Work in progress!

Grako++

Grako++ is a packrat parser runtime for Grako, written in C++. The aim is to provide a Grako-compatible parser generator, but with improved performance over Python.

Differences to Grako

• The output is AST/JSON, but extension types are available.
• The semantics class can be implemented in C++, Cython or pure Python.
• The regular expression syntax is ECMAScript, not Python.
• Some features of Grako are missing, see below in the TODO section.
• State types must implement operator< for std::map (for future flexibility, consider implementing the hash trait, too).

Build

The library is header-files only currently, so you can just copy the files in include/ to a convenient location, or build from there. More formally, you can use cmake.

Build with clang by invoking: cmake -DCMAKE_CXX_COMPILER=/path/to/clang++

Building with g++ < 4.9 requires linking with the boost_regex library, because the std::regex implementation is broken. You can rely on the exported cmake library file to configure the right settings.

Other useful option: -DCMAKE_INSTALL_PREFIX:PATH=/path/to/install

Usage

The grakopp program is used like grako to compile PEG files to source code. There are four different source code output formats that can be specified with the -f/--format option:

--format	--output	Purpose
hpp	_name.hpp	C++ declaration
cpp	_name.cpp	C++ implementation
pxd	name.pxd	Cython declaration
pyx	name.pyx	Cython implementation

For pure C++ parsers, generating the hpp and cpp files is sufficient. For Python integration, the pxd and pyx files are also needed. For "name" you should subsitute the actual name of the parser (either the base name of the PEG file or the argument to the --name option). The filenames are, for now, hard-coded into the source files (the underscore protects the C++ implementation from Cython-generated source files).

Here is an example how to build a parser:

$ ./grakopp -f hpp -o _basic.hpp tests/basic/basic.peg
$ ./grakopp --whitespace="" --no-nameguard -f cpp -o _basic.cpp tests/basic/basic.peg
$ g++ -DGRAKOPP_MAIN -std=c++11 -Iinclude -O4 -o basic _basic.cpp -lboost_regex

You can invoke the parser like a Python parser generated with grako (currently, no options are supported, though, except for an internal option --test that compares the AST with a text fixture file):

$ echo -n e1e2 | ./basic /dev/stdin sequence
[
    "e1",
    "e2"
]

C++ Interface

The following header files exist:

Header	Purpose
grakopp/exceptions.hpp	Parser exceptions
grakopp/buffer.hpp	Buffer for I/O
grakopp/ast.hpp	AST implementation
grakopp/parser.hpp	Parser base class
grakopp/grakopp.hpp	Include all above
grakopp/ast-io.hpp	Optional AST stream I/O

Python Integration

You can build the Python package with setup.py in the normal way, for example:

$ python setup.py build_ext --inplace

To continue the above example:

$ ./grakopp -f pxd -o basic.pxd tests/basic/basic.peg
$ ./grakopp -f pyx -o basic.pyx tests/basic/basic.peg
$ cython -Ipython --cplus basic.pyx
$ g++ -std=c++11 -Iinclude -shared -pthread -fPIC -fwrapv -O2 -Wall -fno-strict-aliasing -I/usr/include/python2.7 -o basic.so basic.cpp _basic.cpp -l boost_regex

You can then use it from Python, see basic-demo.py:

$ PYTHONPATH=python python basic-demo.py
_sequence_ ['e1', 'e2']
['e1', 'e2']

TODO

• Convert more semantics exceptions (beside FailedSemantics) to corresponding AST objects in exc_to_ast.
• python/distutils integration a la cythonize
• automatic compilation a la pyximport
• add namespace
• unicode support?
• regex syntax tests (make sure generated C strings are always proper)
• profile and optimize
• documentation

Grako features missing:

• ignorecase (buffer match, matchre)
• comments skipping
• buffer line parsing and trace output (also in exceptions, and then use grako.exceptions)
• ParseInfo
• rules with arguments
• left recursion
• semantic action "_default" (possible in the WrappedSemantics)

Authors

Copyright (C) 2014 semantics Kommunikationsmanagement GmbH
Written by Marcus Brinkmann <m.brinkmann@semantics.de>
See LICENSE.txt for details.