/clif

Wrapper generator foundation to wrap C++ for Python and other languages using LLVM.

Primary LanguageC++Apache License 2.0Apache-2.0

C++ Language Interface Foundation (CLIF)

CLIF provides a common foundation for creating C++ wrapper generators for various languages.

Overview

It consists of four parts:

  1. Parser
  2. Matcher
  3. Generator
  4. Runtime

Parser

The parser converts a language-friendly C++ API description to the language-agnostic internal format and passes it to the Matcher.

Matcher

The matcher parses selected C++ headers with Clang (LLVM's C++ compiler) and collects type information. That info is passed to the Generator.

Generator

The generator emits C++ source code for the wrapper.

The generated wrapper needs to be built according with language extension rules. Usually that wrapper will call into the Runtime.

Runtime

The runtime C++ library holds type conversion routines that are specific to each target language but are the same for every generated wrapper.

Python CLIF

See complete implementation of a Python wrapper generator in the /python/ subdirectory. Both Python 2 and 3 are supported.

Installation

Prerequisites

  1. We use CMake, so make sure CMake version 3.5 or later is available. (For example, Debian 8 only has version 3.0, so in that case you'll need to install an up-to-date CMake.)

  2. We use Google protobuf for inter-process communication between the CLIF frontend and backend. Version 3.2.0 or later is required. Please install protobuf for both C++ and Python from source, as we will need some protobuf source code later on.

  3. You must have virtualenv installed.

  4. You must have Subversion installed, so we can fetch LLVM.

  5. You must have pyparsing installed, so we can build protobuf. Use pip install 'pyparsing>=2.2.0' to fetch the correct version.

  6. Make sure pkg-config --libs python works (e.g. install python-dev and pkg-config).

Building

The steps below are in INSTALL.sh but outlined here for clarification. The install script sets up a Python virtual environment where it installs CLIF.

  1. Checkout LLVM and Clang source trees (the exact SVN version as specified here is required)

    # We keep it separate of the CLIF tree to avoid pip unwanted copying.
    mkdir $LLVMSRC_DIR
    cd $LLVMSRC_DIR
    svn co https://llvm.org/svn/llvm-project/llvm/trunk@307315 llvm
    cd llvm/tools
    svn co https://llvm.org/svn/llvm-project/cfe/trunk@307315 clang
    ln -sf "$CLIFSRC_DIR/clif" clif
  2. Build and install the CLIF backend. If you use Ninja instead of make your build will go significantly faster. It is used by Chromium, LLVM et al. Look at INSTALL.sh for the directory setup and proper ...flags... to supply the cmake command here:

    # Builds must be done outside of the LLVM tree.
    mkdir ../../build_matcher
    cd ../../build_matcher
    cmake ...flags... $LLVMSRC_DIR/llvm
    make clif-matcher
    make install

    Replace the cmake and make commands with these to use Ninja:

    cmake -G Ninja ...flags... $LLVMSRC_DIR/llvm
    ninja clif-matcher
    ninja -j 2 install

    If the clif-matcher build target is not found, check that you created the correct llvm/tools/clif symlink in the previous step. The CLIF backend builds as part of an LLVM build.

    If you have more than one Python version installed (eg. python2.7 and python3.6) cmake may have problems finding python libraries for the Python you specified as INSTALL.sh argument and uses the default Python instead. To help cmake use the correct Python add the following options to the cmake command (substitute the correct path for your system):

    cmake ... \
      -DPYTHON_INCLUDE_DIR="/usr/include/python3.6" \
      -DPYTHON_LIBRARY="/usr/lib/x86_64-linux-gnu/libpython3.6m.so" \
      -DPYTHON_EXECUTABLE="/usr/bin/python3.6" \
      "${CMAKE_G_FLAGS[@]}" "$LLVM_DIR/llvm"

    NOTE: INSTALL.sh builds only for X86. If you want to build for another architecture, modify it to specify your target architecture, or just remove this restriction (see NOTE in INSTALL.sh).

  3. Get back to your CLIF python checkout and install it using pip.

    cd "$CLIFSRC_DIR"
    cp "$BUILD_DIR/tools/clif/protos/ast_pb2.py" clif/protos/
    pip install .

That version is guaranteed to work. Older versions likely do not work (they lack some APIs); later versions might work, at your own risk.

INSTALL.sh will build and install CLIF for Python (and LLVM Clang as an internal tool) to your system by default in $HOME/opt/clif and $HOME/opt/clif/clang.

To run Python CLIF use $HOME/opt/clif/bin/pyclif.

Using your newly built pyclif

First, try some examples:

cd ~/opt/clif/examples
less README.md

Next, read the Python CLIF User Manual.

For more details please refer to:

  1. CLIF Python Primer
  2. CLIF FAQ

Disclaimer

This is not an official Google product.