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 - 3.5.
To use the configparser
backport instead of the built-in version on both
Python 2 and Python 3, simply import it explicitly as a backport:
from backports import configparser
If you'd like to use the backport on Python 2 and the built-in version on Python 3, use that invocation instead:
import configparser
For detailed documentation consult the vanilla version at http://docs.python.org/3/library/configparser.html.
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 oldSafeConfigParser
with a bunch of tweaks which make it more predictable for users. Don't need interpolation? Simply useConfigParser(interpolation=None)
, no need to use a distinctRawConfigParser
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 (calledBasicInterpolation
) - a new
${buildout:like}
syntax (calledExtendedInterpolation
)
- the classic
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 anotherConfigParser
instance, thanks to its mapping protocol support.many smaller tweaks, updates and fixes
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.
This project uses semver to communicate the impact of various releases while periodically syncing with the upstream implementation in CPython. The changelog serves as a reference indicating which versions incorporate which upstream functionality.
Prior to the 4.0.0
release, another scheme
was used to associate the CPython and backports releases.
This backport was originally authored by Łukasz Langa, the current vanilla
configparser
maintainer for CPython and is currently maintained by
Jason R. Coombs:
To report a security vulnerability, please use the Tidelift security contact. Tidelift will coordinate the fix and disclosure.
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 python-future. The project takes 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
master
branch holds a version of the3.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 but will also work on Python 2.6 and 2.7 (including PyPy). You can check this running the supplied unit tests withtox
.
The process works like this:
- In the
3.x
branch, runpip-run -- sync-upstream.py
, which downloads the latest stable release of Python and copies the relevant files from there into their new locations here and then commits those changes with a nice reference to the relevant upstream commit hash. - I check for new names in
__all__
and update imports inconfigparser.py
accordingly. I run the tests on Python 3. Commit. - I merge the new commit to
master
. I runtox
. Commit. - If there are necessary changes, I do them now (on
master
). Note that the changes should be written in the syntax subset supported by Python 2.6. - I run
tox
. If it works, I update the docs and release the new version. Otherwise, I go back to point 3. I might usepasteurize
to suggest me required changes but usually I do them manually to keep resulting code in a nicer form.
[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. |