Unicode handling

This document explains how Unicode and related issues are handled in CKAN. For a general introduction to Unicode and Unicode handling in Python 2 please read the Python 2 Unicode HOWTO. Since Unicode handling differs greatly between Python 2 and Python 3 you might also be interested in the Python 3 Unicode HOWTO.

CKAN uses the six module to provide simultaneous compatibility with Python 2 and Python 3. All strs are Unicode in Python 3 so the builtins unicode and basestring have been removed so there are a few general rules to follow:

  1. Change all calls to basestring() into calls to six.string_types()

  2. Change remaining instances of basestring to six.string_types

  3. Change all instances of (str, unicode) to six.string_types

  4. Change all calls to unicode() into calls to six.text_type()

  5. Change remaining instances of unicode to six.text_type

These rules do not apply in every instance so some thought needs to be given about the context around these changes.

Note

This document describes the intended future state of Unicode handling in CKAN. For historic reasons, some existing code does not yet follow the rules described here.

New code should always comply with the rules in this document. Exceptions must be documented.

Overall Strategy

CKAN only uses Unicode internally (six.text_type on both Python 2 and Python 3). Conversion to/from ASCII strings happens on the boundary to other systems/libraries if necessary.

Encoding of Python files

Files containing Python source code (*.py) must be encoded using UTF-8, and the encoding must be declared using the following header:

# encoding: utf-8

This line must be the first or second line in the file. See PEP 263 for details.

String literals

String literals are string values given directly in the source code (as opposed to strings variables read from a file, received via argument, etc.). In Python 2, string literals by default have type str. They can be changed to unicode by adding a u prefix. In addition, the b prefix can be used to explicitly mark a literal as str:

x = "I'm a str literal"
y = u"I'm a unicode literal"
z = b"I'm also a str literal"

In Python 3, all str are Unicode and str and bytes are explicitly different data types so:

x = "I'm a str literal"
y = u"I'm also a str literal"
z = b"I'm a bytes literal"

In CKAN, every string literal must carry either a u or a b prefix. While the latter is redundant in Python 2, it makes the developer’s intention explicit and eases a future migration to Python 3.

This rule also holds for raw strings, which are created using an r prefix. Simply use ur instead:

m = re.match(ur'A\s+Unicode\s+pattern')

For more information on string prefixes please refer to the Python documentation.

Note

The unicode_literals future statement is not used in CKAN.

Best Practices

Use io.open to open text files

When opening text (not binary) files you should use io.open instead of open. This allows you to specify the file’s encoding and reads will return Unicode instead of ASCII:

import io

with io.open(u'my_file.txt', u'r', encoding=u'utf-8') as f:
    text = f.read()  # contents is automatically decoded
                     # to Unicode using UTF-8

Text files should be encoded using UTF-8 if possible.

Normalize strings before comparing them

For many characters, Unicode offers multiple descriptions. For example, a small latin e with an acute accent (é) can either be specified using its dedicated code point (U+00E9) or by combining the code points for e (U+0065) and the accent (U+0301). Both variants will look the same but are different from a numerical point of view:

>>> x = u'\N{LATIN SMALL LETTER E WITH ACUTE}'
>>> y = u'\N{LATIN SMALL LETTER E}\N{COMBINING ACUTE ACCENT}'
>>> print x, y
é é
>>> print repr(x), repr(y)
u'\xe9' u'e\u0301'
>>> x == y
False

Therefore, if you want to compare two Unicode strings based on their characters you need to normalize them first using unicodedata.normalize:

>>> from unicodedata import normalize
>>> x_norm = normalize(u'NFC', x)
>>> y_norm = normalize(u'NFC', y)
>>> print x_norm, y_norm
é é
>>> print repr(x_norm), repr(y_norm)
u'\xe9' u'\xe9'
>>> x_norm == y_norm
True

Use the Unicode flag in regular expressions

By default, the character classes of Python’s re module (\w, \d, …) only match ASCII-characters. For example, \w (alphanumeric character) does, by default, not match ö:

>>> print re.match(ur'^\w$', u'ö')
None

Therefore, you need to explicitly activate Unicode mode by passing the re.U flag:

>>> print re.match(ur'^\w$', u'ö', re.U)
<_sre.SRE_Match object at 0xb60ea2f8>

Note

Some functions (e.g. re.split and re.sub) take additional optional parameters before the flags, so you should pass the flag via a keyword argument:

replaced = re.sub(ur'\W', u'_', original, flags=re.U)

The type of the values returned by re.split, re.MatchObject.group, etc. depends on the type of the input string:

>>> re.split(ur'\W+', b'Just a string!', flags=re.U)
['Just', 'a', 'string', '']

>>> re.split(ur'\W+', u'Just some Unicode!', flags=re.U)
[u'Just', u'some', u'Unicode', u'']

Note that the type of the pattern string does not influence the return type.

Filenames

Like all other strings, filenames should be stored as Unicode strings internally. However, some filesystem operations return or expect byte strings, so filenames have to be encoded/decoded appropriately. Unfortunately, different operating systems use different encodings for their filenames, and on some of them (e.g. Linux) the file system encoding is even configurable by the user.

To make decoding and encoding of filenames easier, the ckan.lib.io module therefore contains the functions decode_path and encode_path, which automatically use the correct encoding:

import io
import json

from ckan.lib.io import decode_path

# __file__ is a byte string, so we decode it
MODULE_FILE = decode_path(__file__)
print(u'Running from ' + MODULE_FILE)

# The functions in os.path return unicode if given unicode
MODULE_DIR = os.path.dirname(MODULE_FILE)
DATA_FILE = os.path.join(MODULE_DIR, u'data.json')

# Most of Python's built-in I/O-functions accept Unicode filenames as input
# and encode them automatically
with io.open(DATA_FILE, encoding='utf-8') as f:
    data = json.load(f)

Note that almost all Python’s built-in I/O-functions accept Unicode filenames as input and encode them automatically, so using encode_path is usually not necessary.

The return type of some of Python’s I/O-functions (e.g. os.listdir and os.walk) depends on the type of their input: If passed byte strings they return byte strings and if passed Unicode they automatically decode the raw filenames to Unicode before returning them. Other functions exist in two variants that return byte strings (e.g. os.getcwd) and Unicode (os.getcwdu), respectively.

Warning

Some of Python’s I/O-functions may return both byte and Unicode strings for a single call. For example, os.listdir will normally return Unicode when passed Unicode, but filenames that cannot be decoded using the filesystem encoding will still be returned as byte strings!

Note that if the filename of an existing file cannot be decoded using the filesystem’s encoding then the environment Python is running in is most probably incorrectly set up.

The instructions above are meant for the names of existing files that are obtained using Python’s I/O functions. However, sometimes one also wants to create new files whose names are generated from unknown sources (e.g. user input). To make sure that the generated filename is safe to use and can be represented using the filesystem’s encoding use ckan.lib.munge.munge_filename:

>> ckan.lib.munge.munge_filename(u'Data from Linköping (year: 2016).txt')
u'data-from-linkoping-year-2016.txt'

Note

munge_filename will remove a leading path from the filename.