Monthly Archives: October 2012

C++ Reflection: Type MetaData: Part 3 – Improvements

In our last article we learned how to store information about a classes’s members, however there are a couple key improvements that need to be brought to the MetaData system before moving on.

The first issue is with our RemQual struct. In the previous article we had support for stripping off qualifiers such as *, const or &. We even had support for stripping off an R-value reference. However, the RemQual struct had no support for a pointer to a pointer. It is weird that RemQual would behave differently than RemQual, and so on. To solve this issue we can cycle down, at compile time, the type through the RemQual struct recursively, until a type arrives at the base RemQual definition. Here’s an example:

As you can see, this differs a bit from our previous implementation. The way it works is by passing in a single type to the RemQual struct via typename T. Then, the templating matches the type provided with one of the overloads and feeds the type back into the RemQual struct with less qualifiers. This acts as some sort of compile-time “recursive” qualifier stripping mechanism; I’m afraid I don’t know what to properly call this technique. This is useful for finding out what the “base type” of any given type.

It should be noted that the example code above does not strip pointer qualifiers off of a type. This is to allow the MetaData system to properly provide MetaData instances of pointer types; which is necessary to reflect pointer meta.

It should be noted that in order to support pointer meta, the RemQual struct will need to be modified so it does not strip off the * qualifier. This actually applies to any qualifier you do not wish to have stripped.

There’s one last “improvement” one could make to the RemQual struct that I’m aware of. I don’t actually consider this an improvement, but more of a feature or decision. There comes a time when the user of a MetaData system may want to write a tidbit of code like the following:

Say the user wants to send a message object from one place to another. Imagine this message object can take three parameters of any type, and the reflection system can help the constructor of the message figure out the types of the data at run-time (how to actually implement features like this will be covered once Variants and RefVariants are introduced). This means that the message can take three parameters of any type and then take them as payload to deliver elsewhere.

However, there’s a subtle problem with the “Message ID” in particular. Param1 and Param2 are assumed to be POD types like float or int, however “Message ID” is a const char * string literal. My understanding of string literals in C++ is that they are of the type: const char[ x ], x being the number of characters in the literal. This poses a problem for our templated MetaCreator, in that every value of x will create a new MetaData instance, as the templating treats each individual value of x as an entire new type. Now how can RemQual handle this? It gets increasingly difficult to actually manage Variants and RefVariant constructors for string literals for reasons explained here, though this will be tackled in a later article.

There are two methods of handling string literals that I am aware of; the first is to make use of some form of a light-weight wrapper. A small wrapper object can contain a const char * data member, and perhaps an unsigned integer to represent the length, and any number of utility functions for common string operations (concat, copy, compare, etc). The use of such a wrapper would look like:

The S would be the class type of the wrapper itself, and the constructor would take a const char *. This would require every place in code that handles a string literal to make use of the S wrapper. This can be quite annoying, but has great performance benefits compared to std::string, especially when some reference counting is used to handle the heap allocated const char * data member holding the string data in order to avoid unnecessary copying. Here’s an example skeleton class for such an S wrapper:

As I mentioned before, I found this to be rather annoying; I want my dev team and myself to be able to freely pass along a string literal anywhere and have MetaData handle the type properly. In order to do this, a very ugly and crazy solution was devised. There’s a need to create a RemQual struct for every [ ] type for all values of x. This isn’t possible. However, it is possible to overload RemQual for a few values of x, at least enough to cover any realistic use of a string literal within C++ code. Observe:

The macro ARRAY_OVERLOAD creates a RemQual overload with a value of x. The __COUNTER__ macro (though not standard) increments by one each time the macro is used. This allows for copy/pasting of the ARRAY_OVERLOAD macro, which will generate a lot of RemQual overloads. I created a file with enough overloads to cover any realistically sized string literal. As an alternative to the __COUNTER__ macro, __LINE__ can be used instead, however I imagine it might be difficult to ensure you have one definition per line without any gaps. As far as I know, __COUNTER__ is supported on both GNU and MSVC++.

Not only will the ARRAY_OVERLOAD cover types of string literals, but it will also cover types with array brackets [ ] of any type passed to RemQual.

The second issue is the ability to properly reflect private data members. There are three solutions to reflecting private data that I am aware of. The first is to try to grant access to the MetaData system by specifying that the MetaCreator of the type in question is a friend class. I never really liked the idea of this solution and haven’t actually tried it for myself, and so I can’t really comment on the idea any further than this.

The next possible solution is to make use of properties. A property is a set of three things: a gettor; a settor; a member. The gettor and settor provide access to the private member stored within the class. The user can then specify gettors and/or settors from the ADD_MEMBER macro. I haven’t implemented this method myself, but would definitely like if I find the time to create such a system. This solution is by far the most elegant of the three choices that I’m presenting. Here’s a link to some information on creating some gettor and settor support for a MetaData system like the one in this article series. This can potentially allow a MetaData system to reflect class definitions that the user does not have source code access to, so long as the external class has gettor and settor definitions that are compatible with the property reflection.

The last solution is arguably more messy, but it’s easier to implement and works perfectly fine. I chose to implement this method in my own project because of how little time it took to set up a working system. Like I said earlier, if I have time I’d like to add property support, though right now I simply have more important things to finish.

The idea of the last solution is to paste a small macro inside of your class definitions. This small macro then pastes some code within the class itself, and this code grants access to any private data member by using the NullCast pointer trick. This means that in order to reflect private data, you must have source code access to the class in question in order to place your macro. Here’s what the new macros might look like, but be warned it gets pretty hectic:

The META_DATA macro is to be placed within a class, it places a couple declarations for NullCast, AddMember and RegisterMetaData. The DEFINE_META macro is modified to provide definitions for the method declarations created by the META_DATA macro. This allows the NullCast to retrieve the type to cast to from the DEFINE_META’s TYPE parameter. Since AddMember method is within the class itself, it can now have proper access to private data within the class. The AddMember definition within the class then forwards the information it gathers to the AddMember function within the MetaCreator.

In order for the DEFINE_META api to remain the same as before, the META_DATA macro creates a RegisterMetaData declaration within the class itself. This allows the ADD_MEMBER macro to not need to user to supply to type of class to operate upon. This might be a little confusing, but imagine trying to refactor the macros above. Is the RegisterMetaData macro even required to be placed into the class itself? Can’t the RegisterMetaData function within the MetaCreator call AddMember on the class type itself? The problem with this is that the ADD_MEMBER macro would require the user to supply the type to the macro like this:

This would be yet another thing the user of the MetaData system would be required to perform, thus cluttering the API. I find that by keeping the system as simple as possible is more beneficial than factoring out the definition of RegisterMetaData from the META_DATA macro.

Here’s an example usage of the new META_DATA and DEFINE_META macros:

The only additional step required here is for the user to remember to place the META_DATA macro within the class definition. The rest of the API remains as intuitive as before.

Here’s a link to a compileable (in VS2010) example showing everything I’ve talked about in the MetaData series thus far. The next article in this series will likely be in creating the Variant type for PODs.


C++ Reflection: Type MetaData: Part 2 – Type Reduction and Members

In the last post we learned the very basics of setting up a reflection system. The whole idea is that the user manually adds types into the system using a single simple macro placed within a cpp file, DEFINE_META.

In this article I’ll talk about type deduction and member reflection, both of which are critical building blocks for everything else.

First up is type deduction. When using the templated MetaCreator class:

Whenever you pass in a const, reference, or pointer qualifier an entire new templated MetaCreator will be constructed by the compiler. This just won’t do, as we don’t want the MetaData of a const int to be different at all from an int, or any other registered type. There’s a simple, yet very quirky, trick that can solve all of our problems. Take a look at this:

I’m actually not familiar with the exact terminology to describe what’s going on here, but I’ll try my best. There’s many template overloads of the first RemQual struct, which acts as the “standard”. The standard is just a single plain type T, without any qualifiers and without pointer or reference type. The rest of the templated overloaded version all contain a single typedef which lets the entire struct be used to reference a single un-qualified type by supplying any of the various overloaded types to the struct’s typename param.

Overloads for the R-value reference must be added as well in order to strip down to the bare type T.

Now that we have our RemQual (remove qualifiers) struct, we can use it within our META macros to refer to MetaData types. Take a look at some example re-writes of the three META macros:

The idea is you feed in the typedef’d type from RemQual into the MetaCreator typename param. This is an example of using macros well; there’s no way to screw up the usage of them, and they are still very clean and easy to debug as there isn’t really any abuse going on. Feel free to ignore specific META macros you wouldn’t actually use. I use all three META_TYPE, META and META_STR. It’s a matter of personal preference on what you actually implement in this respect. It will likely be pretty smart to place whatever API is created into a namespace of it’s own, however.

And that covers type deduction. There are some other ways of achieving the same effect, like partial template specialization as covered here, though I find this much simpler.

Next up is register members of structures or classes with the MetaData system. Before anything continues, lets take a look at an example Member struct. A Member struct is a container of the various bits of information we’d like to store about any member:

This member above is actually almost exactly what implementation I have in my own reflection as it stands while I write this; there’s not a lot needed. You will want a MetaData instance to describe the type of data contained, a name identifier, and an unsigned offset representing the member’s location within the containing object. The offset is exceptionally important for automated serialization, which I’ll likely be covering after this article.

The idea is that a MetaData instance can contain various member objects. These member objects are contained within some sort of container (perhaps std::vector).

In order to add a member we’ll want another another very simple macro. There are two big reasons a macro is efficient in this situation: we can use stringize; there’s absolutely no way for the user to screw it up.

Before showing the macro I’d like to talk about how to retrieve the offset. It’s very simple. Take the number zero, and turn this into a pointer to a type of object (class or struct). After the typecasting, use the -> operator to access one of the members. Lastly, use the & operator to retrieve the address of the member’s location (which will be offset from zero by the -> operator) and typecast this to an unsigned integer. Here’s what this looks like:

This is quite the obtrusive line of code we have here! This is also a good example of a macro used well; it takes a single parameter and applies it to multiple locations. There’s hardly any way for the user of this macro to screw up.

NullCast is a function I’ll show just after this paragraph. All it does is return a pointer to NULL (memory address zero) of some type. Having this type pointer to address zero, we then use the ADD_MEMBER macro to provide the name of a member to access. The member is then accessed, and the & operator provides an address to this member with an offset from zero. This value is then typecasted to an unsigned integer and and passed along to the AddMember function within the macro. The stringize operator is also used to pass a string representation of the member to the AddMember function, as well as a MetaData instance of whatever the type of data the member is.

Now where does this AddMember function actually go? Where is it from? It’s actually placed into a function definition. The function AddMember itself resides within the MetaCreator. This allows the MetaCreator to call the AddMember function of the MetaData instance it holds, which then adds the Member object into the container of Members within the MetaData instance.

Now, the only place that this AddMember function can be called from, building from the previous article, is within the MetaCreator’s constructor. The idea is to use the DEFINE_META macro to also create a definition of either the MetaCreator’s constructor, or a MetaCreator method that is called from the MetaCreator’s constructor. Here’s an example:

As you can see this formation is actually very intuitive; it has C++-like syntax, and it’s very clear what is going on here. A GameObject is being registered in the Meta system, and it has members of ID, active, and components being added to the Meta system. For clarity,  here’s what the GameObject’s actual class definition might look like (assuming component based architecture):

// This boolean should always be true when the object is alive. If this is
// set to false, then the ObjectFactory will clean it up and delete this object
// during its inactive sweep when the ObjectFactory’s update is called.
bool active;
std::vector components;

Now lets check out what the new DEFINE_META macro could look like:

The RegisterMetaData declaration is quite peculiar, as the macro just ends there. What this is doing is setting up the definition of the RegisterMetaData function, so that the ADD_MEMBER macro calls are actually lines of code placed within the definition. The RegisterMetaData function should be called from the MetaCreator’s constructor. This allows the user to specify what members to reflect within a MetaData instance of a particular type in a very simple and intuitive way.

Last but not least, lets talk about the NullCast function real quick. It resides within the MetaCreator, as NullCast requires the template’s typename MetaType in order to return a pointer to a specific type of data.

And that’s that! We can now store information about the members of a class and deduce types from objects in an easily customizeable way.

Here’s a link to a demonstration program, compileable in Visual Studio 2010. I’m sure this could compile in GCC with a little bit of house-keeping as well, but I don’t really feel like doing this as I need to get to bed! Here’s the output of the program. The format is . For the object, members and their offsets are printed:

Now you might notice at some point in time, you cannot reflect private data members! This detail will be covered in a later article. The idea behind it is that you require source code access to the type you want to reflect, and place a small tiny bit of code inside to gain private data access. Either that or make the MetaCreator a friend class (which sounds like a messy solution to me).

And here we have all the basics necessary for automated serialization! We can reflect the names of members, their types, and offsets within an object. This lets the reflection register any type of C++ data within itself.

C++ Reflection: Type MetaData: Introduction

Edit: Take these articles about reflection with a grain of salt. They were written when I was learning all this stuff as a tool of education.

Thinking about how to take things a step further in terms of what your tools can do for you is really important in increasing productivity. This is where a form of reflection can come in very handy. I call the reflection system I’m working on by shorthand of “meta” or “MetaData”, though the proper terminology for it would be something more like Class MetaData or Type Reflection. So when I say MetaData I’m referring to data about data, and specifically data about the types of data within your C++ code.

Having a MetaData system allows for information about your data types to be saved during run-time for use during run-time. The C++ compiler compiles away a lot of information, and a lot of this information is very useful. So, a MetaData system saves this information from being compiled away.

So what is the use of such a system? Where here are the things I’ve constructed so far: simple and powerful debugging features; automated serialization for any type registered with Meta; automic function binding to Lua; Variant and RefVariant class types (objects that can hold any type of data registered within Meta). However these aren’t the only things I’ll be using the MetaData system for. One could also apply the MetaData to make simple object factories with hardly any code, or property viewing tables for editors with ease. I’m sure there’s even more uses as well that I just haven’t quite come to terms with. In a way, systems like this can generate tools and functionality for the developer whenever a new class or data type is introduced.

Before we start I want to let the reader know that such a system can be very efficient, efficient enough to run wonderfully within a real-time application like a game. I myself don’t know a lot about efficiency or optimization at a low-level, though I do know other programmers who’ve made such Reflection systems that are very, very fast. So make sure not to have any pre-built misconceptions about a C++ Reflection system before moving on.

I started learning about how to construct a MetaData system from this post. You can take a look if you like, though it won’t be necessary if you just want to learn what I’m trying to teach here. In that post, by a fellow student here at DigiPen, he goes over how to get started with MetaData, but leaves a lot to yet be learned. I’ll be modelling my post after his quite a bit, as he does have good content structure for an introduction post.

I’ll attempt to document what I’ve learned here as honestly, there’s not any resources on constructing a MetaData system like this anywhere as far as I know. The closest thing I could find was from a Game Programming Gems book, the 5th one chapter 1.4, although it required all classes that wanted to participate to inherit from the MetaData class. This isn’t really sufficient if you want to reflect class or structure types you don’t have source code access to, and it doesn’t support the built-in types at all.

Getting Started
To start lets talk about the overall structure of what a MetaData system looks like. I think it’s going to best to draw a diagram of what will be achieved from this article:

Diagram of entire MetaData system layout.

In the above diagram the MetaData objects are the key object in which all important operations are performed. A MetaData object is a non-templated class, allowing them to be stored in a data structure. There is one MetaData object for each type of data registered to the system, and the MetaData object of a corresponding type is a representation of that type. It stores information about whether or not it is a Plain Old Data type (POD), the size of the type, it’s members and methods, and name. Inheritance information can be stored along with multiple inheritance information, though I haven’t even bothered adding that feature in yet as it’s not actually very useful and quite difficult to do properly.

The MetaCreator is a templated class that manages the creation of a single unique MetaData instance. It then adds its instance into the MetaManager which contains it in some sort of map.

The MetaManager is a non-templated class that contains all of the created MetaData instances, and can perform search operations on them to find specific types. I use a map of strings to instances, so I can search by string identifier. I’ve also placed some other small utility functions into my MetaManager as well.

Client Code
Before we get started writing anything, I’d like to try to show some example client code to exemplify why I’ve taken all this time to make a MetaData system in the first place.

As you can see I have a nice DESERIALIZE macro that can deserialize any type of data registered within within the Reflection. My entire serialization file (includes in and out) is only about 400 lines of code, and I implemented my own custom file format. I also have a LuaReference data type, which contains a handle and a MetaData instance, and allows any class to be sent to Lua via handle. Because of my Meta system I can write very generic and powerful code pretty easily.

Getting Started
The first I recommend starting with is to reflect the size of a type, and the name of a type. Here’s what a very simple  MetaData class could like:

This simple class just stores the size and name of a type of data. The next thing required is the ability to create a unique instance of MetaData. This requires a templated class called the MetaCreator.

You should be able to see that by passing in a type, perhaps to the MetaCreator, a single MetaData instance becomes available from the Get function, and is associated with the MetaCreator class. Any type can be passed into the Metatype typename. The MetaCreator constructor initializes the MetaData instance. This is important. In the future you’ll have MetaData for a class that contains MetaData for POD types. However because of out-of-order initialization, some of the types required to construct your class MetaData instance might not be initialized (as in the Init function will not have been called) yet. However if you use the MetaManager<>::Get( ) function, you can retrieve a pointer to the memory location that will be initialized once the MetaCreator of that specific type is constructed. It should be noted that the construction of a MetaCreator happens within a macro, so that there’s absolutely no way of screwing up the type registration (the inside of the macro will become quite… ugly).

Lastly you’ll need a place to store all of your MetaData instances: the MetaManager!

And there we have a nice way to store all of our MetaData instances. The Get function is most useful in retrieving a MetaData instance of a type by string name. Now that we have our three major facilities setup, we can talk about the macros involved in actually registering a type within the MetaData system.

So far so good. Using the DEFINE_META macro it’s pretty easy to add a type to the MetaData system, simply do DEFINE_META( type );. The decltype might be confusing, as it’s new. decltype simply returns the type of an object. This allows the user to retrieve a MetaData instance that corresponds to an object’s type, without knowing what the object’s type is; this lets very generic code be easily written.

NAME_GENERATOR is a bit tricky. Every single instance of a MetaCreator needs to be constructed at global scope- this is the only way to get the Init( ) function to be called, without having to place your DEFINE_META macro in some sort of code scope. Without the constructor of the MetaCreator calling Init, the only way to have any sort of code run by using the DEFINE_META macro is to place it within some scope that is run sometime after main executes. This makes the use of the DEFINE_META macro harder. If you create the MetaCreator at global scope, then you can have the constructor’s code run before main executes at all, making the DEFINE_META macro very easy and simple to use.

So then comes the issue of “what do I call my MetaCreator?” arises. The first thing you might think of is, just call it MetaCreator and make it static. This hides the MetaCreator at file scope, allowing the DEFINE_META macro to be used once per file without any naming conflicts. However, what if you need more than one DEFINE_META in a file? The next solution I thought of was to use token pasting: ## operator. Here’s an example usage of the token pasting technique:

The only problem with this strategy is that you cannot pass in type names with special characters or spaces, as that won’t result in a proper token name. The last solution is to use some sort of unique name generation technique. There are two macros __LINE__ and __FILE__ that can be used to generate a unique name each time the macro is used, so long as it is not used twice on the same line of the same file. The __LINE__ and __FILE__ definitions are a bit tedious to use, but I think I have them working properly, like this:

You have to feed in the __COUNTER__ definition carefully in order to make sure they are translated by the preprocessor into their respective values. A resulting token could look like: GENERATED_NAME__29. This is perfect for creating all of the MetaCreators at global scope on the stack. Without the use of some sort of trick like this, it can be very annoying to have to use a function call to register your MetaData.

Alternatively there are __FILE__ and __LINE__ macros, but they aren’t really necessary as the __COUNTER__ does everything we need. The __COUNTER__ however is, I believe, not actually standard.

So far everything is very straightforward, as far as I can tell. Please ask any questions you have in the comments!

It should be noted that const, &, and * types all create different MetaData instances. As such, there is a trick that can be used to strip these off of an object when using the META macro. This will be covered in a subsequent article.

Example Uses
Now lets check out some example client code of what our code can actually do!

And there we have it! An extremely easy to use DEFINE_META macro that stores name and size information of any type, including class and struct types.

Future Posts
For future posts I look forward to writing about automated Lua binding, Variants and RefVariants, automated Serialization, factory usage, messaging, and perhaps other topics as well! These topics are really rather huge, so please by patient in that it may take a while to cover everything.

Link to second article in series.