by Elliot Woods
C++ an old language. It has been around since 1983. It is often seen as archaic, obtuse, and perhaps just plain out of touch by today's standards. Contrary to this, many people believe it offers the best balance of performance and clarity, and many other people use it fot its strong ecosystem of 3rd party libraries, device support and general acceptance. Even new technologies (like CUDA) often use C++ as their language of choice.
Some more modern languages (such as JavaScript and C#) make programs which run in a very different way to C/C++. They use the concept of a 'virtual machine', which is a very different type of computer than the one which crunches electronic signals on a circuit board. The virtual machine receives and processes program instructions like a real machine, but allows for all sorts of ideas which don't directly translate to silicon electronics, such as dynamic typing and reflection. The virtual machine abstracts the constraints of the processor away from the thinking of the programmer.
C/C++ does not have a virtual machine, which (for the time being) often gives it a performance edge over these newer languages. It is quite strict in that the instructions you type deep down have a 1:1 translation with physical processor instructions. This gives the language some inflexibilities, but means that code can be both understood naturally by a human, and clearly translate to machine code.
In this chapter we'll look at some of the new patterns in the C++ language introduced in C++11, which retain this promise whilst offering new ways of writing code.
auto
Perhaps the most used, and simplest new pattern in C++11 is auto
. You'll love it. And probably won't remember life without it after a day or two. Consider the following...
ofRectangle myRectangle = ofGetCurrentViewport();
ofVec2f rectangleCenter = myRectangle.getCenter();
float rectangleCenterX = rectangleCenter.x;
In this code block, we are declaring 3 variables:
myRectangle
rectangleCenter
rectangleCenterX
On each line of code we are:
ofRectangle
, ofVec2f
, float
respectively)The type of data on the right and left side of the =
is the same. Since C++ is strictly typed (e.g. a function which returns float
will always return float
), it is impossible for the value on the right hand side to ever be anything different. The compiler knows what data type the right hand side of the code will give, e.g. it knows that on line 1 that on the right hand side of the =
is an ofRectangle
. So perhaps if we were to write something like:
auto myRectangle = ofGetCurrentViewport();
auto rectangleCenter = myRectangle.getCenter();
auto rectangleCenterX = rectangleCenter.x;
Then the compiler can do some of the coding for us. In fact, this is exactly the facility of the auto
feature. This code block compiles to exactly the same result as the first code block did. The compiler notices what's on the right hand side and substitutes in the correct type wherever it sees auto
.
auto is going to save you keystrokes. Imagine the following:
//.h
vector<shared_ptr<ofThread> > myVectorOfThreads;
//.cpp
// the old way
vector<shared_ptr<ofThread>>::iterator firstThreadIterator = this->myVectorOfThreads.begin();
// the new way
auto firstThreadIterator = this->myVectorOfThreads.begin();
This makes the code more readable, but also you could take advantage of auto
in other ways.
For example, when you reuse or copy variables, you can use auto along the way, meaning that you only have to define which type you want to use at the beginning and not everywhere else in your code. For example in the above h file, you might change the vector
to a list
, or change shared_ptr<ofThread>
to ofThread *
. These changes would perpetuate automatically to wherever an auto
is being used in the code. Nifty huh?
auto
is not a new typeNote that the following doesn't work:
auto myRectangle = ofGetCurrentViewport();
myRectangle = "look mum! i'm a string!!"; // compiler error!
Remember that auto
isn't a type itself, it's not a magic container that can take any kind of thing (such as var
in C# or another dynamic typed language), it is simply a keyword which gets substituted at compile time, you can imagine that the compiler just writes in for you whatever makes sense on that line where it is. In this case, the first line makes sure that myRectangle
is an ofRectangle
, and you can't assign a string
to an ofRectangle
.
auto
in function argumentsSince the auto
must be implicitly defined by the line of code where it is used, and that this decision is made at compile time, it can not be a function argument, let's see what that means..
Imagine that the following was valid: (NOTE : it isn't!)
float multiplyBy2(auto number) {
return number * 2;
}
int firstNumber = 1;
float secondNumber = 2.0f;
cout << multiplyBy2(firstNumber) << endl;
cout << multiplyBy2(secondNumber) << endl;
Now if this code were valid, then the first time the function is called, the auto
would mean int
, and the second time it would mean float
. Therefore saying that the text auto
is simply substituted with an explicit type where it is written doesn't make sense. So basically you can't use auto
in function arguments (you might want to look into template
instead, which would automatically generate 2 pieces of code for the 2 different types).
auto
as a function return typeI'm not sure why, you just can't. It kinda makes sense that you should be able to, but you just can't, I don't make the rules...
const
and referencesLet's do a const auto
:
ofRectangle rectangle = ofGetCurrentViewport();
//is the same as
auto rectangle = ofGetCurrentViewport();
//and
const ofRectangle rectangle = ofGetCurrentViewport();
//is the same as
const auto rectangle = ofGetCurrentViewport();
Next look at auto &
for when we want reference types.
float x = rectangle.x;
//is the same as
auto x = rectangle.x;
//and
float & x = rectangle.x;
//is the same as
auto & x = rectangle.x;
auto
in variable declarationsauto
takes the type of the right hand side of the =
assignment, and replaces the text auto
with that type at compile time.auto
is not a magic container which can carry any type of data, it simply gets replaced at compile time by whatever is implied in the codeauto
in function arguments, return types and a few other places.const auto
for a const type, and auto &
for a reference typefor (thing : things)
Consider the following common pattern:
vector<ofPixels> mySelfies;
/*
take some snaps here and store in mySelfies
*/
//oh dear, my photos are all in portrait
for(int i=0; i<mySelfies.size(); i++) {
//rotate them from landscape to portrait
mySelfies[i].rotate90(1);
}
vector<ofxMySpace::Friend> myFriends = myspaceClient.getFriends();
//now let's send them to all my friends
for(int i=0; i<myFriends.size(); i++) {
if (myFriends[i].isOnline()) {
for(int i=0; i<mySelfies.size(); i++) {
myFriends[i].sendImage(mySelfies[i]);
}
}
}
Well we're forever doing things like for(int i=0; i<mySelfies.size(); i++) {
, so let's see if we can find a neater way of doing this with for thing in things
which in C++11 we can write as for (thing : things)
...
vector<ofPixels> mySelfies;
/*
take some lovely snaps
*/
//oh dear, my photos are all in portrait
for(auto & mySelfie : mySelfies) {
//rotate them from landscape to portrait
mySelfie.rotate90(1); // (1)
}
auto myFriends = myspaceClient.getFriends();
//now let's send them to all my friends
for(auto & myFriend : myFriends) {
if (myFriend.isOnline()) {
for(auto & mySelfie : mySelfies) {
myFriend.sendImage(mySelfie); // (2)
}
}
}
Notice that for(thing : things)
gels so well with auto
. Also notice that I'm using auto &
since:
(1)
I want to be able to change the contents of the vector, so I need a reference to the vector item rather than a copy of it.(2)
, it makes more sense to use a reference rather than a copy because it's computationally cheaper, and means I don't have to allocate new instances of ofxMySpace::Friend
(which perhaps has many complicated allocations internally which you wouldn't want to repeat by copying the object).for(auto thing : vectorOfThings)
vector
, map
, list
, deque
, set
, etc) and in some more purpose built containers (e.g. ofxGrayCode::DataSet
)for(auto & thing : vectorOfThings)
to use a reference rather than a copyoverride
saves you time not by reducing the amount of typing you do, but by reducing the amount of head-scratching you might do when dealing with virtual
functions. Imagine the following:
class BuildingProjectionMapper {
public:
//...
virtual void mapTheNeoClassicalColumns();
//...
};
class AutoBuildingProjectionMapper : public BuildingProjectionMapper {
public:
void mapTheNeoClassicalColums(); // woops, I spelt column incorrectly
};
Now if I implement AutoBuildingProjectionMapper::mapTheNeoClassicalColums
, it may never get called, and I may be wondering why my function calls are all being handled by the base class. The problem is that the compiler never told me that the function that I was trying to override didn't exist. Here comes override
to the rescue.
class AutoBuildingProjectionMapper : public BuildingProjectionMapper {
public:
void mapTheNeoClassicalColums() override;
};
This tells the compiler that I'm intending to override a virtual
function. In this case, the compiler will tell me that no virtual
function called mapTheNeoClassicalColums
exists, and that therefore my override
is faulty. So following the compiler's complaint I can go in and fix the spelling mistake. Then I can get on with making my Projection Mappening on the town library facade.
This is especially useful when you make changes to your base class, e.g. in oF 0.9.0, many base classes have changed, and by using override
in addons, addon authors can be hard-notified that their code is now invalid and needs updating.
override
at the end of function definitions in a derived class' h file when you are intending to override a virtual
function in the base class
override
is invalid, which might just save you a lot of time hunting for errorsThis one is a little more complex. Rather than making things quicker or safer to write, it adds a new feature to the way you can design your code. As such, we'll introduce the basics of what they do, then make some interesting examples.
Lambda functions are very common in JavaScript, C# and other modern languages, and can be very liberating to use. They can seem more scary than they really are, so I suggest just getting on with some examples to start with. Let's make a simple example:
void normalFunction() {
cout << "Heathcliff, it's me Cathy." << endl;
}
void main() {
auto lambdaFunction = []() {
cout << "So cold. Let me into your window." << endl;
};
normalFunction();
lambdaFunction();
}
Let's take a guess what happens here:
Heathcliff, it's me Cathy.
So cold. Let me into your window.
So the lambda function is just like a normal function, but the biggest differences are that:
lambdaFunction
is defined inside main
).That second one is where the lambda function really comes into its own. Because we can pass it around, store it, throw it away, copy it. Let's make a more concrete example of this.
When we want to pass a lambda function as a variable, we need to introduce a type which can hold that data (since we can't use auto
in function arguments or when defining members of classes). A common type which can store a lambda function is std::function<void()>
. I understand that void()
looks a bit funny at first, but it will make sense when we do a bit more with lambdas later on.
class Graphic {
public:
void draw() {
whatToDraw();
}
std::function<void()> whatToDraw;
};
class ofApp : ofBaseApp {
/*
.. all the usual stuff
*/
protected:
vector<Graphic> graphicElements;
};
void ofApp::setup() {
Graphic circle;
circle.whatToDraw = []() {
ofDrawCircle(50,50,20);
};
this->graphicElements.push_back(circle);
Graphic rectangle;
rectangle.whatToDraw = []() {
ofRect(50,50,200,200);
};
this->graphicElements.push_back(rectangle);
}
void ofApp::draw() {
for(auto & graphic : this->graphicElements) {
graphic.draw();
}
}
We could do this by creating classes which inherit from Graphic
and override the draw
function (e.g. CircleGraphic
and RectangleGraphic
). But instead of defining new classes, we've created 2 instances of Graphic which contain different function definitions by using lambda functions. This is convenient as it's often a pain to make new class definitions when we just want to change a little bit of functionality of a class.
Now how about getting variables in and out of the function? A common pattern with normal functions is using function arguments to get variables into the function, and returning variables to get values out again. Let's see how that works with lambdas.
float multiplyByTwoFunction(float x) {
return x * 2.0f;
}
void test() {
auto multiplyByTwoLambda = [](float x) {
return x * 2.0f;
};
cout << multiplyByTwoFunction(2.0f) << endl;
cout << multiplyByTwoLambda(2.0f) << endl;
}
This would of course print out:
4
4
That's great! But we also have to be careful that because this function now has a float return type and a float argument, it is a different type of function to our earlier lambdaFunction
which took no arguments and returned a string. Note that if we want to pass this function as an argument we'd need to do like:
void performActionOnFloat(float & x, std::function<float(float)> action) {
x = action(x);
}
Specifically, the type which can hold a lambda function is:
std::function<return type(arguments)>
You can use multiple arguments (like you can with normal functions).
Finally, we often need to introduce a 'scope' to the lambda function. The scope means what variables from outside the function can be seen inside the function. We define the scope in-between the []
s in the lambda function definition, for example:
float multiplier = 4.0f;
auto multiplyByOurMultiplier = [multiplier](float x) { // (1)
return x * multiplier;
}
cout << multiplyByOurMultiplier(2.0f); // prints 8
By putting the variable name multiplier
within the []
s of the lambda function definition at (1)
, we now can use the variable multiplier
within the function definition. A really handy thing to put there is [this]
which puts this
within the scope of the function, i.e. makes all the members of the current class available from within the function.
You can: * Add multiple variables to the scope, e.g. [this, multiplier]
* Add variables as references (rather than copies) to the scope, e.g.: [&multiplier]
* Automatically add any variables to the scope which you use inside the function using [=]
(make a copy of the variable at the function definition) or [&]
(use a reference to the variable).
vector
s, etc. Basically, anything you can do with a variable you can do with a lambda function.[=]
, [&]
, [this]
, [myVariableName]
, [&myVariableName]
, etc.Now let's go for a more down-deep example of how you might use lambda functions to save you from getting bogged down. In this example we have a class which talks to a camera device. The particular thing about this camera device is that the camera must run in its own thread, and all operations on the camera must run in that thread.
Imagine our ofApp looks like:
class ofApp : public ofBaseApp {
/*
the usual stuff
*/
CameraClass camera;
ofParameter<float> exposure;
}
class CameraClass : public ofThread {
protected:
void threadedFunction() override {
while (this->isCameraRunning()) {
this->device.getFrame();
this->pixelsLock.lock();
this->pixels.setFromExternalPixels(this->device->getData(), 640, 480, 3);
this->pixelsLock.unlock();
// (1)
}
}
CameraDriver device;
ofPixels pixels;
ofMutex pixelsLock;
}
Now if we want a way to set the exposure on the camera from our ofApp
we need to signal this to the CameraClass
, and have it set the exposure in the right thread (e.g. at (1) which happens between grabbing frames from the camera).
One way of doing this would be to store some data saying that we needed to change the exposure, and what that new exposure value is. Then at (1) we would need to check for that data and have the code to act on it appropriately. The problem with that is that it's really laborious and introduces lots of special cases (e.g. what if something had to read the gain and then change the exposure accordingly?).
A nice workaround is to use lambda functions for this purpose, e.g. :
class CameraClass : public ofThread {
protected:
void threadedFunction() override {
while (this->isCameraRunning()) {
this->device.getFrame();
this->pixelsLock.lock();
this->pixels.setFromExternalPixels(this->device->getData(), 640, 480, 3);
this->pixelsLock.unlock();
this->actionsLock.lock();
for(auto & action : this->actions) {
action(this->device);
}
this->actions.clear();
this->actionsLock.unlock();
}
}
void performInThread(std::function<void(CameraDriver&)> action) {
this->actionsLock.lock();
this->actions.push_back(action);
this->actionsLock.unlock();
}
CameraDriver device;
ofPixels pixels;
ofMutex pixelsLock;
vector<std::function<void(CameraDriver&)> > actions;
ofMutex actionsLock;
}
//somewhere in ofApp
this->camera->performInThread([this](CameraDriver & device) {
device.setExposure(this->exposure);
});
This will then push the instruction into CameraClass, so that at the right time this function is called within the thread managing the camera, taking the this->exposure
variable from the ofApp
and acting on the CameraDriver
defined in CameraClass
. All this happens without having to pass lots of variables around, in a really flexible way. To see how this is implemented in ofxMachineVision, check out the ofxMachineVision::Utils::ActionQueueThread
at https://github.com/elliotwoods/ofxMachineVision/blob/master/src/ofxMachineVision/Utils/ActionQueueThread.cpp .