Skip to content

Latest commit

 

History

History
555 lines (425 loc) · 14.4 KB

GettingStarted.md

File metadata and controls

555 lines (425 loc) · 14.4 KB
Raw

Getting started

From an higher level overview, CppSharp will take a bunch of user-provided C/C++ headers and generate either C++/CLI or C# code that can be compiled into a regular .NET assembly.

Since there are no binary releases yet, the project needs to be compiled from source first.

LLVM/Clang source repositories

SVN repository urls found here: http://clang.llvm.org/get_started.html

Git repository urls found here: http://llvm.org/docs/GettingStarted.html#git-mirror

Compiling on Windows/Visual Studio

  1. Clone CppSharp to <CppSharp>
  2. Clone LLVM to <CppSharp>\deps\llvm
  3. Clone Clang to <CppSharp>\deps\llvm\tools\clang
  4. Run CMake in <CppSharp>\deps\llvm\build and compile solution in RelWithDebInfo mode
  5. Run GenerateProjects.bat in \build
  6. Build generated solution in Release.

Building in Release is recommended because else the Clang parser will be excruciatingly slow.

Due to some incompatible changes in recent LLVM versions, Windows builds need to use the following set of revisions that are known to work.

LLVM r202563 / Git mirror revision c9bf74fdc543c2da90f334e0bf8e34b128c8a615.

Clang r202562 / Git mirror revision 30577e6c4f5ba66b8dc832ed9955b1a7475f788a.

Compiling on Mac OS X (experimental)

  1. Clone CppSharp to <CppSharp>
  2. Clone LLVM to <CppSharp>\deps\llvm
  3. Clone Clang to <CppSharp>\deps\llvm\tools\clang
  4. Run CMake in <CppSharp>\deps\llvm\build and compile solution in RelWithDebInfo mode The following CMake variables should be enabled:
    • LLVM_ENABLE_CXX11 (enables C++11 support)
    • LLVM_ENABLE_LIBCXX (enables libc++ standard library support)
    • LLVM_BUILD_32_BITS for 32-bit builds (defaults to 64-bit)
  5. Run premake5 gmake in \build
  6. Build generated makefiles:
    • 32-bit builds: config=release_x32 make
    • 64-bit builds: config=release_x64 make

Compiling on Linux (experimental)

Build dependencies:

If you do not have native build tools you can install them first with:

sudo apt-get install build-essential gcc-multilib g++-multilib
sudo apt-get install ninja-build cmake

Additionaly we depent on a somewhat recent version of Mono (.NET 4.5). If you're using an Ubuntu-based distribution you can install an up-to-date version from: https://launchpad.net/~directhex/+archive/monoxide

sudo add-apt-repository ppa:directhex/monoxide
sudo apt-get update
sudo apt-get install mono-devel

Getting Premake:

Download a recent Premake version from: http://sourceforge.net/projects/premake/files/Premake/nightlies/premake-dev-linux.zip/download

Extract the binary inside premake5 to <CppSharp>/build.

Cloning CppSharp:

git clone https://github.com/mono/CppSharp.git

Cloning and building LLVM/Clang:

pushd && cd CppSharp/deps

git clone https://github.com/llvm-mirror/llvm.git
git clone https://github.com/llvm-mirror/clang.git llvm/tools

cd llvm && mkdir build && cd build

cmake -G Ninja -DCLANG_BUILD_EXAMPLES=false -DCLANG_ENABLE_ARCMT=false \
  -DCLANG_ENABLE_REWRITER=false -DCLANG_ENABLE_STATIC_ANALYZER=false \
  -DCLANG_INCLUDE_DOCS=false -DCLANG_INCLUDE_TESTS=false \
  -DLLVM_BUILD_32_BITS=false -DCLANG_BUILD_DOCS=false \
  -DCLANG_BUILD_EXAMPLES=false -DLLVM_TARGETS_TO_BUILD="X86" ..

ninja
popd

Building CppSharp:

cd CppSharp/build
./premake5 gmake
cd gmake
make

This will compile the default target for the architecture, for instance, the debug_x32 target for 32-bits X86.

You can change the target by invoking make as:

config=release_x32 make

Additionaly if you need more verbosity from the builds invoke make as:

verbose=true make

Generating bindings

Suppose we have the following declarations in a file named Sample.h and we want to bind it to .NET.

class Foo
{
public:

	int a;
	float b;
};

int FooAdd(Foo* foo);

The easiest way to get started with CppSharp is to create a new class and implement the ILibrary interface.

Each implemented method will be called by the generator during different parts of the binding process.

public interface ILibrary
{
	/// Setup the driver options here.
	void Setup(Driver driver);

	/// Setup your passes here.
	void SetupPasses(Driver driver, PassBuilder passes);

	/// Do transformations that should happen before passes are processed.
	void Preprocess(Driver driver, Library lib);

	/// Do transformations that should happen after passes are processed.
	void Postprocess(Driver driver, Library lib);
}

Then you just need to call the ConsoleDriver.Run static method with your an instance of your class to start the generation process. Like this:

ConsoleDriver.Run(new SampleLibrary());

Now let's drill through each of the interface methods in more detail:

  1. void Setup(Driver driver)

This is the first method called and here you should setup all the options needed for Clang to correctly parse your code. You can get at the options through a property in driver object. The essential ones are:

Parsing

  • Defines
  • Include directories
  • Headers
  • Libraries
  • Library directories

Generator

  • Output language (C# or C++/CLI)
  • Output namespace
  • Output directory

Here's how the setup method could be implemented for binding the sample code above:

void Setup(Driver driver)
{
    var options = driver.Options;
    options.GeneratorKind = GeneratorKind.CSharp;
    options.LibraryName = "Sample";
    options.Headers.Add("Sample.h");
    options.Libraries.Add("Sample.lib");
}

Pretty simple! We just need to tell the name of our library, what headers to process and the path to the compiled library containing the exported symbols of the declarations. And of course, what kind of output language we want.

This is enough to get the generator outputting some bindings:

public unsafe partial class Foo : IDisposable
{
    [StructLayout(LayoutKind.Explicit, Size = 8)]
    public struct Internal
    {
        [FieldOffset(0)]
        public int a;

        [FieldOffset(4)]
        public float b;

        [SuppressUnmanagedCodeSecurity]
        [DllImport("Sample.Native", CallingConvention = CallingConvention.ThisCall,
            EntryPoint="??0Foo@@QAE@XZ")]
        public static extern System.IntPtr Foo0(System.IntPtr instance);
    }

    public System.IntPtr _Instance { get; protected set; }

    internal Foo(Foo.Internal* native)
        : this(new System.IntPtr(native))
    {
    }

    internal Foo(Foo.Internal native)
        : this(&native)
    {
    }

    internal Foo(System.IntPtr native)
    {
        _Instance = native;
    }

    public Foo()
    {
        _Instance = Marshal.AllocHGlobal(8);
        Internal.Foo0(_Instance);
    }

    public void Dispose()
    {
        Dispose(disposing: true);
        GC.SuppressFinalize(this);
    }

    protected virtual void Dispose(bool disposing)
    {
        Marshal.FreeHGlobal(_Instance);
    }

    public int a
    {
        get
        {
            var _ptr = (Internal*)_Instance.ToPointer();
            return _ptr->a;
        }

        set
        {
            var _ptr = (Internal*)_Instance.ToPointer();
            _ptr->a = value;
        }
    }

    public float b
    {
        get
        {
            var _ptr = (Internal*)_Instance.ToPointer();
            return _ptr->b;
        }

        set
        {
            var _ptr = (Internal*)_Instance.ToPointer();
            _ptr->b = value;
        }
    }
}

public partial class SampleSample
{
    public struct Internal
    {
        [SuppressUnmanagedCodeSecurity]
        [DllImport("Sample.Native", CallingConvention = CallingConvention.Cdecl,
            EntryPoint="?FooAdd@@YAHPAVFoo@@@Z")]
        public static extern int FooAdd0(System.IntPtr foo);
    }

    public static int FooAdd(Foo foo)
    {
        var arg0 = foo._Instance;
        var ret = Internal.FooAdd0(arg0);
        return ret;
    }
}

The generator creates one managed class corresponding to each native class. It also creates one class per file to hold all the free functions in the native code, since free functions are not supported under C#. By default, it will use the library name followed by the file name, hence SampleSample.

Each bound class will have an Internal nested type which is used for interop with the native code. It holds the native P/Invoke declarations and has the same size as the native class, as it will be used to pass an instance of the object whenever value-type semantics are expected in the native code.

Alright, now that we know how to generate barebones bindings, let's see how we can clean them up.

Cleaning up after you

CppSharp provides plenty of support for customization of the generated code.

A couple of things could be improved in the generated code above:

  • Property names do not follow the .NET naming convention
  • FooAdd function could be an instance method instead of a static method

The main mechanism provided to customize the code are passes.

A pass is a simple tree visitor whose methods get called for each declaration that was parsed from the headers (or translation units).

CppSharp already comes with a collection of useful built-in passes and we will now see how to use them to fix the flaws enumerated above.

  1. void SetupPasses(Driver driver, PassBuilder passes)

New passes are added to the generator by using the API provided by PassBuilder.

void SetupPasses(Driver driver, PassBuilder passes)
{
	passes.RenameDeclsUpperCase(RenameTargets.Any);
	passes.FunctionToInstanceMethod();
}

Re-generate the bindings and voila:

public unsafe partial class Foo : IDisposable
{
    [StructLayout(LayoutKind.Explicit, Size = 8)]
    public struct Internal
    {
        [FieldOffset(0)]
        public int a;

        [FieldOffset(4)]
        public float b;

        [SuppressUnmanagedCodeSecurity]
        [DllImport("Sample.Native", CallingConvention = CallingConvention.ThisCall,
            EntryPoint="??0Foo@@QAE@XZ")]
        public static extern System.IntPtr Foo0(System.IntPtr instance);

        [SuppressUnmanagedCodeSecurity]
        [DllImport("Sample.Native", CallingConvention = CallingConvention.Cdecl,
            EntryPoint="?FooAdd@@YAHPAVFoo@@@Z")]
        public static extern int Add0(System.IntPtr instance);
    }

    public System.IntPtr _Instance { get; protected set; }

    internal Foo(Foo.Internal* native)
        : this(new System.IntPtr(native))
    {
    }

    internal Foo(Foo.Internal native)
        : this(&native)
    {
    }

    internal Foo(System.IntPtr native)
    {
        _Instance = native;
    }

    public Foo()
    {
        _Instance = Marshal.AllocHGlobal(8);
        Internal.Foo0(_Instance);
    }

    public void Dispose()
    {
        Dispose(disposing: true);
        GC.SuppressFinalize(this);
    }

    protected virtual void Dispose(bool disposing)
    {
        Marshal.FreeHGlobal(_Instance);
    }

    public int A
    {
        get
        {
            var _ptr = (Internal*)_Instance.ToPointer();
            return _ptr->a;
        }

        set
        {
            var _ptr = (Internal*)_Instance.ToPointer();
            _ptr->a = value;
        }
    }

    public float B
    {
        get
        {
            var _ptr = (Internal*)_Instance.ToPointer();
            return _ptr->b;
        }

        set
        {
            var _ptr = (Internal*)_Instance.ToPointer();
            _ptr->b = value;
        }
    }

    public int Add()
    {
        var ret = Internal.Add0(_Instance);
        return ret;
    }
}

Custom processing

Now that the bindings are looking good from a .NET perspective, let's see how we can achieve more advanced things by using the remaining overloads in the interface.

  1. void Preprocess(Driver driver, Library lib);
  2. void Postprocess(Driver driver, Library lib);

As their comments suggest, these get called either before or after the the passes we setup earlier are run and they allow you free reign to manipulate the declarations before the output generator starts processing them.

Let's say we want to change the class to provide .NET value semantics, drop one field from the generated bindings and rename the FooAdd function.

void Postprocess(Driver driver, Library lib)
{
  	lib.SetClassAsValueType("Foo");
  	lib.SetNameOfFunction("FooAdd", "FooCalc");
  	lib.IgnoreClassField("Foo", "b");
}

Re-generate the bindings and this is what we get:

public unsafe partial struct Foo
{
    [StructLayout(LayoutKind.Explicit, Size = 8)]
    public struct Internal
    {
        [FieldOffset(0)]
        public int a;

        [SuppressUnmanagedCodeSecurity]
        [DllImport("Sample.Native", CallingConvention = CallingConvention.ThisCall,
            EntryPoint="??0Foo@@QAE@XZ")]
        public static extern System.IntPtr Foo0(System.IntPtr instance);

        [SuppressUnmanagedCodeSecurity]
        [DllImport("Sample.Native", CallingConvention = CallingConvention.Cdecl,
            EntryPoint="?FooAdd@@YAHPAVFoo@@@Z")]
        public static extern int Calc0(System.IntPtr instance);
    }

    internal Foo(Foo.Internal* native)
        : this(new System.IntPtr(native))
    {
    }

    internal Foo(Foo.Internal native)
        : this(&native)
    {
    }

    internal Foo(System.IntPtr native)
    {
        var _ptr = (Internal*)native.ToPointer();
        A = _ptr->a;
    }

    internal Internal ToInternal()
    {
        var _native = new Foo.Internal();
        _native.a = A;
        return _native;
    }

    internal void FromInternal(Internal* native)
    {
        var _ptr = native;
        A = _ptr->a;
    }

    public int A;

    public int Calc()
    {
        var _instance = ToInternal();
        var ret = Internal.Calc0(new System.IntPtr(&_instance));
        FromInternal(&_instance);
        return ret;
    }
}

This is just a very small example of what is available. Since the entire AST is accessible, pretty much every customization one might need is possible.

The methods we called are in fact just regular .NET extension methods provided by the library to be used as helpers for common operations.

Where to go from here

Hopefully now you have a better idea of how the generator works and how to setup simple customizations to get the outputs better mapped to .NET.

This barely touched the surface of what can be done with CppSharp, so please check out the user and developer reference manuals for more information.