Julia interface to libclang and C wrapper generator

First Commit


Last Touched

25 days ago

Commit Count

415 commits



Clang.jl provides a Julia language wrapper for libclang: the stable, C-exported interface to (a subset of) the LLVM Clang compiler. The libclang API documentation provides background on the functionality available through libclang, and thus through the Julia wrapper. The Clang.jl repository also hosts related tools built on top of libclang functionality.

If you are unfamiliar with the Clang AST and plan to access the internals of this library (as opposed to accessing the 'wrap-c' bindings), a good starting point is the Introduction to the Clang AST.

Getting Started

Installation requirements are listed down page.

Clang.jl Documentation

Example notebooks:

Usage Examples

Parsing C with Clang and Julia

C wrapper generator

Clang.jl includes a generator to create Julia wrappers for C libraries from a collection of header files. The following declarations are currently supported:

  • function: translated to Julia ccall, with full type translation
  • typedef: translated to Julia typealias to underlying intrinsic type
  • enum: translated to const value symbol

  • struct: partial struct support may be enabled by setting WrapContext.options.wrap_structs = true


Clang.jl tends to be used on large codebases, often with multiple API versions to support. Building a generator requires some customization effort, so for small libraries the initial investment may not pay off.

  • Gtk.jl: Julia Gtk bindings (uses heavily customized generator)
  • CUDArt.jl: Bindings to the CUDA Runtime library
  • CUFFT.jl: Bindings to CUDA FFT library
  • Sundials.jl: interface to the LLNL "SUite of Nonlinear and DIfferential/ALgebraic equation Solvers" package
  • libCURL.jl: wrapper of cURL
  • VideoIO.jl: Julia bindings for libav/ffmpeg
  • WCSLIB.jl
  • NIDAQ.jl: wrapper for NIDAQ data acquisition library



libclang and libclang-dev (versions >= 3.1) are required. libclang may optionally be built by the Julia build system by setting BUILD_LLVM_CLANG=1 in julia/Make.user. You might also install clang manually, e.g. via sudo apt-get install clang-3.7.

When using a distribution packaged version, note the latest version may only be available from an optional pre-release archive (e.g. Ubuntu PPA).


To install using the Julia package manager, use: Pkg.add("Clang"). Note that you first need to make sure, that the following are in your PATH variable: llvm-config and clang. If not present in a default Julia location, libclang needs to be in PATH (on Windows), or LD_LIBRARY_PATH (Linux), or DYLD_LIBRARY_PATH (OS X).

Background Information

libclang wrapper status

The cindex.jl libclang wrapper encompasses most CXType, CXCursor, and C++ related functions. A small convenience API is provided for basic tasks related to index generation, and retrieval of cursor kind and type information.

All testing and development so far has been on Ubuntu 12.04.1 (64-bit)

cindex.jl Usage

  using Clang.cindex

  julia> topcu = cindex.parse_header("Index.h") # Parse Index.h, returning the root cursor
  julia> children(topcu).size

  julia> cu = cindex.search(top, "clang_parseTranslationUnit")[1]
  julia> cu_kind(cu) == CIndex.CurKind.FUNCTIONDECL
  julia> name(cu)
  "clang_parseTranslationUnit(CXIndex, const char *, const char *const *, int, struct CXUnsavedFile *, unsigned int, unsigned int)"

  julia> tkl = tokenize(cu)
  julia> for tk in tkl
         println(tk) # use tk.text to get underlying string

See the examples/ and util/ folders for further usage scenarios. Clang.jl is partially self-generating, including parsing of the enums (util/genCIndex_h.jl)

There is a small convenience API exported by CIndex:

tu_init     # init and parse file to TranslationUnit
tu_cursor   # get top cursor from TranslationUnit
children    # gets CXCursor children; accessed by ref.
cu_kind     # CIndex.CurKind enum
ty_kind     # CIndex.TypKind enum
name        # get the DisplayName of a CXCursor
spelling    # get spelling of a CXCursor or CXType
cu_file     # file in which cursor was declared
is_function # CXCursor or CXType
is_null     # CXCursor
tokenize    # get tokens underlying given cursor

For documentation of wrapped CIndex functions, see:


For a nice introduction to some capabilities and limitations (fewer now) of the libclang API, see this post:



Most libclang functions accept and return small structs by value. As Julia does not (yet) have full struct support, the current Clang.jl implementation includes a C++ wrapper. clang functions are wrapped by a C++ function that memcpys data to/from a Julia-owned memory array (passed by pointer)

see src/cindex_base.jl and deps/src/wrapclang.h for generated wrappers.

Some functions are manually wrapped for convenience and ease of implementation, or to provide a more Julia-friendly API.

see src/Clang.jl and deps/src/wrapclang.cpp


Clang.jl is licensed under the MIT license.