libconf

libconf is a pure-Python reader/writer for configuration files in libconfig format, which is often used in C/C++ projects. It's interface is similar to the json module: the four main methods are load(), loads(), dump(), and dumps().

Example usage:

import io, libconf
>>> with io.open('example.cfg') as f:
...     config = libconf.load(f)
>>> config
{'capabilities': {'can-do-arrays': [3, 'yes', True],
                  'can-do-lists': (True,
                                   14880,
                                   ('sublist',),
                                   {'subgroup': 'ok'})},
 'version': 7,
 'window': {'position': {'h': 600, 'w': 800, 'x': 375, 'y': 210},
            'title': 'libconfig example'}}

>>> config['window']['title']
'libconfig example'
>>> config.window.title
'libconfig example'

>>> print(libconf.dumps({'size': [10, 15], 'flag': True}))
flag = True;
size =
[
    10,
    15
];

The data can be accessed either via indexing (['title']) or via attribute access .title.

Character encoding and escape sequences

The recommended way to use libconf is with Unicode objects (unicode on Python2, str on Python3). Input strings or streams for load() and loads() should be Unicode, as should be all strings contained in data structures passed to dump() and dumps().

In load() and loads(), escape sequences (such as \n, \r, \t, or \xNN) are decoded. Hex escapes (\xNN) are mapped to Unicode characters U+0000 through U+00FF. All other characters are passed though as-is.

In dump() and dumps(), unprintable characters below U+0080 are escaped as \n, \r, \t, \f, or \xNN sequences. Characters U+0080 and above are passed through as-is.

Writing libconfig files

Reading libconfig files is easy. Writing is made harder by two factors:

• libconfig's distinction between int and int64: 2 vs. 2L
• libconfig's distinction between lists and arrays, and the limitations on arrays

The first point concerns writing Python int values. Libconf dumps values that fit within the C/C++ 32bit int range without an "L" suffix. For larger values, an "L" suffix is automatically added. To force the addition of an "L" suffix even for numbers within the 32 bit integer range, wrap the integer in a LibconfInt64 class.

Examples:

dumps({'value': 2})                # Returns "value = 2;"
dumps({'value': 2**32})            # Returns "value = 4294967296L;"
dumps({'value': LibconfInt64(2)})  # Explicit int64, returns "value = 2L;"

The second complication arises from distinction between lists and arrays in the libconfig language. Lists are enclosed by () parenthesis, and can contain arbitrary values within them. Arrays are enclosed by [] brackets, and have significant limitations: all values must be scalar (int, float, bool, string) and must be of the same type.

Libconf uses the following convention:

• it maps libconfig ()-lists to Python tuples, which also use the () syntax.
• it maps libconfig []-arrays to Python lists, which also use the [] syntax.

This provides nice symmetry between the two languages, but has the drawback that dumping Python lists inherits the limitations of libconfig's arrays. To explicitly control whether lists or arrays are dumped, wrap the Python list/tuple in a LibconfList or LibconfArray.

Examples:

# Libconfig lists (=Python tuples) can contain arbitrary complex types:
dumps({'libconf_list': (1, True, {})})

# Libconfig arrays (=Python lists) must contain scalars of the same type:
dumps({'libconf_array': [1, 2, 3]})

# Equivalent, but more explit by using LibconfList/LibconfArray:
dumps({'libconf_list': LibconfList([1, True, {}])})
dumps({'libconf_array': LibconfArray([1, 2, 3])})

Comparison to other Python libconfig libraries

Pylibconfig2 is another pure-Python libconfig reader. It's API is based on the C++ interface, instead of the Python json module. It's licensed under GPLv3, which makes it unsuitable for use in a large number of projects.

Python-libconfig is a library that provides Python bindings for the libconfig++ C++ library. While permissively licensed (BSD), it requires a compilation step upon installation, which can be a drawback.

I wrote libconf (this library) because both of the existing libraries didn't fit my requirements. I had a work-related project which is not open source (ruling out pylibconfig2) and I didn't want the deployment headache of python-libconfig. Further, I enjoy writing parsers and this seemed like a nice opportunity :-)

Release notes

• 2.0.1, released on 2019-11-21
  • Allow trailing commas in lists and arrays for improved compatibility with the libconfig C implementation. Thanks to nibua-r for reporting this issue.
• 2.0.0, released on 2018-11-23
  • Output validation for dump() and dumps(): raise an exception when dumping data that can not be read by the C libconfig implementation. This change may raise exceptions on code that worked with <2.0.0!
  • Add LibconfList, LibconfArray, LibconfInt64 classes for more fine-grained control of the dump()/dumps() output.
  • Fix deepcopy() of AttrDict classes (thanks AnandTella).
• 1.0.1, released on 2017-01-06
  • Drastically improve performance when reading larger files
  • Several smaller improvements and fixes
• 1.0.0, released on 2016-10-26:
  • Add the ability to write libconf files (dump() and dumps(), thanks clarkli86 and eatsan)
  • Several smaller improvements and fixes
• 0.9.2, released on 2016-09-09:
  • Fix compatibility with Python versions older than 2.7.6 (thanks AnandTella)