is actually a friend class, so it can access private members as well

class Enclosing {
    private:
       int x = 30;
    public:
       class Nested {
           public:
              void nested_func(Enclosing *e) {
                cout<<e->x;  // works fine: nested class can access
                             // private members of Enclosing class, if placed in
                             // either private or public
              }
       }; 
       Nested n_;   // Nested must have been declared
       void func(){
           n_.nested_func(this); 
       }
}; // declaration Enclosing class ends here

========================================================================

Constructors

========================================================================

High level summary

Compiler will generate funcs that are public, inline, non virtual (except virtual destructor for a child class) :
1. default ctor (if there's no other ctor), and the big five
2. default destructor
3. default copy operations:
  - ctor
  - copy assignment, =
4. default move operations (C++11), Each performs memberwise copy and assignment, as well as base-class parts. If no support for move, then copy
  - ctor
  - move assignment, =
ctor's initializer doesn't have a complete this pointer yet.
Rules of three, five, zero
- "Rule of three" means: if one of copy ctor, =, destructor is declared, then others should be declared as well. same reasoning as move operations.
  - STL containers with dynamic memory management all have the three implemented.
- Rule of Five just includes the move ctor and =
- Rule of zero means if you don't touch Any of the above, don't declare anything. If you've touched any of them, follow the previous rules.
Initializations
- direct initialziation uses () or {} obj(args); or obj{args}
- copy initialization uses = for contruction.

Declare something as default, if the default memberwise behaviour satisfies your need:

  class Foo{
      Foo() = default; 
      ~Foo() = default; 
      Foo(const Foo&) = default;  //copy ctor
      Foo& operator= (const Foo&) = default;//= 
      Foo(Foo&& ) = default;      //move copy ctor
      Foo& operator= (Foo&&) = default;    //move =
    }

templated copy ctor, = has no effect on the five, yayy

  class Foo{
  //copy ctor from anything, even though it could be default constructor, but you can still do rule of zero  
  template <typename T>
  Foo(const T& rhs);    
    }

What Ctors are synthesized

Will automatically synthesize all three even if you've had some of them. article

You provide \ synthesize Dtor Copy assignment Copy ctor Move ctor Move assignment

dtor v v x x

Copy= v v x x

Copy ctor v v x x
1. Dtor, copy, copy =, technically rule of 3 but not enforced, c++98 2. Exception: can't synthesize operator = (copy =) , reference member in the class (have to be initialized).
  1. Const member (have to be initialized first)
Also, if you have some of the above, "there're complex resource management, no move ctor, move assignment",
1. rule of 5, but to keep backward compatibility with c++98, only move and move = are enforced
If you have move ctor, move = , then it can't be c++98, so rule of 5 holds

You provide \ synthesize Dtor Copy assignment Copy ctor Move ctor Move assignment

Move ctor x x x x x

Move assignment x x x x x
No default ctor in the following cases: "maybe the developer doesn't want a default ctor"

Copy = (weird)

Move

Direct ctor, Class(int x, double y)

You provide \ synthesize	Dtor	Copy assignment	Copy ctor	Move ctor	Move assignment
dtor		v	v	x	x
Copy=	v		v	x	x
Copy ctor	v	v		x	x

You provide \ synthesize	Dtor	Copy assignment	Copy ctor	Move ctor	Move assignment
Move ctor	x	x	x	x	x
Move assignment	x	x	x	x	x

Copy = (weird)
Move
Direct ctor, Class(int x, double y)

Conversion and Explicit

Conversion Ctor(ctor can take 1 arg) and Explicit (doesn't allow implicit conversion, but can still be casted explicitly)

class oop
{
public:
    // this is actually a conversion ctor. Can convert 
    oop (int i = 3, int j = 4.0): i_(i), j_(j){}
    bool operator == (const oop& rhs){ return (rhs.i_ == i_) & (rhs.j_ == j_); }
private:
    int i_; 
    int j_;
};

class oop_explicit
{
public:
    explicit oop_explicit (int i = 3, int j = 4.0): i_(i), j_(j){}
    bool operator == (const oop_explicit& rhs){ return (rhs.i_ == i_) & (rhs.j_ == j_); }
private:
    int i_; 
    int j_;
};

int main()
{
    oop o;
    // conversion. Will not happen if ctor is explicit
    cout<<(o == 1)<<endl;

    oop_explicit oe; 
    // Explicit prohibits this
    // oe = 4;

    // you can still cast for explicit 
    oe = (oop_explicit)4;
}

Explicit should be in front of non-copy ctor (such as direct initialization), but doesn't work with copy constructor (using =)

class regex{
  explicit regex(char*);  // direct initialization permits this
}
std::regex rx = nullptr;    //won't compile, as = is copy initialization
regexes.push_back(nullptr);     // won't compile, because vector::push_back uses copy initialization
regexes.emplace_back(nullptr)   //compiles, since vector::emplace_back() uses direct initialization, though we might get undefined behavior

Details about Ctors

Copy constructor

No need for copy constructor, if the default is fine. The default copy constructor will recursively iterate thru the class.

Copy elision happens during =, so after we create a temp object, we don't have to copy it. This is called "copy elision"

class B { 
public:     
  B(const char* str = "\0") //default constructor { 
      cout << "Constructor called" << endl; 
  }     
    
  // this must be a reference, else you are going to call copy constructor itself when copying for the first time.
  B(const B &b)  //copy constructor { 
      cout << "Copy constructor called" << endl; 
  }  
}; 
 
int main() {  
  B ob = "copy me";  // only default contructor is called for creating a temp object, but we don't copy it
  return 0; 
}

If we use "-fno-elide-constructors" for g++ compiler, then copy ctor will be called

when would copy assignment throw? not enough memory/the copy ctor throws one explicitly

Copy Assignment Operator: This is called when the object has been initialized
- Assignment operator is actually copying, too. Assignment operator returns a reference, so you can do chaining a = b =c
- E.g
```
  class Foo{
      Foo& operator = (const Foo& f){
          if (&f = this) return *this;	//Ego check
          Foo.a = f.a;
          return *this; 
        }
    }
```
- Copy vs assignment operator: if object hasn't been declared, copy is called; else, assignment is called
```
class Foo{
Foo(); 
Foo(const Foo& f){cout<<"assignment"; }
Test& operator= (const Foo& t){cout<<"copy";}
}
Foo T1, T2; 
T1 = T2;		//calls = since T1 already exits
Foo T3 = T2;	// calls copy constructor
```
- Cautions
  1. Assignment Operator is Not defined if:
    - You have const non-static member (of course, they should be defined right on the spot, or in intializer list)
    - Your data member doesn't have an accessible copy constructor (ofc, you're copying stuff over)
    - Your base class doesn't have an accessible copy ctor
move constructor and assignment
- What is move constructors? Take the pointer to the resource, then nullify the moved-from object (x-value, in a "valid but unspecified state").
  - unique_ptr -> nullptr
  - std::thread -> empty thread, not joinable
  - std::future
  - std::string: random value...
- move assignment reminder
```
Obj& operator=(Obj&&){
  std::cout<<__FUNCTION__<<"RICO TODO "<<std::endl;
  return *this;
}
```
- when move ctor is not called:
  - Move operations rely on std::move, and if there's no support for move (no move ctor or assignment), then copy is performed
  - Some containers, such as std::vector::push_back(), std::stack::push(), have strong exception safety. So you move contructor must have noexcept, otherwise, copy ctor will be used.
- Cautions:
  - You may need a move constructor for std::vector resizing.
  - default move ctor can be ill-formed, if one of your members is not movable
    - Store smart pointers to that instead

Use another ctor (delegation)

ctor(int i) : ctor(i, 1,2,3)    //always have the longer ctor here
{}

Initialization

good read. For more info about initialization basics, check out c++_basics
When Initializer List must be used (Not to be confused with std::initializer_list)
- list initializer is to "unify" initialization. So, it wors with ctors with different args. e.g, vector{beg, end} also works. see code
- for const members, you need to use initializer list. Reasons is we don't need to default construct it.
- reference members must be initialized using initializer list.
```
class Foo{
    public:
    Foo(int a, int b); 
  }

class Main{
    public: 
      MyClass(int a, Foo f) : x(a), Foo(a,a){} // pass in the arguments directly for const member
    const int x; 
    const Foo foo; 
  }
```
- Note: carefule when const & is a member variable: it does create a temp copy as it would in function. but it will quickly get out of scope. Assigning that to an rvalue will result in dangling reference
```
using Fn = std::function<bool (void)>; 
class{
    Fn& fn1_;   // BAD: needs to be declared in initialization. You can't pass in nullptr, that's a rvalue. 
    const Fn& fn2_ = nullptr;   // BAD: dangling pointer. 
}
```

copy initialization:

Two ways to initialize const members: ctor or in place initialization. The "in place" intiailization is also called "default member initialization".

// in place member initialization
class Foo{
  private:
    Obj obj_;     // requires default initialization
    Obj obj2 = Obj(1,2,3) //copy ctor
    Obj obj3 = {1,2,3}    //implicit conversion from 1,2,3 to Obj, which requires ctor Obj(int, int int) NOT to have explicit, and copy ctor
    Obj obj4{1,2,3}     // list initialization, not copy initialization
}

list initalization:
- struct with no ctor, and all public members:
  1. One form is of list initialization is "aggregate initialization". the declaration order must match that in struct. Not to be confused with designated initialization in C.
  2. when working with pointers, make_unique() doesn't support direct initialization, but new does.
```
struct Foo{
  int i; 
  double j;
};
// note it has to be Foo{}, not Foo()
new Foo{2, 3.0}
```
- In C++, aggregate also needs:
  1. No custom ctor,
  2. No virtual functions, no base classes
  3. No protected, private static members
- Stuff doesn't work with list initialization:
  1. [!] std::vector::emplace_back since it doesn't take lists
```
std::vector<MyStruct>vec;
vec.emplace_back(arg1, arg2);       // don't compile.
```
  2. [!] std::make_unique, cuz arg can be deduced as things that can cause narrowing, e.g., int, but actually char. But works with std::unique_ptr(new T{1,2}). Verified. See code

Evolution of ctor

C++98: constructor + initializer list

C++ 11: in class non-static member initialization

// since C++11:
struct SimpleType {
  int field = 0;                        // works now!
  std::string name { "Hello World "} // alternate way with { }
}

C++ 17: in class static member declaration with inline

struct OtherType {
  static const int value = 10;
  static inline std::string className = "Hello Class";
}

========================================================================

Overloading

========================================================================

Operator Overloading

ctor compatible with 1 arg (aka conversion ctor): see the constructor section

conversion operators

class Int{
	operator double (){return 3.0;} // This is conversion operator
    // explicit conversion
    explicit operator int*() const { return nullptr; }
	}; 
int main(){
    Int i; 
    double a = i;	//using conversion operator. 
    double b = static_cast<Int>(i)	//using conversion operator

    int* p = static_cast<int*>(x); // OK: sets p to null
    //  int* q = x; // Error: no implicit conversion, explicit conversion only
}

Function Overloading

rvalue reference, lvalue reference funcs:

class Bar{
    public: 
        Foo& getFoo()& {cout<<"lvalue foo is returned "<<endl; return foo;}	//foo is already lvalue

        // cannot be overloaded with two getFoo()&& functions
        // Foo& getFoo()&& {
        //    cout<<"rvalue foo is returned "<<endl; 
        //    return foo;
        // }	

        // good
        Foo&& getFoo()&& {cout<<"rvalue foo is returned "<<endl; return std::move(foo);}

========================================================================

Inheritance

========================================================================

Short summary
- You CANNOT call a function defined only in derived class through parent class ptr
- All code is here: see code

name overloading bw function and variable

// in c
int a;
void a(){}
int main(){
  a();    // error: variable cannot be used as function.
}

// in cpp
class Foo{
  int a;
  void a(){}    // this is not allowed either
}

// case 3:
class Foo{
  void a(){}
  void foo(){
    int a;
    this->a();    // this is allowed, because we are not relying on implicit invoking.
  }
}

Inheritance: object slicing from up-casting (references, or pointers, derive -> base)

Base part is built first, then the derived part
- default base ctor is called. That's why you want to call the copy/move ctor explicitly from derived class. This makes sense because the derived class object naturally has a base part to copy
```
  class Foo{}; 
  class Bar : public Foo{
      Bar(const Bar& b) : Foo(b){}
      Bar(Bar&& b) : Foo(std::move(b)){}
    };
```

when passed by value, the object gets sliced off

  class Foo{
      virtual void eat(); 
    }; 
  class Bar:public Foo{
      void eat override{}; 
    }; 

  Eat_value(Foo f){f.eat()};  //calls Foo::eat()
  Eat_ref(const Foo& f){f.eat();}     //can call Bar::eat() or Foo::eat()

template inheritance:

Base ctor must come before everything else

template <typename T>
class Foo
{
    public:
      Foo (T t){}
    protected: 
      int num_;
};
class Bar : public Foo<int>{
    Bar::Bar() : Foo(1), basic_info_(){}
}

Multiple Inheritance:

class TA : public Faculty, public Student {
}

Polymorphism

Vtable and virtual function
- virtual allows polymorphism, which is to call child class function thru base class reference or ptr. The derived class function is called using vtable, created during compile time. The virtual function is resolved late during runtime.
  - "dynamic dispatch" is the word for achiecving function overriding (polymorphism), NOT overloading. You need the virtual keyword, otherwise, NO POLYMORPHISM IS ACHIEVED
- Every class constructor will create a vtable, then create a vptr to point to the vtable. (For derived class, same).
- When you call a virtual function, compiler uses base_class ptr -> find the right vptr -> access the right vtable -> find the right function for the current class.

call virtual function in base: read below for some common senarios

Class Base
{   
    void func(){cout<<"base"<<endl;}
    virtual void Base_func(){cout<<"base_only"<<endl; }
    virtual void f(){g();};
    virtual void g(){//Do some Base related code;}
};

Class Derived : public Base
    void func(){cout<<"Derived"<<endl;}
    virtual void f(){Base::f();};
    virtual void g(){//Do some Derived related code};
};

int main()
{
    Base *pBase = new Derived;      //**call Base::base() -> Derived::Derived()**
    pBase -> func();		// calls base func, because it's not virtual. 
    pBase -> base_only();		//calls the Base base_only, since it's virtual, it's in the vtable. 
    pBase->f();         // **call Derived::f() -> Base::f() -> Derived::g()**  
    return 0;  
}

Notes:
- If there are "this" pointer, it will be resolved as well.
- note that Base::g() won't be called, because you can choose not to implement Derived::g() at all. Think the derived class object is completed, so if you want to call it, use Base::g()
  - There is one exception, where Base::g() will be called, that'd be during the constructor phase of Base, because then you don't have a completed Derived object yet (remember stack unwinding.)

Abstract function (or pure function), abstract class
- if an implementation of a fucntion is not known in base class declaration, then we have an abstract function, which is virtual ... = 0 in .h file.
  - An abstract class has AT LEAST one abstract function. The purpose is to serve as a base, enforce every child class to define it.
  - if an abstract class has no implementation for ANY function, and (it can have member variables), then it's called interface class
```
class MyInterface
{
public:
    virtual ~MyInterface() {}
    virtual void Method1() = 0;
    virtual void Method2() = 0;
};
```
  - we cannot create an object for an abstract class

override

Motivation: It's so easy to make mistakes for overriding exactly the same functions!, That's why we need override keyword. Here you have a few cases.

class Base{
  virtual void func1 const(); 
  virtual void func2 () &&; 
  virtual void func3 (int x) ; 
  void func4 (int x) ; 
}
class Derived : public Base{
  void func1 ();		// Not Overloading, because you need const. 
  void func2 () &;	// ref qualifiers should match as well 
  void func3 (unsigned int x) ;		//type not matching
  void func4 (int x) ;		//no virtual keyword in base
}

========================================================================

Destructor

========================================================================

To destruct object properly:
1. Be EXTREAMLY CAREFUL about the destruction sequence: main dtor (body), members in reverse order in initializer list.
2. but we almost never explicitly call the dtor. else, there'd be double-freeing at the end of the dtor
3. If you have a while loop running: have something like this: So certain items may already be destructed when your main loop might be destructed, which causes crashes
```
  //params for destruction
  std::atomic<bool> stop_requested_;
  std::atomic<bool> run_stopped_;

   ~Foo() {
    // stop the main loop
    stop_requested_ = true;
    while(!run_stopped_.load());
   }

  void run(){
    while(!stop_request_.load()){
        ...
    }
    run_stopped_ = true;
  }
```
  - if you have a thread pool, destruct it after your object
  - So place thread pool as the last item in member declaration, so the task in thread pool can be terminated properly
  - Caveat: if you have another thread loop the major task loop, this will still get stuck. solution: an atomic that signals that.

On some compilers, there might be no errors to call dtor several times, but this will technically lead to undefined behavior:

Obj obj; 
//lifetime is over here. see https://en.cppreference.com/w/cpp/language/lifetime
obj.~Obj();   
obj.~Obj(); 
obj.~Obj();

Private Destructor - will not affect constructor` - will not let main() or other non-child functions delete the object.

Virtual and Pure virtual destructor

need a virtual dtor for ALL polymorphism

class Base
{
    public:
    // should be this 
    // virtual ~Base() {}
    ~Base() { cout << "Base Destructor\n"; }
};

class Derived:public Base
{
    public:
    ~Derived() { cout<< "Derived Destructor\n"; }
}; 
int main()
{
    Base* b = new Derived;     // Upcasting
    delete b;       // not deleting the derived part, because there is virtual keyword before base dtor
}

Need pure virtual dtor for abstract class (same reason as virtual function). But also because we are not overriding the base class dtor, but calling it after the derived class dtor (which is different than other polymorphism). derived_class_1 dtor ... base_class dtor, we must define its body

class Base
{
    public:
    virtual ~Base() = 0;     // Pure Virtual Destructor
};

// Definition of Pure Virtual Destructor
Base::~Base() 
{ 
    cout << "Base Destructor\n"; 
} 

class Derived:public Base
{
    public:
    ~Derived() 
    { 
        cout<< "Derived Destructor"; 
    }
};

========================================================================

Small Details about OOP

========================================================================

Forward declatation - if your class needs another class that depends on yours (circular reference), you need to do this:
```
    class IAttack;
    class Mob{
        IAttack...
    }

    //Attack.cpp
    class Mob;
    class IAttack{
        Mob mob;
    }
```
- only declared, not defined yet. It's just telling the compiler that "it will be defined soon", so ok to use for pointers, references, not for the whole object.
```
// foo.hpp
class Foo{
    public: 
      Foo() : bar(Bar()){}    //havent' seen the definition yet. 
      Bar bar_; 
};

//foo.cpp
class Bar{}; 
```
  - seems like you can completely define the nested class outside of the hpp (forward-declaration). But this works only with pointers, Not concrete objects (tested)
    - deleting a pointer of incomplete type will lead to undefined behaviour.

Reminders

Reminder: class functions are like:

template<typename T>
class ThreadSafeQueue{
  ThreadSafeQueue ();
}; 

template<typename T>
ThreadSafeQueue<T>::ThreadSafeQueue(){}

Reminder: noexcept needs to be in both the declaration and the definition:
```
class Foo{
  void f()noexcept;
}; 
void Foo::f() noexcept{}
```
Expensive: delete (calls destructor)

========================================================================

Template

========================================================================

Basics
- T inside <> is called template argument
- Always define a template outside of a function (including main, ofc)
- template argument is always determined at compile time.
  - which means it's a constexpr. So use a constexpr function for that.
member functions of a template class need to be defined in .hpp
- Rationale: when compiled, template declaration needs to be known. Also, the template needs to see functions definition during compile time. Compiler don't remember the details of another .cpp, so if defined in a .cpp, the main .cpp see either the tempate code, or Foo<int>, but not both at the same time.

Correct way to declare template functions:

template <typename T>
class Thread_Safe_Stack{
  template<typename Q>
  void push(Q&& object); 
}; 

template <typename T>
Thread_Safe_Stack<T>::Thread_Safe_Stack(){}

template <typename T>
template <typename Q>
void Thread_Safe_Stack<T>::push(Q&& object){
    stack_.push(std::forward<Q>(object));
}

function template

template<class T> 
void foo(T t);       // Function template, itself is not a function, but a "generator" that generate function definitions.

templates can be used to deduce size_
```
 template<unsigned N>
     const_str(const char(&arr)[N]);
```
- here, it's not template, we are using a non-type parameter: they can be:
  - Pointers
  - references
  - Integral ConstExpr

Template Specialization When you have a primary template, but you want to have something different for a specific type, use template specialization.

template<typename T>
void is_voi(T input){};

// explicit specialization for T = char
template<>
void is_voi <char> (char input){};

//equivalent to above, since template arguments can be deducted from the function arguments.
template<>
void is_voi(char input){}

Cautions:
- Specialized function template must come after the generic template.
- Template specialization cannot be private

Overloading. (TODO)
- The loading sequence is: exact function > specialized function template > generic function template
- Function overloading is functions with the same name but with different args (args only, return type doesn't matter). Use overloading if you only have different args to pass in.
  - Function overloading is not overriding, they're different things. Polymorphism is to have the same function in different derived class.
  - Use polymorphism when you have the same arguments.
```
struct Base{
    void shadow(std::string){}
};
struct Derived: Base{
    use Base::shadow;
    void shadow(int){}
};
int main(){
    Derived derived;
    derived.shadow(std::string{});                  //Won't work unless you have
}
```

Variadic template: template that takes in any number of inputs, uysing ellipsis (...)

template <typename T, typename ... T2>
void foo(T var1, T2 ... vars){      //... is always between type and arg name
  cout<<var1<<endl; 
  foo(vars); 
}
print(1,2,3.15, "var");       //must have this as var 1 has been popped out     //must have this as var 1 has been popped out

========================================================================

Advanced Uses of Templates

========================================================================

Template Metaprogramming (TMP)
- Motivation: Compile time program, that can express any program. (turing complete).
- May not be super practical, interesting tho
- Example
```
  template <unsigned int n>
  struct Factorial{enum{val = val * Factorial<val - 1>::val};
  }

  template <>
  struct Factorial<0> { enum{val = 1}; }

  Factorial<8>::val;  // this returns 8!
```
  - when there's a new argument, compiler creates a new instance of the template
    - Factorial<8>::val is a different instance than Factorial<7>::val
  - why use enumeration ?//TODO
    - Because enum const expression types return in compile time
    - Why can you do Factorial::stuff? //not sure

type_traits

std::is_integral
This is a template
```
template <typename> struct is_integral
```

Usage

#include <type_traits>
cout<<std::is_integral<int>::value<<endl;
cout<<std::is_integral<int&>::value<<endl;    //returns false

std::is_same<class A, class B>
- returns true if A and B are the same
- cv qualifier (const and volatile) and reference is taken into account

std::decay::type

removes cv qualifier and references
converts array and function types to pointers ??

Use:

typename = std::enable_if<std::is_same<Person typename std::decay<T>::type>::value, T>::type;     //C++11
typename = std::enable_if_t<std::is_same<Person, std::decay_t<T>>::value, T>//C++14, with std::enable_if_t (omit typename), std::decay_t (omit ::type)

std::is_base_of<class A, class B>
- returns true if B is derived from A, or A itself
std::is_constructible<std::string, B>::value: constexpr true or false, if B can be used to construct std::string
- std::enable_if
- template
```
#include <type_traits>
template <bool cond, class T>
struct enable_if{}

template <class T>
struct enable_if<true, T>{
    using type = T; 
  }
```
- so when false is passed into cond, there's no type in struct. But because of SFINAE, i.e, errors will be ignored during template match searching, the compiler will keep going and try to find a best match

Usage: When you have the exact number of template params for two different template functions

template <class T, 
  class = typename std::enable_if<std::is_integral<T>::value, T>::type>    
  //you cannot do class std::enable_if ... here, have to be typename. This is a template argument
  // the templated function has two template args, one is T, one is the dependent type. So you can technically by pass this by do_stuff<double, float>
  // class = ... means the name of the param is optional. 
void do_stuff(){
  cout<<"YEEHAA"<<endl;     
}
int main() {
  do_stuff<int>(); 
}

Another example: two iterators

template <class _InputIterator >
vector (_InputIterator _first,
         typename std::enable_if<std::is_input_iterator<_InputIterator>::value, _InputIterator>::type _last)

template <class T>
vector (size_t size, T val)

Pros:
- No obscure error msg from having the same template and passing in the wrong type
So if enable_if is false, there's no ::type passed in. But why no error msgs
- Because error in function matching (resolving overloading) is fine, the compiler will keep searching for a match
- But in the function body, a compiler error is irreversable.

typename vs class (TODO): typename can be used to define a type that's dependent on template arg.
- Without typename the compiler doesn't know if this is a template type or not
```
template <typename param_t>
class Foo{
    typedef typename param_t NEWTYPE; 
  }

// or
typename some_template<T>::some_type     
```
class = TODO?

========================================================================

Typedef and Struct

========================================================================

struct

The only difference between struct and class is by default, members in struct are public.
You can optionally include a constructor:
```
struct Foo{int x; int y}
Foo foo = {x_val, y_val}; // pre c++11
Foo foo{x_val, y_val};    // since c++11
```
- in a struct, memories might be continguous if the members are of the same type, but don't bank on that. This is done using structural padding (TODO)

in C, struct is very different

Definition

typedef struct S{
    ...
} T; 
//equivalent to struct S{}; typedef S T; 
void (T something) //valid, because this is in the T is in the global namespace
void (struct S)  // valid. 
void (S something) //not valid, because S is not valid "type", but struct S is.

struct 里面是没有member function 的。

typedef

define a pointer

typedef int* int_ptr
//or
typedef struct lol{...}
lol* ptr;

C _Struct_Oop_Template - RicoJia/notes GitHub Wiki

OOP

Members

Regular Members

Static Member functions, variables

Accessibility

Basic

Who can access what?

Constructors

High level summary

What Ctors are synthesized

Conversion and Explicit

Details about Ctors

Initialization

Overloading

Operator Overloading

Function Overloading

Inheritance

Polymorphism

Destructor

Small Details about OOP

Template

Advanced Uses of Templates

Typedef and Struct

struct

typedef

⚠️ GitHub.com Fallback ⚠️

C _Struct_Oop_Template - RicoJia/notes GitHub Wiki

OOP

Members

Regular Members

Static Member functions, variables

Accessibility

Basic

Who can access what?

Constructors

High level summary

What Ctors are synthesized

Conversion and Explicit

Details about Ctors

Initialization

Overloading

Operator Overloading

Function Overloading

Inheritance

Polymorphism

Destructor

Small Details about OOP

Template

Advanced Uses of Templates

Typedef and Struct

struct

typedef

⚠️ **GitHub.com Fallback** ⚠️

⚠️ GitHub.com Fallback ⚠️