1.14.0.5
Hummingbird
A modern user interface library for games
JavaScript integration

Hummingbird interprets and executes JavaScript code that can be used for scripting the UI. The execution itself is performed by a JavaScript virtual machine. Hummingbird uses different VMs on different platforms to provide the best available option. On platforms where code JIT compilation is allowed like Windows and Android, Hummingbird uses V8, while on iOS it uses the JavaScriptCore framework.

All currently implemented JavaScript functions and objects can be viewed in the JavaScript DOM API section in the documentation.

All communication between JavaScript and the game goes through the engine module.

For ways to update the UI without writing JavaScript, take a look at the data binding feature.

There are two ways for invoking native code from JavaScript and vice-versa.

  • Events
  • Calls

Events

Events allow to call multiple handlers in both directions, but they can not return any value. To register a JavaScript handler for an event use the engine.on method.

C++ triggering JavaScript

EngineTrigger_ToUI.png

First the JavaScript code has to attach a handler for the event using engine.on:

engine.on('PlayerScoreChanged', function (new_score) {
var scoreDisplay = document.getElementById('scoreDisplay');
scoreDisplay.innerHTML = new_score;
});

Then we trigger the event in C++, using the cohtml::View::TriggerEvent method:

// set player score to 10000
view->TriggerEvent("PlayerScoreChanged", 10000);

There are also global functions in the cohtml namespace which can be used such as:

cohtml::TriggerEvent(view, "ScoreChanged", 10000);

Every TriggerEvent call starts with a C++ macro:

COHTML_TRIGGER_EVENT_SCOPE(TriggerEvent_N_Params, RuntimeName);

By default this macro does nothing and is compiled out. You can, however, add custom code there by defining COHTML_TRIGGER_EVENT_SCOPE before it's used in View.h. For example, you can add some profiling code such as:

#define COHTML_TRIGGER_EVENT_SCOPE(X, RuntimeName) \
struct GeneratedClass_##X { \
GeneratedClass_##X(const char* name) \
: EventName(name) { \
/* Start profiling */ \
} \
~GeneratedClass_##X() { \
/* End profiling */ \
} \
const char* EventName; \
} __instance(RuntimeName);
Warning
The first argument of the macro is a compile time token and can be used for creating new tokens during compilation. The second argument is a runtime string literal and is valid within the trigger event scope. If you try using it as a compile-time token you'll always get "name".
Note
In order for the binding to work you need to include the CoherentHummingbird.js library in your HTML page.

JavaScript triggering C++

EngineTrigger_ToGame.png

Registering C++ functions for events triggered by JavaScript should happen in the handler for the cohtml::IViewListener::OnReadyForBindings event.

Event Handlers

Event handlers are registered with the cohtml::View::RegisterForEvent method. They cannot return any value to JavaScript, but may trigger an event for JavaScript. There may be more than one C++ handler for a given event. There also may be both C++ and JavaScript handlers for the same event.

class Game
{
public:
void Quit()
{
}
} g_Game;
class GameIViewListener : public cohtml::IViewListener
{
public:
virtual void OnReadyForBindings()
{
m_View->RegisterForEvent("OnQuitClicked",
cohtml::MakeHandler(&g_Game, &Game::Quit));
}
};

Triggering the event from JavaScript looks like:

engine.on('OnQuitClicked', function () {
ShowMessage('Bye');
});
function OnClick() {
// this will execute both Game::Quit in C++
// and ShowMessage('Bye') in JavaScript
engine.trigger('OnQuitClicked');
});

cohtml::MakeHandler uses template argument deduction to guess the signature of the handler. This requires several specializations of cohtml::FunctorTraits and overloads of cohtml::EventHandler::InvokeStub. To extend the number of arguments for event handlers supported by Hummingbird, you have to add additional specializations and overloads.

Call Handlers

Promises are used to return results from C++ to JavaScript. Hummingbird promises are modeled after the PromisesA specification.

EngineCall.png

Call handlers are registered with the cohtml::View::BindCall method. There may be only one handler for a given call and the handler may return a result.

class Game
{
public:
std::string GetPlayerName()
{
return m_PlayerName;
}
} g_Game;
class GameIViewListener : public cohtml::IViewListener
{
public:
virtual void OnReadyForBindings()
{
m_View->BindCall("getPlayerName", cohtml::MakeHandler(&g_Game,
&Game::GetPlayerName));
}
};

To get the player name in the view is:

// call 'Game::GetPlayerName' with callback for the result
engine.call('getPlayerName').then(function (name) {
var playerName = document.getElementById('playerName');
playerName.innerHTML = name;
});
Note
For better performance it is recommended to use BindCall with engine.call when you have single handler and RegisterForEvent with engine.trigger when you have more that one handlers.

Exposing C++ Types

To be able to use your C++ types as arguments or results for event and call handlers, the C++ type must be exposed to Hummingbird. To expose a C++ type to Hummingbird use the following pattern:

class Player
{
public:
std::string Name;
double GetScore() const;
void SetScore(double);
};
// called by Hummingbird when an instance of Player goes through
// the C++ / JavaScript boundary
void CoherentBind(cohtml::Binder* binder, Player* player)
{
if (auto type = binder->RegisterType("Player", player))
{
type.Property("name", &Player::Name)
.Property("score", &Player::GetScore, &Player::SetScore)
;
}
}
Note
When exposing custom properties with the .Property syntax, a string literal must be provided. The string literal should stay "alive" at least until the OnBindingsReleased View callback is called.

The CoherentBind overload must be visible in all places where a Player instance is exposed to JavaScript, i.e. in any of the following cases:

  • a Player instance is used as an argument to TriggerEvent
  • a C++ handler takes a Player as an argument
  • a C++ handler returns a Player to JavaScript

In case the CoherentBind overload for the Player class is not visible, you will get the following compilation error:

include\cohtml\binding\binding.h(57): error C2338: cohtml::Overload_CoherentBind_For_Your_Type<T*>
1>          include\cohtml\binding\binding.h(191) : see reference to function template instantiation 'void CoherentBind<T>(cohtml::Binder *,T *)' being compiled
1>          with
1>          [
1>              T=Player
1>          ]

Going down the template instantiation stack, you can find where you are using Player without the CoherentBind overload visible.

You can define the CoherentBind overload for Player in any C++ source (.cpp) file, but you'll need to have a declaration for it, that is included everywhere Player is exposed to Hummingbird.

Depending on the structure of your project, you may consider the following patterns:

  • If Player is dedicated only for the UI - then add the declaration of CoherentBind for Player in the Player header. This way the overload will be visible everywhere Player is visible.
  • If Player is generic game type - add a PlayerBinding.h or a MySubsystemBindings.h header with the declaration of CoherentBind for Player (or for all the types in the particular game subsystem). After that make sure to include the header where Player is used with Hummingbird.
Note
If you are using namespaces, then the CoherentBind overload for Player has to be either in the namespace of Player or in the cohtml namespace. This way it will be found using argument dependent lookup.
Warning
You have to call cohtml::View::TriggerEvent with a valid instance of your class. Hummingbird uses this instance to cache the exposed properties and in certain cases might need to use this instance. For example in the case of virtual getters or setters, or virtual inheritance, the instance pointer might need to be adjusted before calling a getter or a setter for a property and getting this adjustment is using the virtual table pointer.

STL and container types

Hummingbird has built-in support for most STL containers and std:: classes.

C++ Type JavaScript Type Header
std::string String <cohtml\Binding\String.h>
std::vector Array <cohtml\Binding\Vector.h>
std::map Object <cohtml\Binding\Map.h>
std::pair Object <cohtml\Binding\Pair.h>
C style array Array <cohtml\Binding\Array.h>
std::shared_ptr Stored object's type <cohtml\Binding\SharedPointer.h>
std::unique_ptr Stored object's type <cohtml\Binding\UniquePointer.h>
raw pointer Stored object's type <cohtml\Binding\TypeTraits.h>

Stored object's type is the type of the unwrapped object. For example, the JavaScript type of std::shared_ptr<std::string*> is String.

Support for additional containers can be added in a similar way.

Support for a custom pointer type can be added in a similar way. All you have to do is to specialize PointerTrait for it. For example:

template <typename T>
struct PointerTrait<MyPointer<T>> : TrueType
{
typedef T StoredType;
static void* Deref(MyPointer<StoredType>& ptr) // Returns a pointer to the stored object.
{
return ptr.myDeref(); // In case of myDeref returns a pointer to the stored object.
}
};
template <typename T>
struct PointerTrait<MyPointer<const T>> : TrueType
{
typedef T StoredType;
static void* Deref(MyPointer<const StoredType>& ptr)
{
return const_cast<StoredType*>(ptr.myDeref());
}
};

C++ and JavaScript Communication

After registering the Player type, events can be triggered in both directions with instances of Player as an argument and call handlers can return Players as return types.

class Game
{
public:
void Start()
{
m_View->TriggerEvent("StartWithPlayer", m_Player);
}
private:
Player m_Player;
} g_Game;

Then in JavaScript, we receive the object with the specified properties. The value of each property is the same as in the moment of triggering the event. The JavaScript callback may store a reference to the object, but its properties WILL NOT be synchronized with the actual g_Game.m_Player in the game.

engine.on('StartWithPlayer', function (player) {
var playerName = document.getElementById('playerName');
playerName.innerHTML = player.name;
var scoreDisplay = document.getElementById('scoreDisplay');
scoreDisplay.innerHTML = player.score;
});

If you want to call a C++ handler with an instance of Player created in JavaScript there is one important detail - the object must have a property __Type with value Player (the same name of the type we gave to cohtml::Binder::RegisterType in CoherentBind for Player. Otherwise Hummingbird cannot treat the object as an instance of Player.

function CreatePlayer() {
var player = {
__Type: 'Player', // set the type name
name: "MyTestPlayer",
score: 0
};
engine.call('CreatePlayer', player).then(function (success) {
if (success) {
ShowMessage('Welcome ' + player.name);
} else {
ShowMessage('Sorry, try another name');
}
});
});
bool CreatePlayer(const Player& player)
{
if (player.Name.find(' ') == std::string::npos)
{
// create the player
return true;
}
else
{
// sorry, no spaces in player name
return false;
}
}

For some calls it's possible that there is no meaningful value to return. For example -

Item Player::GetItem(int slot)
{
if (HasItemAt(slot))
{
return GetItemAt(slot);
}
else
{
// notify the view that there is not item at this slot
m_View->SetScriptError(cohtml::SCE_NoResult, "no item at slot");
return Item();
}
}
engine.call('GetItem', slot).then(function (item) {
ShowMessage('Item at slot ' + slot + ' costs ' + item.Price);
},
// called when there is no item at this slot
function (errorType, message) {
console.log('could not get item at slot ' + slot);
}
});

Customizing Promises

Custom promises have been deprecated! Now we are supporting ECMAScript 6 promises

JavaScript extensions

Hummingbird provides extensions on-top of the vanilla DOM-related JavaScript as defined by the standard. The extensions are aimed to provide better performance and a more natural workflow to UI developers. The prime example of such an extension are the Node.leftXX/Node.topXX methods. In vanilla JS, to update the position of an element you have to call Node.top/Node.left, however the functions take a string parameter. To update an element's left property you have to do:

myNode.left = newPos + "px";

The newPos variable is a number in pixels, but the runtime will have to convert it to a string, append the 'px' specifier and send it to the native code. The native code at that moment will convert it back to a number to re-calculate the position. This is very inefficient, it creates JavaScript garbage and unnecessary work.

In Hummingbird you can just say:

myNode.leftPX = newPos;

No garbage is generated and the calculation is much faster both in JS and native code. A complete list of extensions is available in the JavaScript DOM API section of the documentation.

Exporting C++ objects by reference

Hummingbird supports exporting C++ objects by reference. This avoids copying the C++ values to the JavaScript heap and makes the communication between the game and the UI faster. However the user is responsible for keeping the exposed C++ object pointers valid while they are exposed to JavaScript.

Warning
Destroying or moving a C++ object exposed by reference to another location, without notifying the JavaScript universe will cause undefined behavior when the variable is used by JavaScript.

The API to expose a C++ object by reference is the same as normal values. The difference is that to expose a reference the object has to be wrapped in a cohtml::ByRef holder.

Once an object is exposed by reference to JavaScript, any subsequent TriggerEvents using ByRef with the same object will result in the same JavaScript variable send to JavaScript. This allows for 1-to-1 mapping between the C++ and JavaScript objects.

When an exposed object is about to be destroyed (or moved) its reference must be removed from JavaScript to avoid using invalid pointers (use-after-free). To notify JavaScript that a referenced object is no longer available call the cohtml::View::DestroyExposedObject method with the address of the object being destroyed. It will remove the reference from the JavaScript values and any attempt to access the properties of the object throw an exception in JavaScript.

Note
When a complex object is exposed by references, all of its subobjects are also exposed by reference. This means that the JavaScript should avoid keeping references to them.

Here is a example usage of exposing objects by reference.

struct Nameplate
{
float X;
float Y;
float Health;
};
// the same API for exposing the Nameplate struct
//
void CoherentBind(cohtml::Binder* binder, Nameplate* nameplate)
{
if (auto type = binder->RegisterType("nameplate", nameplate))
{
type.Property("x", &Nameplate::X)
.Property("y", &Nameplate::Y)
.Property("health", &Nameplate::Health)
;
}
}
class NameplateManager
{
public:
void CreateNameplates(unsigned count)
{
// Create and fill nameplates
// ...
for (auto i = 0; i < count; ++i)
{
// Send a reference to the nameplate to JavaScript
m_View->TriggerEvent("nameplates.create",
cohtml::ByRef(&m_Nameplates[i]));
}
}
void DestroyNameplates()
{
for (auto& nameplate: m_Nameplates)
{
// Notify the JavaScript universe that this reference is no
// longer valid. All accessses to the variable will throw
// exception in JavaScript.
m_View->DestroyExposedObject(&nameplate);
}
}
// The game requires to modify a nameplate, so we store it in the
// modified list in order to update the UI later
Nameplate& ModifyNameplate(unsigned id)
{
auto& nameplate = m_Nameplates[id];
m_Modified.push_back(&nameplate);
return nameplate;
}
// Update the modified nameplates since the last update
// Again expose the nameplates by reference. This way the JavaScript
// will receive the very same JavaScript object as the one created.
void Update()
{
for (auto modified : m_Modified)
{
m_View->TriggerEvent("nameplates.update",
cohml::ByRef(modified));
}
m_Modified.clear();
}
private:
cohtml::View* m_View;
// Beware that the vector may grow and relocate the nameplates.
// If this happens all the exposed references must be destroyed and
// recreated.
std::vector<Nameplate> m_Nameplates;
std::vector<Nameplate*> m_Modified;
};

To take advantage of this C++ classes the JavaScript side can do:

// this will point the exposed by C++ nameplate object
function updateNameplate() {
this.position.leftVW = this.X; // will take the current X member
this.position.topVH = this.Y; // will take the current Y member
// update the health with the current value, without fractional part
this.health.textContent = this.Health.toFixed();
}
function createNameplate(nameplate) {
// create the DOM elements for the nameplate
var plate = createNameplateDOM();
// extend the C++ nameplate with additional JavaScript values - the DOM
// elements that has to be updated when the nameplate is changed
nameplate.position = plate.style;
nameplate.health = plate.childNodes.item(1);
// extend the C++ nameplate with JavaScript function for the update
nameplate.update = updateNameplate();
nameplate.update();
document.body.appendChild(plate);
}
engine.on('nameplates.create', createNameplate);
engine.on('nameplates.update', function (nameplate) {
// This the same JavaScript value, so it already has an update method
// and references to the DOM elements
nameplate.update();
}

Calling C++ methods from JavaScript

When a C++ object is exposed by reference to JavaScript, its methods can also be called from JavaScript.

class Game
{
void Initialize(cohtml::View* view)
{
// this game object will be accessing as the global "game" variable
// in JavaScript
view->ExposeAsGlobal("game", this);
}
~Game()
{
// make sure JavaScript won't crash us
m_View->DestroyExposedObject(this);
}
bool LoadSave(const char* name);
Quit();
};
void CoherentBind(cohtml::Binder* binder, Game* game)
{
if (auto type = binder->RegisterType("Game", game))
{
type.Method("loadSave", &Game::LoadSave)
.Method("quit", &Game::Quit)
;
}
}

After this, JavaScript can use the game variable as any other object.

quitButton.addEventListener('touchstart', function() { game.quit(); }, false);
// ...
function LoadGame(name) {
if (!game.loadSave(name)) { // calls the C++ method
ShowError();
}
}