1
0
mirror of https://github.com/wqking/eventpp.git synced 2024-12-27 00:17:02 +08:00
eventpp/doc/eventqueue.md
2024-03-13 08:37:10 +08:00

270 lines
12 KiB
Markdown

# Class EventQueue reference
<!--begintoc-->
## Table Of Contents
* [Description](#a2_1)
* [API reference](#a2_2)
* [Header](#a3_1)
* [Template parameters](#a3_2)
* [Public types](#a3_3)
* [Member functions](#a3_4)
* [Inner class EventQueue::DisableQueueNotify](#a3_5)
* [Internal data structure](#a2_3)
<!--endtoc-->
<a id="a2_1"></a>
## Description
EventQueue includes all features of EventDispatcher and adds event queue features. Note: EventQueue doesn't inherit from EventDispatcher, don't try to cast EventQueue to EventDispatcher.
EventQueue is asynchronous. Events are cached in the queue when `EventQueue::enqueue` is called, and dispatched later when `EventQueue::process` is called.
EventQueue is equivalent to the event system (QEvent) in Qt, or the message processing in Windows API.
For basic functions such as add/remove listeners, please refer to the [EventDispatcher document](eventdispatcher.md).
<a id="a2_2"></a>
## API reference
<a id="a3_1"></a>
### Header
eventpp/eventqueue.h
<a id="a3_2"></a>
### Template parameters
```c++
template <
typename Event,
typename Prototype,
typename Policies = DefaultPolicies
>
class EventQueue;
```
EventQueue has the exactly same template parameters with EventDispatcher. Please reference [EventDispatcher document](eventdispatcher.md) for details.
<a id="a3_3"></a>
### Public types
`QueuedEvent`: the data type of event stored in the queue. It's declaration in pseudo code is,
```c++
struct EventQueue::QueuedEvent
{
EventType event;
std::tuple<ArgTypes...> arguments;
// get the event
EventType getEvent() const;
// get the argument of index N
// same as std::get<N>(queuedEvent.arguments)
template <std::size_t N>
NthArgType getArgument() const;
};
```
`event` is the EventQueue::Event, `arguments` are the arguments passed in `enqueue`.
<a id="a3_4"></a>
### Member functions
#### constructors
```c++
EventQueue();
EventQueue(const EventQueue & other);
EventQueue(EventQueue && other) noexcept;
EventQueue & operator = (const EventQueue & other);
EventQueue & operator = (EventQueue && other) noexcept;
```
EventQueue can be copied, moved, assigned, and move assigned.
Note: the queued events are not copied, moved, assigned, or move assigned, only the listeners are performed with these operations. That's to say, the queued events are not duplicated when an EventQueue is copied or assigned.
#### enqueue
```c++
template <typename ...A>
void enqueue(A && ...args);
template <typename T, typename ...A>
void enqueue(T && first, A && ...args);
```
Put an event into the event queue. The event type is deducted from the arguments of `enqueue`.
All copyable arguments are copied to internal data structure. All non-copyable but movable arguments are moved.
EventQueue requires the arguments either copyable or movable.
If an argument is a reference to a base class and a derived object is passed in, only the base object will be stored and the derived object is lost. Usually shared pointer should be used in such situation.
If an argument is a pointer, only the pointer will be stored. The object it points to must be available until the event is processed.
`enqueue` wakes up any threads that are blocked by `wait` or `waitFor`.
The time complexity is O(1).
The two overloaded functions have similar but slightly difference. How to use them depends on the `ArgumentPassingMode` policy. Please reference the [document of policies](policies.md) for more information.
Note: the arguments life time may be longer than expected. `EventQueue` copies the arguments into internal data structure, after the event is dispatched, the data is cached for next usage, so the arguments won't be destroyed until the data is reused. This is for performance optimization. This is usually not an issue, but if you pass large data in shared pointer, the data may be in the memory for longer time than necessary.
#### process
```c++
bool process();
```
Process the event queue. All events in the event queue are dispatched once and then removed from the queue.
The function returns true if any events were processed, false if no event was processed.
The listeners are called in the thread same as the caller of `process`.
Any new events added to the queue during `process()` are not dispatched during current `process()`.
`process()` is efficient in single thread event processing, it processes all events in the queue in current thread. To process events from multiple threads efficiently, use `processOne()`.
Note: if `process()` is called from multiple threads simultaneously, the events in the event queue are guaranteed dispatched only once.
#### processOne
```c++
bool processOne();
```
Process one event in the event queue. The first event in the event queue is dispatched once and then removed from the queue.
The function returns true if one event was processed, false if no event was processed.
The listener is called in the thread same as the caller of `processOne`.
Any new events added to the queue during `processOne()` are not dispatched during current `processOne()`.
If there are multiple threads processing events, `processOne()` is more efficient than `process()` because it can split the events processing to different threads. However, if there is only one thread processing events, 'process()' is more efficient.
Note: if `processOne()` is called from multiple threads simultaneously, the events in the event queue are guaranteed dispatched only once.
#### processIf
```c++
template <typename Predictor>
bool processIf(Predictor && predictor);
```
Process the event queue. Before processing an event, the event is passed to `predictor` and the event will be processed only if `predictor` returns true. If `predictor` returns false, the event will not be processed and be kept in the queue, then `processIf` will continue processing next event in the queue.
`predictor` is a callable object(function, lambda, etc) that takes exactly the same arguments as `EventQueue::enqueue` or have no arguments, and returns a boolean value. eventpp will pass the arguments properly.
`processIf` returns true if any event was dispatched, false if no event was dispatched.
`processIf` has some good use scenarios:
1. Process certain events in certain thread. For example, in a GUI application, the UI related events may be only desired to be processed in the main thread. In such case, `predictor` may return true for any UI events, and return false for any non-UI events.
```c++
template <typename Predictor>
bool processUntil(Predictor && predictor);
```
Process the event queue. Before processing an event, the event is passed to `predictor`. If `predictor` returns true, `processUntil` stops any further processing and returns. If `predictor` returns false, `processUntil` will process the underlying events.
`predictor` is a callable object(function, lambda, etc) that takes exactly the same arguments as `EventQueue::enqueue` or have no arguments, and returns a boolean value. eventpp will pass the arguments properly.
`processUntil` returns true if any event was dispatched, false if no event was dispatched.
`processUntil` has a good use case that limits the process time to simulate "timeout". For example, in a game engine, the event process may be limited to only several milliseconds, the remaining events will be processed in next game loop. In such situation, the `predictor` can return true when time out.
#### emptyQueue
```c++
bool emptyQueue() const;
```
Return true if there is no any event in the event queue, false if there are any events in the event queue.
Note: in multiple threading environment, the empty state may change immediately after the function returns.
Note: don't write loop as `while(! eventQueue.emptyQueue()) {}`. It's dead loop since the compiler will inline the code and the change of empty state is never seen by the loop. The safe approach is `while(eventQueue.waitFor(std::chrono::nanoseconds(0))) ;`.
#### clearEvents
```c++
void clearEvents();
```
Clear all queued events without dispatching them.
This is useful to clear any references such as shared pointer in the queued events to avoid cyclic reference.
#### wait
```c++
void wait() const;
```
`wait` causes the current thread to block until the queue is not empty.
Note: though `wait` has work around with spurious wakeup internally, the queue is not guaranteed not empty after `wait` returns.
`wait` is useful when a thread processes the event queue. A sample usage is,
```c++
for(;;) {
eventQueue.wait();
eventQueue.process();
}
```
The code works even if it doesn't `wait`, but doing that will waste CPU power resource.
#### waitFor
```c++
template <class Rep, class Period>
bool waitFor(const std::chrono::duration<Rep, Period> & duration) const;
```
Wait for no longer than *duration* time out.
Return true if the queue is not empty, false if the return is caused by time out.
`waitFor` is useful when a event queue processing thread has other condition to check. For example,
```c++
std::atomic<bool> shouldStop(false);
for(;;) {
while(! eventQueue.waitFor(std::chrono::milliseconds(10)) && ! shouldStop.load()) ;
if(shouldStop.load()) {
break;
}
eventQueue.process();
}
```
#### peekEvent
```c++
bool peekEvent(EventQueue::QueuedEvent * queuedEvent);
```
Retrieve an event from the queue. The event is returned in `queuedEvent`.
```c++
struct EventQueue::QueuedEvent
{
TheEventType event;
std::tuple<ArgumentTypes...> arguments;
};
```
`queuedEvent` is a EventQueue::QueuedEvent struct. `event` is the EventQueue::Event, `arguments` are the arguments passed in `enqueue`.
If the queue is empty, the function returns false, otherwise true if an event is retrieved successfully.
After the function returns, the original even is still in the queue.
Note: `peekEvent` doesn't work with any non-copyable event arguments. If `peekEvent` is called when any arguments are non-copyable, compile fails.
#### takeEvent
```c++
bool takeEvent(EventQueue::QueuedEvent * queuedEvent);
```
Take an event from the queue and remove the original event from the queue. The event is returned in `queuedEvent`.
If the queue is empty, the function returns false, otherwise true if an event is retrieved successfully.
After the function returns, the original even is removed from the queue.
Note: `takeEvent` works with non-copyable event arguments.
#### dispatch
```c++
void dispatch(const QueuedEvent & queuedEvent);
```
Dispatch an event which was returned by `peekEvent` or `takeEvent`.
<a id="a3_5"></a>
### Inner class EventQueue::DisableQueueNotify
`EventQueue::DisableQueueNotify` is a RAII class that temporarily prevents the event queue from waking up any waiting threads. When any `DisableQueueNotify` object exist, calling `enqueue` doesn't wake up any threads that are blocked by `wait`. When the `DisableQueueNotify` object is out of scope, the waking up is resumed. If there are more than one `DisableQueueNotify` objects, the waking up is only resumed after all `DisableQueueNotify` objects are destroyed.
`DisableQueueNotify` is useful to improve performance when batching adding events to the queue. For example, in a main loop of a game engine, `DisableQueueNotify` can be created on the start in a frame, then the game adding events to the queue, and the `DisableQueueNotify` is destroyed at the end of a frame and the events are processed.
To use `DisableQueueNotify`, construct it with a pointer to event queue.
Sample code
```c++
using EQ = eventpp::EventQueue<int, void ()>;
EQ queue;
{
EQ::DisableQueueNotify disableNotify(&queue);
// any blocking threads will not be waken up by the below two lines.
queue.enqueue(1);
queue.enqueue(2);
}
// any blocking threads are waken up here immediately.
// any blocking threads will be waken up by below line since there is no DisableQueueNotify.
queue.enqueue(3);
```
<a id="a2_3"></a>
## Internal data structure
EventQueue uses three `std::list` to manage the event queue.
The first busy list holds all nodes of queued events.
The second idle list holds all idle nodes. After an event is dispatched and removed from the queue, instead of freeing the memory, EventQueue moves the unused node to the idle list. This can improve performance and avoid memory fragment.
The third list is a local temporary list used in function `process()`. During processing, the busy list is swapped to the temporary list, all events are dispatched from the temporary list, then the temporary list is returned and appended to the idle list.