I am entering a realm that is new to me, but basically I need to implement callbacks in C++. I am designing a toolkit for myself to use to simplify my life. Basically it is a .dll plugin that will be exposing a lot of functions to my other .dll plugins.
One of these functions is HookEvent(const char *event_name, void *callback) which will allow me to hook different events that get fired. Here would be an example...
Example_Plugin1.dll does HookEvent("player_spawn", &Plugin1::Event_PlayerSpawn);
Example_Plugin2.dll does HookEvent("player_spawn", &Plugin2::Event_PlayerSpawn);
I need to figure out the best (and preferably easiest) method of setting up a callbacks system that will work well for this. I have been reading up on C++ callbacks for a few hours now, and found quite a few different approaches.
I assume the easiest thing to do would be make a template, and use typedef bool (ClassName::*EventHookCallback)(IGameEvent, bool); After that, I am a bit foggy.
I also read that Delegates or a .NET style events system are other possible approaches. I am already somewhat confused, so I don't want to confuse myself more, but figured it was worth asking.
Here is a link to the C++ .NET style events system I was reading about.
http://cratonica.wordpress.com/2010/02/19/implementing-c-net-events-in-c/
So what do you guys suggest? Any tips as far as implementing it would be most appreciated.
If you want generalized event firing Boost.Signals2 might be applicable.
The Boost.Signals2 library is an
implementation of a managed signals
and slots system. Signals represent
callbacks with multiple targets, and
are also called publishers or events
in similar systems. Signals are
connected to some set of slots, which
are callback receivers (also called
event targets or subscribers), which
are called when the signal is
"emitted."
Even if you don't need this level of flexibility you should be able to simplify the function binding in your code using Boost.Bind, or the C++0x equivalents.
EDIT:
There's an excellent discussion from Herb Sutter of the issues you could face here. You could use this for guidance if you decide you don't need the full Boost feature set, and so roll your own.
How about using Qt Signal and Slot? It does what callbacks do but without the messiness of making anything not part of your callback parameters global.
Boost.Signals would be my choice, combined with things like boost::bind and Boost.Function.
I would use an abstract base class as a plugin interface. (And in fact, I have used a pattern like the one below before.)
Library, PluginIfc.h:
class PluginIfc {
public:
virtual ~PluginIfc() = 0;
virtual bool EventCallback(const char* event_name, IGameEvent, bool) = 0;
};
// For Windows, add dllexport/dllimport magic to this declaration.
// This is the only symbol you will look up from the plugin and invoke.
extern "C" PluginIfc* GetPlugin();
Plugin:
#include <PluginIfc.h>
class Plugin1 : public PluginIfc {
public:
virtual bool EventCallback(const char* event_name, IGameEvent, bool);
Plugin1& get() { return the_plugin_obj; }
bool Event_PlayerSpawn(IGameEvent, bool);
// ...
private:
std::vector<std::string> _some_member;
static Plugin1 the_plugin_obj; // constructed when plugin loaded
};
Plugin1 Plugin1::the_plugin_obj;
PluginIfc* GetPlugin() { return &Plugin1::get(); }
This way, your plugin classes can easily have members, and C++'s virtual call mechanism takes care of giving you a good this pointer in EventCallback.
It may be tempting to make a virtual method per event type, say just make Event_PlayerSpawn and similar methods virtual. But then whenever you want to add an event type, if this means changing class PluginIfc, your old compiled plugins are no longer compatible. So it's safer to use a string event identifier (for extensibility) and have the main callback sort events off to more specific methods.
The major drawback here (as compared to a signal-slot type implementation) is that all callbacks must take the same set of arguments. But your question sounded like that would be adequate. And it's often possible to work within that limitation by making sure the set of arguments is very flexible, using strings to be parsed or Any-style objects.
Sounds like you might be interested in how to build your own plugin framework. The problems you'll encounter are likely the same. Have a look at this nice Dr Dobbs article Building Your Own Plugin Framework.
Hope this helps!
Implementing your own callback system is non-trivial.
My understanding is that your aim is to map event types to specific callback functions.
E.g. if "player_spawn" event is risen the &Plugin1::Event_PlayerSpawn will be called.
So what you should do is the following:
1) Define all the events of interest. Make them as generic as possible. They can
encapsulate any information you need
2) Create a Registrar. I.e. a class that all modules register their interest for specific
methods. E.g. Registrar.register(player_spawn,this,Event_PlayerSpawn);
3) Registrar has a queue of all subscribers.
4) You can also have a uniform interface for the modules. I.e. all module implement a specific function but based on event's data can do different things
5) When an event occurs, all the subscribers interested for the specific event get notified by calling the appropriate function
6)Subscriber can de-register when ever is need
Hope this helps.
Related
I'm working on a piece of software that is constructed from a series of "modules". Modules can be connected together to form the full application (one module might go to another, sort of an implied state machine). Each module can render to the screen, get updates and access state from other modules. Note that the modules are still within the same process, so no IPC needs to be designed into this.
However, these modules do not directly depend on each other. There is a singleton object that has the sole purpose of managing message passing between the modules. When you want to register for an event from any module:
CPostMaster::Instance().RegisterEvent("TheEventName", [](std::string const& data) { /* the callback */ });
The data variable is serialized data. Can be anything, but usually is XML or JSON. To send the event you do:
std::string serialized_data = /* serialized data, do this before calling */;
CPostMaster::Instance().SendEvent("TheEventName", serialized_data);
The 2nd parameter is optional.
Having a "master authority" for message passing has a drawback: The events themselves can't send varying parameters without utilizing some sort of serialization or type erasure (removes type safety from the picture and impacts performance).
But it also has the benefit of strict/strong coupling not being required, which means that at any given time a different module can be responsible for sending a specific event without the receiving modules having to change.
The alternative seems to be not using a singleton, and instead each module receives an object that it can use to subscribe to. This could get messy especially when you are passing them around everywhere, it will quickly mean that functions start taking boilerplate parameters.
What is a good design for message passing in a system such as this? How can it be improved and be made manageable? Type safety and open/close principles are important here. I think it's OK to have direct dependencies across modules so long as they can be mocked (for unit testing) and easily swapped out should modules change without severely impacting the whole system (but this is part of the open/close principle).
First: I dislike singletons. The only singleton I accept is a singleton manager (some sort of central instance distributor) that handles a defined init and deinit of all "singletons" in a defined order.
But back to your problem:
Your title already has the solution: Define a message interface. If you want type-safety define an IMessage with common attributes.
Then define specializations of IMessage which then are consumed by your callbacks.
The tricky part is: You will need RTTI for that, which is odd in c++, I know but might be worth the benefits, if you are restricted to gcc or visual studio, you could make use of those types, or implement some simple RTTI in the IMessage itself to avoid dynamic_cast.
To avoid boilerplate code in a callback which checks and casts around the IMessage I would provide a utility function (pseudo code, adjust for pointers, references, smart ptrs, const correctness etc.)
T SafeCast<T>(IMessage message);
depending on the implementation of your compiler you should add restrictions to T to be of a sub type of IMessage and what should happen when the cast fails (exception, nullptr, etc).
Alternatively: Check how others have solved this (maybe Qt's Signals&Slots or something in Boost)
I would make the sub modules dependent on a parent class (in your case the singleton). Then you could pass this object's reference along the line, to be used in the modules.
Module(Handler& h) : _h(h) { }
void do_stuff(){
_h.RegisterEvent("TheEventName", [](std::string const& data)
{ /* the callback */ })
Then I would register your Module class itself, or another class, as an Event, and on the Handler side, I would formalize the messaging in a way that you'd get multiple callbacks instead of just one. You'd have to formalize your message though, but you'd have type safety instead of passing strings.
For example the handler, while parsing a message, he'd call:
_callback.start(); //signals the start of a message
_callback.signalParam1(1); //calls Module.signalParam(int);
_callback.signalParam2("test"); //calls Module.signalParam2(const char*);
_callback.end();
Your Module would need to implement those.
After programming for sometime with the iOS and Mac objective C frameworks, I have come to love the generic notification pattern implemented by the NSNotificationCenter and NSNotification classes. Coming back to C++, which has always been my language of choice for most things, I find myself trying to replicate this pattern and believe there should really already be a generic implementation of similar C++ classes offering support for it out there.
It does seem like the pattern is somewhat more difficult to implement in C++ than Objective C because of the more dynamic nature of the later, but it seems far from impossible. I've looked through the boost libraries as they are generally awesome and was sad not to find my luck there. Although the boost::bind, boost::lamda, boost::function seem like they do most of the work. Have I missed something obvious? Is there anything already existing out there that would allow me to easily replicate NSNotification/NSNotificationCenter behaviour?
In theory you could create a class that has a vector of function pointers to call when a certain notification is called - A class that has a dictionary where the objects are the vectors of functions to call when a notification is pushed
In addition to the boost packages mentioned in other answers, another option is poco::NotificationCenter.
This implementation is closer to the Cocoa notification framework, as specifically discussed on Poco's documentation:
The NotificationCenter class is basically a C++ implementation of the
NSNotificationCenter class found in Apple's Cocoa (or OpenStep).
Following #anno's recommendation to look at boot::signal, it does after examination seem like a possible option although it is, as expected, not as straight-forward as the objective C solutions. Looking through the boost::signal tutorial, I thought I would go through the most relevant aspects for the problem at hand.
To create notification senders:
Consider a simple news delivery service, where clients connect to a news provider that then sends news to all connected clients as information arrives. The news delivery service may be constructed like this:
class NewsItem { /* ... */ };
boost::signal<void (const NewsItem&)> deliverNews;
The objective of deliverNews is to inform observers that a NewsItem has been generated.
Observers can be added as follows (using the boost::bind library):
Clients that wish to receive news updates need only connect a function object that can receive news items to the deliverNews signal. For instance, we may have a special message area in our application specifically for news, e.g.,:
struct NewsMessageArea : public MessageArea
{
public:
// ...
void displayNews(const NewsItem& news) const
{
messageText = news.text();
update();
}
};
// ...
NewsMessageArea newsMessageArea = new NewsMessageArea(/* ... */);
// ...
deliverNews.connect(boost::bind(&NewsMessageArea::displayNews, newsMessageArea, _1));
To address the problem of removing observers which have been deallocated from the list, boost::signal offers the following solution
However, what if the user closes the news message area, destroying the
newsMessageArea object that deliverNews knows about? Most likely, a
segmentation fault will occur. However, with Boost.Signals one need
only make NewsMessageArea trackable, and the slot involving
newsMessageArea will be disconnected when newsMessageArea is
destroyed. The NewsMessageArea class is made trackable by deriving
publicly from the boost::signals::trackable class, e.g.:
struct NewsMessageArea : public MessageArea, public boost::signals::trackable
{
// ...
};
At this time there is a significant limitation to the use of trackable
objects in making slot connections: function objects built using
Boost.Bind are understood, such that pointers or references to
trackable objects passed to boost::bind will be found and tracked.
I'm looking for a solution for this problem in C or C++.
edit: To clarify. This is on a linux system. Linux-specific solutions are absolutely fine. Cross-plaform is not a concern.
I have a service that runs in its own thread. This service is a class with several methods, some of which need to run in the own service's thread rather than in the caller's thread.
Currently I'm using wrapper methods that create a structure with input and output parameters, insert the structure on a queue and either return (if a "command" is asynchronous) or wait for its execution (if a "command" is synchronous).
On the thread side, the service wakes, pops a structure from the queue, figures out what to execute and calls the appropriate method.
This implementation works but adding new methods is quite cumbersome: define wrapper, structure with parameters, and handler. I was wondering if there is a more straightforward means of coding this kind of model: a class method that executes on the class's own thread, instead of in the caller's thread.
edit - kind of conclusion:
It seems that there's no de facto way to implement what I asked that doesn't involve extra coding effort.
I'll stick with what I came up with, it ensures type safeness, minimizes locking, allows sync and async calls and the overhead it fairly modest.
On the other hand it requires a bit of extra coding and the dispatch mechanism may become bloated as the number of methods increases. Registering the dispatch methods on construction, or having the wrappers do that work seem to solve the issue, remove a bit of overhead and also remove some code.
My standard reference for this problem is here.
Implementing a Thread-Safe Queue using Condition Variables
As #John noted, this uses Boost.Thread.
I'd be careful about the synchronous case you described here. It's easy to get perf problems if the producer (the sending thread) waits for a result from the consumer (the service thread). What happens if you get 1000 async calls, filling up the queue with a backlog, followed by a sync call from each of your producer threads? Your system will 'play dead' until the queue backlog clears, freeing up those sync callers. Try to decouple them using async only, if you can.
There are several ways to achieve this, depending upon the complexity you want to accept. Complexity of the code is directly proportional to the flexibility desired. Here's a simple one (and quite well used):
Define a classes corresponding to each functionality your server exposes.
Each of these classes implements a function called execute and take a basic structure called input args and output args.
Inside the service register these methods classes at the time of initialization.
Once a request comes to the thread, it will have only two args, Input and Ouput, Which are the base classes for more specialized arguments, required by different method classes.
Then you write you service class as mere delegation which takes the incoming request and passes on to the respective method class based on ID or the name of the method (used during initial registration).
I hope it make sense, a very good example of this approach is in the XmlRpc++ (a c++ implementation of XmlRpc, you can get the source code from sourceforge).
To recap:
struct Input {
virtual ~Input () = 0;
};
struct Ouput {
virtual ~Output () = 0;
};
struct MethodInterface {
virtual int32_t execute (Input* __input, Output* __output) = 0;
};
// Write specialized method classes and taking specialized input, output classes
class MyService {
void registerMethod (std::string __method_name, MethodInterface* __method);
//external i/f
int32_t execute (std::string __method, Input* __input, Output* __output);
};
You will still be using the queue mechanism, but you won't need any wrappers.
IMHO, If you want to decouple method execution and thread context, you should use Active Object Pattern (AOP)
However, you need to use ACE Framework, which supports many OSes, e.g. Windows, Linux, VxWorks
You can find detailed information here
Also, AOP is a combination of Command, Proxy and Observer Patterns, if you know the details of them, you may implement your own AOP. Hope it helps
In addition to using Boost.Thread, I would look at boost::function and boost::bind. That said, it seems fair to have untyped (void) arguments passed to the target methods, and let those methods cast to the correct type (a typical idiom for languages like C#).
Hey now Rajivji, I think you have it upside-down. Complexity of code is inversely proportional to flexibility. The more complex your data structures and algorithms are, the more restrictions you are placing on acceptable inputs and behaviour.
To the OP: your description seems perfectly general and the only solution, although there are different encodings of it. The simplest may be to derive a class from:
struct Xqt { virtual void xqt(){} virtual ~Xqt(){} };
and then have a thread-safe queue of pointers to Xqt. The service thread then just pops the queue to px and calls px->xqt(), and then delete px. The most important derived class is this one:
struct Dxqt : Xqt {
xqt *delegate;
Dxqt(xqt *d) : delegate(d) {}
void xqt() { delegate->xqt(); }
};
because "all problems in Computer Science can be solved by one more level of indirection" and in particular this class doesn't delete the delegate. This is much better than using a flag, for example, to determine if the closure object should be deleted by the server thread.
I have a project with a large codebase (>200,000 lines of code) I maintain ("The core").
Currently, this core has a scripting engine that consists of hooks and a script manager class that calls all hooked functions (that registered via DLL) as they occur. To be quite honest I don't know how exactly it works, since the core is mostly undocumented and spans several years and a magnitude of developers (who are, of course, absent). An example of the current scripting engine is:
void OnMapLoad(uint32 MapID)
{
if (MapID == 1234)
{
printf("Map 1234 has been loaded");
}
}
void SetupOnMapLoad(ScriptMgr *mgr)
{
mgr->register_hook(HOOK_ON_MAP_LOAD, (void*)&OnMapLoad);
}
A supplemental file named setup.cpp calls SetupOnMapLoad with the core's ScriptMgr.
This method is not what I'm looking for. To me, the perfect scripting engine would be one that will allow me to override core class methods. I want to be able to create classes that inherit from core classes and extend on them, like so:
// In the core:
class Map
{
uint32 m_mapid;
void Load();
//...
}
// In the script:
class ExtendedMap : Map
{
void Load()
{
if (m_mapid == 1234)
printf("Map 1234 has been loaded");
Map::Load();
}
}
And then I want every instance of Map in both the core and scripts to actually be an instance of ExtendedMap.
Is that possible? How?
The inheritance is possible. I don't see a solution for replacing the instances of Map with instances of ExtendedMap.
Normally, you could do that if you had a factory class or function, that is always used to create a Map object, but this is a matter of existing (or inexistent) design.
The only solution I see is to search in the code for instantiations and try to replace them by hand. This is a risky one, because you might miss some of them, and it might be that some of the instantiations are not in the source code available to you (e.g. in that old DLL).
Later edit
This method overriding also has a side effect in case of using it in a polymorphic way.
Example:
Map* pMyMap = new ExtendedMap;
pMyMap->Load(); // This will call Map::Load, and not ExtendedMap::Load.
This sounds like a textbook case for the "Decorator" design pattern.
Although it's possible, it's quite dangerous: the system should be open for extension (i.e. hooks), but closed for change (i.e. overriding/redefining). When inheriting like that, you can't anticipate the behaviour your client code is going to show. As you see in your example, client code must remember to call the superclass' method, which it won't :)
An option would be to create a non-virtual interface: an abstract base class that has some template methods that call pure virtual functions. These must be defined by subclasses.
If you want no core Map's to be created, the script should give the core a factory to create Map descendants.
If my experience with similar systems is applicable to your situation, there are several hooks registered. So basing a solution on the pattern abstract factory will not really work. Your system is near of the pattern observer, and that's what I'd use. You create one base class with all the possible hooks as virtual members (or several one with related hooks if the hooks are numerous). Instead of registering hooks one by one, you register one object, of a type descendant of the class with the needed override. The object can have state, and replace advantageously the void* user data fields that such callbacks system have commonly.
Actually, this question seems to have two parts:
How to implement pattern matching?
How to implement send and receive (i.e. the Actor model)?
For the pattern matching part, I've been looking into various projects like App and Prop. These look pretty nice, but couldn't get them to work on a recent version (4.x) of g++. The Felix language also seems to support pattern matching pretty well, but isn't really C++.
As for the Actor model, there are existing implementations like ACT++ and Theron, but I couldn't find anything but papers on the former, and the latter is single-threaded only [see answers].
Personally, I've implemented actors using threading and a thread-safe message queue. Messages are hash-like structures, and used these together with a number of preprocessor macros to implemented simple pattern matching.
Right now, I can use the following code to send a message:
(new Message(this))
->set("foo", "bar")
->set("baz", 123)
->send(recipient);
And the following to do simple pattern matching (qDebug and qPrintable are Qt-specific):
receive_and_match(m)
match_key("foo") { qDebug("foo: %s", qPrintable(m->value("foo").toString())); }
or_match_key("baz") { qDebug("baz: %d", m->value("baz").toInt()); }
or_match_ignore
end_receive
However, this looks a bit hackish to me, and isn't very robust.
How would you do it? Did I miss any existing work?
As for the Actor model, there are
existing implementations like ACT++
and Theron, but I couldn't find
anything but papers on the former, and
the latter is single-threaded only.
As the author of Theron, I was curious why you believe it's single-threaded?
Personally, I've implemented actors
using threading and a thread-safe
message queue
That's how Theron is implemented.. :-)
Ash
One of the important things about erlang is how the features are used to make robust systems.
The send/recieve model is no-sharing, and explicitly copying.
The processes themselves are lightweight threads.
If you did desire the robust properties of the erlang model, you would be best to use real processes and IPC rather than threads.
If you want robust message passing though you may end up wanting to serialize and deserialise the contents. Especially with type safety.
Pattern matching in C++ isn't always pretty but there will be a good pattern for this - you will end up creating a dispatcher object that uses some form of polymorphism to get what you want.
Although if you are not careful you end up with xml over pipes :)
Really, if you want the erlang model you really want to use erlang. If there are slow bits, I'm sure you can augment your program using a foreign function internet.
The problem about re-implementing parts, is you won't get a good cohesive library and solution. The solutions you have already don't look much like C++ anymore.
I'm currently implementing an actor library for C++ called "acedia" (there's nothing yet about it on google) that uses "type matching". The library is a project for my master thesis and you can send any kind of data to an actor with it.
A small snippet:
recipient.send(23, 12.23f);
And on the recipient side you can either analyze the received message like this:
Message msg = receive();
if (msg.match<int, float>() { ... }
... or you can define a rule set that invokes a function or method for you:
void doSomething(int, float);
InvokeRuleSet irs;
irs.add(on<int, float>() >> doSomething);
receiveAndInvoke(irs);
It's also possible to match both on type and on value:
Message msg = receive();
if (msg.match<int, float>(42, WILDCARD) { ... }
else if (msg.match<int, float>() { ... }
The constant "WILDCARD" means, that any value will be acceptet. Pass no arguments is equal set all arguments to "WILDCARD"; meaning that you only want to match the types.
This is certainly a small snippet. Also you can use "case classes" like in Scala. They are comparable to "atomics" in erlang. Here is a more detailed example:
ACEDIA_DECLARE_CASE_CLASS(ShutdownMessage)
ACEDIA_DECLARE_CASE_CLASS(Event1)
ACEDIA_DECLARE_CASE_CLASS(Event2)
To react to the defined case classes you can write an actor like this:
class SomeActor : public Actor
{
void shutdown() { done = true; }
void handleEvent1();
void handleEvent1();
public:
SomeActor() : done(false) { }
virtual void act()
{
InvokeRuleSet irs;
irs
.add(on<ShutdownMessage>() >> method(&SomeActor::shutdown))
.add(on<Event1>() >> method(&SomeActor::handleEvent1))
.add(on<Event2>() >> method(&SomeActor::handleEvent2))
;
while (!done) receiveAndInvoke(irs);
}
};
To create a new actor and start it, all you have to write is:
Acedia::spawn<SomeActor>();
Although the library not even reached beta stadium the shown snippets work and i have a first application running on it. One major goal of the library is to support distributed programming (also across a network).
Your question is a while ago, but if you're interested in it: let me know! :)
You can mimic the behavior using Qt's signal/slot mechanism, especially since Qt's signal/slot supports multithread.
I would definitely be interested in looking at your "acedia" library and would love to help in any way that I could. Erlang has some wonderful constructs and C++ could definitely benefit from such a library.
Today I hostet the library at sourceforge: https://sourceforge.net/projects/acedia/
As I said before it's an early release. But feel free to critique it!
Today, if you want erlang style robust actors in C++, and pattern matching,
maybe Rust is the answer.
Of course this wasn't around publically when the OP asked ~5years ago, and as of april 2014 it still isn't v1.0 yet - but its been progressing very well and is definitely stabilizing, enough of the language core is stable I think.
And ok its not C++, but it has the same approach to memory management as C++, except that it supports lightweight tasks with no shared memory by default (then provides controlled library features for sharing - "Arc");
It can directly call (and directly expose) 'extern C' functions. You can't share templated library headers with C++ - but you can write generics that mimick C++ collection classes (and vica versa) to pass references to data-structures across.