configparser

The ancient ConfigParser module available in the standard library 2.x has seen a major update in Python 3.2. This is a backport of those changes so that they can be used directly in Python 2.6 - 2.7.

To use configparser instead of ConfigParser, simply replace:

import ConfigParser

with:

import configparser

For detailed documentation consult the vanilla version at http://docs.python.org/3/library/configparser.html.

Why you’ll love configparser

Whereas almost completely compatible with its older brother, configparser sports a bunch of interesting new features:

• full mapping protocol access (more info):

  >>> parser = ConfigParser()
  >>> parser.read_string("""
  [DEFAULT]
  location = upper left
  visible = yes
  editable = no
  color = blue
  
  [main]
  title = Main Menu
  color = green
  
  [options]
  title = Options
  """)
  >>> parser['main']['color']
  'green'
  >>> parser['main']['editable']
  'no'
  >>> section = parser['options']
  >>> section['title']
  'Options'
  >>> section['title'] = 'Options (editable: %(editable)s)'
  >>> section['title']
  'Options (editable: no)'
  

• there’s now one default ConfigParser class, which basically is the old SafeConfigParser with a bunch of tweaks which make it more predictable for users. Don’t need interpolation? Simply use ConfigParser(interpolation=None), no need to use a distinct RawConfigParser anymore.

• the parser is highly customizable upon instantiation supporting things like changing option delimiters, comment characters, the name of the DEFAULT section, the interpolation syntax, etc.

• you can easily create your own interpolation syntax but there are two powerful implementations built-in (more info):

  • the classic %(string-like)s syntax (called BasicInterpolation)
  • a new ${buildout:like} syntax (called ExtendedInterpolation)

• fallback values may be specified in getters (more info):

  >>> config.get('closet', 'monster',
  ...            fallback='No such things as monsters')
  'No such things as monsters'
  

• ConfigParser objects can now read data directly from strings and from dictionaries. That means importing configuration from JSON or specifying default values for the whole configuration (multiple sections) is now a single line of code. Same goes for copying data from another ConfigParser instance, thanks to its mapping protocol support.

• many smaller tweaks, updates and fixes

A few words about Unicode

configparser comes from Python 3 and as such it works well with Unicode. The library is generally cleaned up in terms of internal data storage and reading/writing files. There are a couple of incompatibilities with the old ConfigParser due to that. However, the work required to migrate is well worth it as it shows the issues that would likely come up during migration of your project to Python 3.

The design assumes that Unicode strings are used whenever possible [1]. That gives you the certainty that what’s stored in a configuration object is text. Once your configuration is read, the rest of your application doesn’t have to deal with encoding issues. All you have is text [2]. The only two phases when you should explicitly state encoding is when you either read from an external source (e.g. a file) or write back.

Versioning

This backport is intended to keep 100% compatibility with the vanilla release in Python 3.2+. To help maintaining a version you want and expect, a versioning scheme is used where:

• the first three numbers indicate the version of Python 3.x from which the backport is done
• a backport release number is provided after the r letter

For example, 3.3.0r1 is the first release of configparser compatible with the library found in Python 3.3.0.

A single exception from the 100% compatibility principle is that bugs fixed before releasing another minor Python 3.x.y version will be included in the backport releases done in the mean time. This rule applies to bugs only.

Maintenance

This backport is maintained on BitBucket by Łukasz Langa, the current vanilla configparser maintainer for CPython:

• configparser Mercurial repository
• configparser issue tracker

Change Log

3.3.0r2

• updated the fix for #16820: parsers now preserve section order when using __setitem__ and update

3.3.0r1

• compatible with 3.3.0 + fixes for #15803 and #16820
• fixes BitBucket issue #4: read() properly treats a bytestring argument as a filename
• ordereddict dependency required only for Python 2.6
• unittest2 explicit dependency dropped. If you want to test the release, add unittest2 on your own.

3.2.0r3

• proper Python 2.6 support
  • explicitly stated the dependency on ordereddict
  • numbered all formatting braces in strings
• explicitly says that Python 2.5 support won’t happen (too much work necessary without abstract base classes, string formatters, the io library, etc.)
• some healthy advertising in the README

3.2.0r2

• a backport-specific change: for convenience and basic compatibility with the old ConfigParser, bytestrings are now accepted as section names, options and values. Those strings are still converted to Unicode for internal storage so in any case when such conversion is not possible (using the ‘ascii’ codec), UnicodeDecodeError is raised.

3.2.0r1

• the first public release compatible with 3.2.0 + fixes for #11324, #11670 and #11858.

Conversion Process

This section is technical and should bother you only if you are wondering how this backport is produced. If the implementation details of this backport are not important for you, feel free to ignore the following content.

configparser is converted using 3to2. Because a fully automatic conversion was not doable, I took the following branching approach:

• the 3.x branch holds unchanged files synchronized from the upstream CPython repository. The synchronization is currently done by manually copying the required files and stating from which CPython changeset they come from.
• the 3.x-clean branch holds a version of the 3.x code with some tweaks that make it independent from libraries and constructions unavailable on 2.x. Code on this branch still must work on the corresponding Python 3.x. You can check this running the supplied unit tests.
• the default branch holds necessary changes which break unit tests on Python 3.2. Additional files which are used by the backport are also stored here.

The process works like this:

1. I update the 3.x branch with new versions of files. Commit.
2. I merge the new commit to 3.x-clean. Check unit tests. Commit.
3. If there are necessary changes that can be made in a 3.x compatible manner, I do them now (still on 3.x-clean), check unit tests and commit. If I’m not yet aware of any, no problem.
4. I merge the changes from 3.x-clean to default. Commit.
5. If there are necessary changes that cannot be made in a 3.x compatible manner, I do them now (on default). Note that the changes should still be written using 3.x syntax. If I’m not yet aware of any required changes, no problem.
6. I run ./convert.py which is a custom 3to2 runner for this project.
7. I run the unit tests with unittest2 on Python 2.x. If the tests are OK, I can prepare a new release. Otherwise, I revert the default branch to its previous state (hg revert .) and go back to Step 3.

NOTE: the default branch holds unconverted code. This is because keeping the conversion step as the last (after any custom changes) helps managing the history better. Plus, the merges are nicer and updates of the converter software don’t create nasty conflicts in the repository.

This process works well but if you have any tips on how to make it simpler and faster, do enlighten me :)

Footnotes

[1]

To somewhat ease migration, passing bytestrings is still supported but they are converted to Unicode for internal storage anyway. This means that for the vast majority of strings used in configuration files, it won’t matter if you pass them as bytestrings or Unicode. However, if you pass a bytestring that cannot be converted to Unicode using the naive ASCII codec, a UnicodeDecodeError will be raised. This is purposeful and helps you manage proper encoding for all content you store in memory, read from various sources and write back.

[2]

Life gets much easier when you understand that you basically manage text in your application. You don’t care about bytes but about letters. In that regard the concept of content encoding is meaningless. The only time when you deal with raw bytes is when you write the data to a file. Then you have to specify how your text should be encoded. On the other end, to get meaningful text from a file, the application reading it has to know which encoding was used during its creation. But once the bytes are read and properly decoded, all you have is text. This is especially powerful when you start interacting with multiple data sources. Even if each of them uses a different encoding, inside your application data is held in abstract text form. You can program your business logic without worrying about which data came from which source. You can freely exchange the data you store between sources. Only reading/writing files requires encoding your text to bytes.