biojppm / regen

regen

regen is a python3 package providing C/C++ reflection and source-code generation. You provide your own code generation templates , and have full control over where the generated code goes. A flexible system of annotations is used, so that you can pass meta-values to your code generation templates.

regen is ideal to be used as a pre-build step for your C/C++ project, but it can also be used separately on a file by file basis. Examples of application are object-tree iteration utilities, or property systems with arbitrary per-property annotations, or maintenance-free enum strings.

This is a very fresh pre-alpha project. It is buggy and its interface will change.

How it works

regen receives two inputs: your C/C++ source code and the code generation templates. The code generation templates are written for Python's jinja2 template engine. regen uses libclang to parse your C/C++ code and produce an Abstract Syntax Tree (AST). This tree is processed to extract the features, which are then passed to each generator.

You can see regen being used in the project's examples folder.

Quick examples

Enum stringification

Consider a C++ header, let's name it myenum.h:

#pragma once

C4_ENUM()
typedef enum {
    FOO,
    BAR,
    BAZ
} MyEnum;

Now use the following python code for parsing and generating, saved as regen.py:

import c4.regen as regen

egen = regen.EnumGenerator(
    # extract enums tagged with this macro
    tag="C4_ENUM",
    # header preamble
    hdrp="""\
#include "enum_pairs.h"
"""
    # template for code in header files
    hdr="""\
template<> const EnumPairs< {{enum.type}} > enum_pairs();
""",
    # template for code in source files
    src="""\
template<> const EnumPairs< {{enum.type}} > enum_pairs()
{
    static const EnumAndName< {{enum.type}} > vals[] = {
        {% for e in enum.symbols %}
        { {{e.name}}, "{{e.name}}"},
        {% endfor %}
    };
    EnumPairs< {{enum.type}} > r(vals);
    return r;
}
"""
)
writer = regen.ChunkWriterGenFile()

#------------------------------------------------------------------------------
if __name__ == "__main__":
    regen.run(writer, egen, [])

Now run regen to parse the source code and generate your code:

python regen.py myenum.h

The command above generates myenum.gen.h:

#ifndef _MYENUM_GEN_H_
#define _MYENUM_GEN_H_

#include "enum_pairs.h"
#include "myenum.h"

template<> const EnumPairs< MyEnum > enum_pairs();
#endif // _MYENUM_GEN_H_

and also myenum.gen.cpp:

#include "myenum.gen.h"

template<> const EnumPairs< MyEnum > enum_pairs()
{
    static const EnumAndName< MyEnum > vals[] = {
        { FOO, "FOO"},
        { BAR, "BAR"},
        { BAZ, "BAZ"},
    };
    EnumPairs< MyEnum > r(vals);
    return r;
}

Running

Finding libclang

regen uses libclang-py3, which is a python wrapper for the libclang library. The current version of libclang-py3 requires libclang 3.8. regen tries to find libclang 3.8 by querying llvm-config --libdir (if llvm-config --version reports 3.8) or llvm-config-3.8 --libdir if the first fails. If this also fails, then you can still use the option --clang-libdir.

(This version dependency needs to be fixed; this will probably be done by using different branches).

libclang on windows

libclang is hard to use on windows, but it is useable. While its rough edges are rounded out by the clang developers, we need to deal with its windows problems:

• The official installer for version 3.8.1 on the LLVM site is broken with VS2015 Update 3, so it won't work out of the box when the C++ library is used. It needs to be compiled from source and patched (AFAIK there's no 3.8.2 release).
• clang 3.9.1 needs to be run with the Visual Studio developer environment, or it will cause a linker error (no kernel32).
• Use of the flag -fms-compatibility-version=19 is required (even after compiling).

For this and other reasons it is sometimes good to compile clang from source. To make this task easier, regen has a clang build project, which downloads the source code from llvm, clang and extra tools, patches it as needed, compiles and installs. You can use it like this:

cd regen/tools/clang-build
mkdir build
cd build
cmake -DCLANG_VERSION=3.8.1 ..
cmake --build --config Release .

You can compile several versions at once. For example, to compile versions 3.8.1, 3.9.1 and 4.0.0 in a single swoop, you can configure with this command instead:

cmake -DCLANG_VERSION="3.8.1;3.9.1;4.0.0" ..

Installing

From PyPi

regen installation is easy with the Python package repository. This will install regen along with its dependencies:

pip install regen

From source

git clone https://github.com/biojppm/regen.git
cd regen
pip install .

For development

Setting up regen for development is easy:

git clone https://github.com/biojppm/regen.git
cd regen
pip install -r requirements_dev.txt
pip install -e .

*Windows notes*. The examples rely extensively on symbolic link files. This works as expected in Unix and Mac, but symbolic links were only recently introduced in Windows. Git already allows you to use symbolic links in Windows, but the process is convoluted. Before cloning the repo, you must first enable symlinks in windows. Then you need to pass an option to git clone to ensure that the files are really symbolic links. The clone command thus needs to be:

git clone -c core.symlinks=true https://github.com/biojppm/regen.git

License

cmany is permissively licensed under the MIT license.