Intermediate C++ Game Programming Tutorial 21

From Chilipedia
Revision as of 05:54, 23 October 2019 by R vdw (Talk | contribs) (Video Timestamp Index)

Jump to: navigation, search

Smart pointers. I know pointers, I have the smartest pointers. In this video we learn about std::unique_ptr<>, which is by far the most frequently-used and important smart pointer. std::shared_ptr<> can go suck on an egg.

Topics Covered

  • std::unique_ptr<>
  • How to pass (and not to pass) smart pointers
  • Custom deleters

Video Timestamp Index

Tutorial 21

  • Risks of memory leaking when using pointers and (dynamically allocated) heap memory 0:30
    • Because you forget to delete memory
    • After copy constructing objects that contain pointers to ("that own") dynamically allocated memory: upon leaving scope, your program will try to delete the same memory twice, leading to an Access Violation exception
    • Calling a function the returns a pointer to dynamically allocated memory without assigning it, "dropping" that pointer (it is not being owned anymore)
  • Smart pointers / unique pointers can help avert these inadvertant memory leaks 2:39
    • They are RAII enabled, i.e. they take care of memory management (garbage collection) for you
    • No need to call delete on them, they will free the memory of the objects that they own in their destructors, which get called when they go out of scope
    • You #include <memory> and create a unique pointer through std::unique_ptr<T> p( new T );. However, from now on, we avoid the use of new in our code
  • Creating unique pointer in modern C++: std::unique_ptr<T> p = std::make_unique<T>() 3:57
    • This can be shortened to auto p = std::make_unique<T>()
    • To create a unique pointer to an array of n: auto p = std::make_unique<T[]>( n ). The unique pointer will auto destroy all objects of type T when leaving scope
    • Unique pointers can be dereferenced *p, or dereferenced to access a member p->someMember
    • If you want to call a function by the underlying raw (/naked) pointer, use SomeFunc( p.get() );. Unique pointer does not implicitly convert to a raw pointer to prevent you from accidently passing it to a function that might want to accept ownership of that pointer
  • Example: functions in a class to transfer ownership of a resource through smart pointers 7:00
    • Instead of implementing the member function T* Take() and void Give(T*)
    • We implement std::unique_ptr<T> Take() and void Give( std::unique_ptr<T> ) and avoid the risk of unintended memory leaks
    • When you are transferring an object via function, always pass unique pointers by value
  • You control ownership of the resource object by using move semantics 9:12
    • You cannot copy a unique pointer; you have to use objectHolder.Give( std::move( p ) ); where p is a unique pointer to a resource.
    • When implementing with smart pointers, you don't need to define destructors as this is taken care of by the smart pointer class: Rule-of-0
    • If you use unique pointers correctly, you're not going to get memory leaks. It leads to: safer code, less code, and speed of execution (oh yeah, triple whammy)
    • If you have a unique pointer in your class, the compiler won't let you copy construct (copy constructor is deleted) / shallow copy, as this would generate two pointers to the same memory block
    • The default move constructor will still work
  • Using a factory pattern (creating a factory function) to create unique pointers 11:47
    • Example: std::unique_ptr<T> MakeT() {return std::make_unique<T>();}
    • As MakeT() will return an rvalue, you don't need to call std::move on it, and you can write
      objectHolder.Give( MakeT() );
  • Implement smart pointers in the MemeFighter game 13:58
  • WORK-IN-PROGRESS

Bonus Video

The addition of smart pointers to Surface has allowed us to reduce the amount of code we need to write for Surface, making it simpler and correct by default. But we still need to write logic for those pesky copy operations... What if we used a container like std::vector! Then all that copying bullshit will be taken care of for us.

This is what we do in the bonus video. The only problem is, although adding std::vector will not make our release build any slower (the compiler optimizes out all of the extra abstraction for us), it does make the debug build slow. Usually, we don't worry about performance under debug, but for stuff like PutPixel, since we're calling it potentially millions of times per second, if our debug build is too slow it will become unusable for development and testing.

So the second goal of this bonus video is to optimize the debug build so that PutPixel-related operation run very fast, regardless of the fact that we are using std::vector to manage the array of pixels. The end result is that we can achieve rendering speeds which are over 100x faster than without the optimized debug configuration.

Notes

Although Chili uses std::unique_ptr<> in Surface, which allows us to = default the move members and leave the destructor undeclared, the astute student will realize that Surface should actually be setting the width and height of the donor surface to 0 when pilfering, so we actually still need to declare move members in this case. This is done in the Bonus Video where we perfect Surface.

Also note one major mistake: when we convert MemeFighter to use std::unique_ptr for owning its Weapon, we remove the destructor (because we no longer need to manually delete the heap Weapon object). However, we still need to mark the destructor as virtual so that the proper derived destructor gets called in polymorphic situations. The correct course should have been to change the destructor to virtual ~MemeFighter() = default; (a commit has been pushed after the fact that fixes this issue).

Source Code

See also