Python extension wrapping the ICU C++ API
README file for PyICU
Welcome to PyICU, a Python extension wrapping the ICU C++ libraries.
ICU stands for "International Components for Unicode". These are the i18n libraries of the Unicode Consortium. They implement much of the Unicode Standard, many of its companion Unicode Technical Standards, and much of Unicode CLDR.
The PyICU source code is hosted at https://gitlab.pyicu.org/main/pyicu.
The ICU homepage is https://icu.unicode.org/
See also the CLDR homepage at http://cldr.unicode.org/
PyICU is a python extension implemented in C++ that wraps the C/C++ ICU library.
It is known to also work as a PyPy extension.
pkg-config and the ICU libraries and headers are already installed,
building PyICU from the sources on PyPI
involves more than just a
pip call. Many operating systems distribute
pre-built binary packages of ICU and PyICU, see below.
Mac OS X
# install libicu (keg-only) brew install pkg-config icu4c # let setup.py discover keg-only icu4c via pkg-config export PATH="/usr/local/opt/icu4c/bin:/usr/local/opt/icu4c/sbin:$PATH" export PKG_CONFIG_PATH="$PKG_CONFIG_PATH:/usr/local/opt/icu4c/lib/pkgconfig"
Install PyICU with the same C++ compiler as your Python distribution (more info):
# EITHER - when using a gcc-built CPython (e.g. from Homebrew) export CC="$(which gcc)" CXX="$(which g++)" # OR - when using system CPython or another clang-based CPython, ensure system clang is used (for proper libstdc++ https://gitlab.pyicu.org/main/pyicu/issues/5#issuecomment-291631507): unset CC CXX # avoid wheels from previous runs or PyPI pip install --no-binary=:pyicu: pyicu
ICU and PyICU binaries are both available via Macports as well. The same limitations about mixing binaries may apply.
# see versions available /opt/local/bin/port search pyicu sudo /opt/local/bin/port install ...
apt-get update # EITHER - from apt directly https://packages.debian.org/source/stable/pyicu apt-get install python3-icu # OR - from source apt-get install pkg-config libicu-dev pip install --no-binary=:pyicu: pyicu
Ubuntu: similar to Debian, there is a pyicu package available via
Alpine Linux: there is a pyicu package available via
NetBSD: there is a pyicu package available via
OpenBSD: there is a pyicu package available via
Other operating systems: see below.
Please, refer to next section for building Python, ICU and PyICU from sources. The current section is about building only PyICU from sources, with all dependencies such as Python and ICU already present.
Before building PyICU the ICU libraries must be built and installed. Refer to each system's instructions for more information.
PyICU is built from sources with
setuptools or with
pkg-configis available (the
icu-configprogram is deprecated as of ICU 63.1)
pkg-config --cflags --libs icu-i18n
If this command returns an error or doesn't return the paths expected then ensure that the
setup.pycontain correct values for your platform. Starting with ICU 60,
-std=c++11must appear in your CFLAGS or be the default for your C++ compiler.
either build and install PyICU with
python setup.py build sudo python setup.py install
or build PyICU with
buildand install it with
python -m build sudo python -m pip install dist/PyICU-<version>-<platform>.whl
either test PyICU with
python setup.py test
or test PyICU with
python -m pytest
Building PyICU, Python 3 and ICU from sources
The instructions at note_855 contain the complete steps for building everything from sources into a self-contained directory, without modifying any system directories. They were made and tested on an M1 Mac but they can be modified and reused for any unix environment. In particular, they outline how to build PyICU from sources without icu-config or pkg-config being present.
Mac OS X Make sure that
DYLD_LIBRARY_PATHcontains paths to the directory(ies) containing the ICU libs.
Linux & Solaris Make sure that
LD_LIBRARY_PATHcontains paths to the directory(ies) containing the ICU libs or that you added the corresponding
Windows Make sure that
PATHcontains paths to the directory(ies) containing the ICU DLLs.
See the CHANGES file for an up to date log of changes and additions.
There is no API documentation for PyICU. The API for ICU is documented at https://unicode-org.github.io/icu-docs/apidoc/released/icu4c/ and the following patterns can be used to translate from the C++ APIs to the corresponding Python APIs.
The ICU string type, UnicodeString, is a type pointing at a mutable array of UChar Unicode 16-bit wide characters and is described here. The Python 3 str type is described here and here. The Python 2 unicode type is described here.
Because of their differences, ICU's and Python's string objects are not merged into the same type when crossing the C++ boundary but converted.
ICU APIs taking
UnicodeString arguments have been overloaded to also
accept arguments that are Python 3
str or Python 2
str objects are auto-decoded into ICU strings using the
To convert a Python 3
bytes or a Python 2
str object encoded in an
encoding other than
utf-8 to an ICU
UnicodeString use the
UnicodeString(str, encodingName) constructor.
ICU's C++ APIs accept and return
UnicodeString arguments in several
ways: by value, by pointer or by reference.
When an ICU C++ API is documented to accept a
parameter, it is safe to assume that there are several corresponding
PyICU python APIs making it accessible in simpler ways:
For example, the
'UnicodeString &Locale::getDisplayName(UnicodeString &)'
can be invoked from Python in several ways:
The ICU way
>>> from icu import UnicodeString, Locale >>> locale = Locale('pt_BR') >>> string = UnicodeString() >>> name = locale.getDisplayName(string) >>> name <UnicodeString: 'Portuguese (Brazil)'> >>> name is string True <-- string arg was returned, modified in place
The Python way
>>> from icu import Locale >>> locale = Locale('pt_BR') >>> name = locale.getDisplayName() >>> name 'Portuguese (Brazil)'
UnicodeStringobject was allocated and converted to a Python
A UnicodeString can be converted to a Python unicode string with Python 3's
str() or Python 2's
unicode() constructor. The usual
and[:]`` operators are all available, with the additional
twists that slicing is not read-only and that ``+=`` is also available since a
UnicodeString is mutable. For example:
>>> name = locale.getDisplayName() 'Portuguese (Brazil)' >>> name = UnicodeString(name) >>> name <UnicodeString: 'Portuguese (Brazil)'> >>> str(name) 'Portuguese (Brazil)' >>> len(name) 19 >>> str(name) 'Portuguese (Brazil)' >>> name 't' >>> name[12:18] <UnicodeString: 'Brazil'> >>> name[12:18] = 'the country of Brasil' >>> name <UnicodeString: 'Portuguese (the country of Brasil)'> >>> name += ' oh joy' >>> name <UnicodeString: 'Portuguese (the country of Brasil) oh joy'>
The C++ ICU library does not use C++ exceptions to report errors. ICU
C++ APIs return errors via a
UErrorCode reference argument. All such
APIs are wrapped by Python APIs that omit this argument and throw an
ICUError Python exception instead. The same is true for ICU APIs
taking both a
ParseError and a
UErrorCode, they are both to be
For example, the
'UnicodeString &DateFormat::format(const Formattable &, UnicodeString &, FieldPosition &, UErrorCode &)' API, documented here is invoked from Python with:
>>> from icu import DateFormat, Formattable >>> df = DateFormat.createInstance() >>> df <SimpleDateFormat: M/d/yy h:mm a> >>> f = Formattable(940284258.0, Formattable.kIsDate) >>> df.format(f) '10/18/99 3:04 PM'
Of course, the simpler
'UnicodeString &DateFormat::format(UDate, UnicodeString &)' documented here can be used too:
>>> from icu import DateFormat >>> df = DateFormat.createInstance() >>> df <SimpleDateFormat: M/d/yy h:mm a> >>> df.format(940284258.0) '10/18/99 3:04 PM'
ICU uses a double floating point type called
UDate that represents the
number of milliseconds elapsed since 1970-jan-01 UTC for dates.
In Python, the value returned by the
function is the number of seconds since 1970-jan-01 UTC. Because of this
difference, floating point values are multiplied by 1000 when passed to
UDate and divided by 1000 when returned as
datetime objects, with or without timezone information, can
also be used with APIs taking
UDate arguments. The
objects get converted to
UDate when crossing into the C++ layer.
Many ICU API take array arguments. A list of elements of the array element types is to be passed from Python.
StringEnumeration has three
unext() which returns
str objects in Python 3
unicode objects in Python 2 and
snext() which returns
UnicodeString objects. Any of these methods can be used as an iterator,
using the Python built-in
For example, let
e be a
e = TimeZone.createEnumeration() [s for s in e] # a list of 'str' objects [s for s in iter(e.unext, '')] # a list of 'str' or 'unicode' objects [s for s in iter(e.snext, '')] # a list of 'UnicodeString' objects
TimeZone type may be wrapped with an
ICUtzinfo type for
usage with Python's
datetime type. For example:
from datetime import datetime tz = ICUtzinfo(TimeZone.createTimeZone('US/Mountain')) datetime.now(tz)
or, even simpler:
tz = ICUtzinfo.getInstance('Pacific/Fiji') datetime.now(tz)
To get the default time zone use:
defaultTZ = ICUtzinfo.getDefault()
To get the time zone's id, use the
tzid attribute or coerce the time
zone to a string:
ICUtzinfo.getInstance('Pacific/Fiji').tzid -> 'Pacific/Fiji' str(ICUtzinfo.getInstance('Pacific/Fiji')) -> 'Pacific/Fiji'
The unit tests have more examples of actual PyICU usage.
There are also a few samples ported from ICU C/C++.
Last but not least, this cheat sheet has useful examples.
Release history Release notifications | RSS feed
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.