Version Notice

You need to use the correct branch of c2ffi for your version of LLVM/Clang:

• 3.4: branch llvm-3.4 (unsupported)
• 3.5: branch llvm-3.5 (unsupported)
• 3.6: branch llvm-3.6 (unsupported)
• 3.7: branch llvm-3.7 (unsupported)
• 4.0.0: branch llvm-4.0.0 (deprecated)
• 5.0.0: branch llvm-5.0.0 current

Developement will always take place in llvm-X.Y, according to the appropriate version of LLVM. The master branch has been removed. Check out the appropriate version for your LLVM.

c2ffi

This is a tool for extracting definitions from C, C++, and Objective C headers for use with foreign function call interfaces. For instance:

#define FOO (1 << 2)

const int BAR = FOO + 10;

typedef struct my_point {
    int x;
    int y;
    int odd_value[BAR + 1];
} my_point_t;

enum some_values {
    a_value,
    another_value,
    yet_another_value
};

void do_something(my_point_t *p, int x, int y);

Running c2ffi on this, we can get the following JSON output:

[
{ "tag": "const", "name": "BAR", "location": "/home/rpav/test.h:3:11", "type": { "tag": ":int" }, "value": 14 },
{ "tag": "struct", "name": "my_point", "id": 0, "location": "/home/rpav/test.h:5:16", "bit-size": 544, "bit-alignment": 32, "fields": [{ "tag": "field", "name": "x", "bit-offset": 0, "bit-size": 32, "bit-alignment": 32, "type": { "tag": ":int" } }, { "tag": "field", "name": "y", "bit-offset": 32, "bit-size": 32, "bit-alignment": 32, "type": { "tag": ":int" } }, { "tag": "field", "name": "odd_value", "bit-offset": 64, "bit-size": 480, "bit-alignment": 32, "type": { "tag": ":array", "type": { "tag": ":int" }, "size": 15 } }] },
{ "tag": "typedef", "name": "my_point_t", "location": "/home/rpav/test.h:9:3", "type": { "tag": ":struct", "name": "my_point", "id": 0 } },
{ "tag": "enum", "name": "some_values", "id": 0, "location": "/home/rpav/test.h:11:6", "fields": [{ "tag": "field", "name": "a_value", "value": 0 }, { "tag": "field", "name": "another_value", "value": 1 }, { "tag": "field", "name": "yet_another_value", "value": 2 }] },
{ "tag": "function", "name": "do_something", "location": "/home/rpav/test.h:17:6", "variadic": false, "parameters": [{ "tag": "parameter", "name": "p", "type": { "tag": ":pointer", "type": { "tag": "my_point_t" } } }, { "tag": "parameter", "name": "x", "type": { "tag": ":int" } }, { "tag": "parameter", "name": "y", "type": { "tag": ":int" } }], "return-type": { "tag": ":void" } }
]

Because this uses Clang as a parser, the C, C++, or Objective C is fully and correctly parsed.

Building

This requires Clang and LLVM of the appropriate version, which you can obtain from the repository, by download.

You should be able to build c2ffi with out-of-the-box clang-3.7 on your dist. However, see "Notes" below for some things to watch out for.

c2ffi uses cmake. This is relatively easy to use. However, if you built clang++ with special options (e.g., libc++, libc++abi, libcxxrt, etc), see Notes below.

c2ffi/ $ mkdir build
c2ffi/ $ cd build/
build/ $ cmake ..
  :
  : Output
  :
build/ $ make
  :
  : Output
  :
build/ $ ./bin/c2ffi
Usage: c2ffi [options ...] FILE

Options:
      -I, --include        Add a "LOCAL" include path
      -i, --sys-include    Add a <system> include path
      -D, --driver         Specify an output driver (default: json)

      -o, --output         Specify an output file (default: stdout)
      -M, --macro-file     Specify a file for macro definition output

      -N, --namespace      Specify target namespace/package/etc

      -A, --arch           Specify the target triple for LLVM
                           (default: x86_64-unknown-linux-gnu)
      -x, --lang           Specify language (c, c++, objc, objc++)

Drivers: json, sexp, null

Now you have a working c2ffi. If not, see Notes.

Notes

• Packaged clang binaries should now work. But, because these appear to be build with gcc, it is not possible to build c2ffi with clang! So use gcc in this case.

• You need llvm/clang dev libraries. libclang.so is not enough. If you don't have libclangAST.a, you have to install any -dev type packages in your dist. Not all dists may package these.

• If you build with clang and get link errors about ABI functions, you may need to link to -lc++abi or similar.

• If you build with clang and get link errors about random LLVM and Clang functions, you need to build with gcc, because your clang was built with gcc.

• Building on OSX may require specifying LIBCLANG_CPPFLAGS=/usr/local/include or wherever you installed LLVM. And you will have to build LLVM, because Apple's build does not seem to include the appropriate headers or libraries. (Not verified recently.)

• If you're seeing compiler errors, you probably checked out the wrong branch. Verify your clang -v vs your git branch.

Usage

There are generally two steps to using c2ffi:

• Generate output for a particular header or file, gathering macro definitions (with the -M <file>.c parameter)

• Generate output for macro definitions by running c2ffi again on the generated file (without -M)

This is due to the preprocessor being a huge hack (see below). However, once this is done, you should have two files with all the necessary data for your FFI bindings.

Currently JSON is the default output. This is in a rather wordy hierarchical format, with each object having a "tag" field which describes it. All objects are contained in an array. This should make it fairly easy (or at least far easier than parsing C yourself) to transform into language-specific bindings.

This format may be documented at some point, but for now, you'll have to look at the input and the output! I recommend a pretty-printing reformatter for the JSON. Patches to produce prettier output will be accepted. ;-)

Errors

You may encounter errors if the code in question is not correct. Presumably, most of the time, you will be running c2ffi on existing, known-working code.

In this case, the most likely "error" you will encounter will look like this:

Skipping invalid Decl:
FunctionDecl 0x21e05f0 </usr/include/glib-2.0/glib/gmacros.h:328:22, /usr/include/glib-2.0/glib/deprecated/gthread.h:282:65> g_cond_timed_wait 'int (GCond *, GMutex *, GTimeVal *)' extern
|-ParmVarDecl 0x21e0480 <line:280:42, col:58> cond 'GCond *'
|-ParmVarDecl 0x21e04f0 <line:281:42, col:58> mutex 'GMutex *'
`-ParmVarDecl 0x21e0560 <line:282:42, col:58> timeval 'GTimeVal *' invalid

This usually means that Clang didn't find a header, and it doesn't know about one of the types referenced. Look at the top of your error output. Missing header errors will often appear there.

You should specify any necessary additional include paths with -i(for system headers, i.e. those using <brackets>) or -I (for local headers, i.e. those using "quotes").

Generally, any issue relating to an error with C, includes, or the like is not a bug with c2ffi. However, c2ffi should not abort or crash; any such error is certainly a bug with c2ffi.

Language Support

C

C support should be fairly complete. Formerly variadic functions and bitfield support was incomplete. These should now be fully-supported.

Note however that bitfield support is platform- and sometimes compiler-specific; if your platform ABI does not provide a strict definition, expect the layout of structs which use bitfields to be undefined.

C++

C++ support should be fairly complete. This outputs everything as it normally would for C, as well as namespace, classes, methods, and class hierarchy (including base class offsets).

Template support is limited to instantiated templates (including both classes/structs/unions and functions). c2ffi can output a new .hpp file using the -T parameter with explicit instantiations for those it finds declared but not instantiated. E.g.,

template<typename T>
class C { T t; };

typedef class C<int> C_int;

Using c2ffi -T file.T.hpp ..., this will produce the following, which #includes the original to produce a complete definition:

#include "original.cpp"
template class C<int>;

Note: The behavior of this has changed. This used to produce a file which did not include the original. You can now use -D null to output only the .T.hpp file, and then produce full output from that. This simpifies the process.

ObjC

Basic support at least exists. I am not an Objective C person and don't really have a great way to use or test the output, or verify that all the useful features are included.

If you send me example source along with some information about what would be useful, I can try to accommodate. If you write a translator for the JSON to an ObjC bridge, let me know and I will link it below.

ObjC++

Untested.

Importing

Processing the JSON into a usable format is fairly straightforward. Some care must be given to handle anonymous types (e.g., typedef struct { ... } type_t;), but writing these is fairly trivial overall.

The following language bindings exist for c2ffi:

• cl-autowrap: Create bindings in Commonn Lisp from a .h with c2ffi using a simple (c-include "file.h")

• c2ffi-ruby: Uses the JSON from c2ffi to produce a nicely-formatted Ruby file for ruby-ffi.

New Output Drivers

If you're feeling motivated, it should be fairly simple to produce a new output driver. Look in src/drivers/ and you can see the source for JSON, Sexp (lisp symbolic expressions), and possibly some others.

You will need to do the following:

• Create a new subclass of OutputDriver in src/drivers/; copying one of the existing ones is probably the easiest.

• Add this file to src/Makefile.am

• Add the factory function to src/OutputDriver.cpp.

• Write your code!

The Preprocessor

The preprocessor handling is, as was noted, a huge hack. This is due entirely to the fact that #define macros can contain just about anything, and thus it's not easy to tell if they are useful values or syntax hackery.

For this, c2ffi uses a simple heuristic:

• If there are arithmetic operators (+, -, *, <<, etc), parens, numbers, and identifiers, it's treated as "useful".

• If only ints are found, it's treated as an __int128_t; if floats are found, it's treated as a double; if a string is found, a char*

Why the odd __int128_t? Because without more parsing (and technically, without context), it can't be determined as signed or unsigned. So this is declared with very large capacity which will hold the entire range of signed and unsigned 64-bit ints.

If you're dealing with unsigned 128-bit int constants, you'll have to do it yourself. I personally haven't seen any.

License

This is currently GPL2, but it will almost certainly be moved to LGPL2, as I would like to make a shared library which you can load definitions at runtime. It may be moved to BSD or similar at some point in the future.