Enable Unicode input and display when running Python from Windows console.
A Python package to enable Unicode input and display when running Python from Windows console.
When running Python in the standard console on Windows, there are several problems when one tries to enter or display Unicode characters. The relevant issue is http://bugs.python.org/issue1602. This package solves some of them.
First, when you want to display Unicode characters in Windows console, you have to select a font able to display them. Similarly, if you want to enter Unicode characters, you have to have you keyboard properly configured. This has nothing to do with Python, but is included here for completeness.
The standard stream objects (sys.stdin, sys.stdout, sys.stderr) are not capable of reading and displaying Unicode characters in Windows console. This has nothing to do with encoding, since even sys.stdin.buffer.raw.readline() returns b"?\n" when entering α and there is no encoding under which sys.stdout.buffer.raw.write displays α.
The streams module provides several alternative stream objects. stdin_raw, stdout_raw, and stderr_raw are raw stream objects using WinAPI functions ReadConsoleW and WriteConsoleW to interact with Windows console through UTF-16-LE encoded bytes. The stdin_text, stdout_text, and stderr_text are standard text IO wrappers over standard buffered IO over our raw streams, and are intended to be primary replacements to sys.std* streams. Unfortunately, other wrappers around std*_text are needed (see below), so there are more stream objects in streams module.
The function streams.enable installs chosen stream objects instead of the original ones. By default, it chooses appropriate stream objects itself. The function streams.disable restores the original stream objects (these are stored in sys.__std*__ attributes by Python).
After replacing the stream objects, also using print with a string containing Unicode characters and displaying Unicode characters in the interactive loop works. For input, see below.
Python interactive loop doesn’t use sys.stdin to read input so fixing it doesn’t help. Also the input function may or may not use sys.stdin depending on whether sys.stdin and sys.stdout have the standard filenos and whether they are interactive. See http://bugs.python.org/issue17620 for more information.
To solve this, we install a custom readline hook. Readline hook is a function which is used to read a single line interactively by Python REPL. It may also be used by input function under certain conditions (see above). On Linux, this hook is usually set to GNU readline function, which provides features like autocompletion, history, …
The module readline_hook provides our custom readline hook, which uses sys.stdin to get the input and is (de)activated by functions readline_hook.enable, readline_hook.disable.
As we said, readline hook can be called from two places – from the REPL and from input function. In the first case the prompt is encoded using sys.stdin.encoding, but in the second case sys.stdout.encoding is used. So Python currently makes an assumption that these two encodings are equal.
Python tokenizer, which is used when parsing the input from REPL, cannot handle UTF-16 or generally any encoding containing null bytes. Because UTF-16-LE is the encoding of Unicode used by Windows, we have to additionally wrap our text stream objects (std*_text). Thus, streams module contains also stream objects stdin_text_transcoded, stdout_text_transcoded, and stderr_text_transcoded. They basically just hide the underlying UTF-16-LE encoded buffered IO, and sets encoding to UTF-8. These transcoding wrappers are used by default by streams.enable.
There are additional issues on Python 2.
Since default Python 2 strings correspond to bytes rather than unicode, people are usually calling print with bytes argument. Therefore, sys.stdout.write and sys.stderr.write should support bytes argument. That is why we add stdout_text_str and stderr_text_str stream objects to streams module. They are used by default on Python 2.
When we enter a Unicode literal into interactive interpreter, it gets processed by the Python tokenizer, which is bytes-based. When we enter u"\u03b1" into the interactive interpreter, the tokenizer gets essentially b'u"\xce\xb1"' plus the information that the encoding used is UTF-8. The problem is that the tokenizer uses the encoding only if sys.stdin is a file object (see https://hg.python.org/cpython/file/d356e68de236/Parser/tokenizer.c#l797). Hence, we introduce another stream object streams.stdin_text_fileobj that wraps stdin_text_transcoded and also is structurally compatible with Python file object. This object is used by default on Python 2.
The check for interactive streams done by raw_input unfortunately requires that both sys.stdin and sys.stdout are file objects. Besides stdin_text_fileobj for stdin we could use also stdout_text_str_fileobj for stdout. Unfortunately, that breaks print.
Using print statement or function leads to calling PyFile_WriteObject with sys.stdout as argument. Unfortunately, its generic write method is used only if it is not a file object. Otherwise, PyObject_Print is called, and this function is file-based, so it ends with a fprintf call, which is not something we want. In conclusion, we need stdout not to be a file object.
Given the situation described, the best solution seems to be reimplementing raw_input and input builtin functions and monkeypatching __builtins__. This is done by our raw_input module on Python 2.
Install the package from PyPI via pip install win-unicode-console (recommended), or download the archive and install it from the archive (e.g. pip install win_unicode_console-0.4.zip), or install the package manually by placing directory win_unicode_console and module run.py from the archive to the site-packages directory of your Python installation.
The top-level win_unicode_console module contains a function enable, which install various fixes offered by win_unicode_console modules, and a function disable, which restores the original environment. By default, custom stream objects are installed as well as a custom readline hook. On Python 2, raw_input and input functions are monkeypatched. For customization, see the sources. The logic should be clear.
Generic usage of the package is just calling win_unicode_console.enable() whenever the fixes should be applied and win_unicode_console.disable() to revert all the changes. Note that it should be a responsibility of a Python user on Windows to install win_unicode_console and fix his Python environment regarding Unicode interaction with console, rather than of a third-party developer enabling win_unicode_console in his application, which adds a dependency. Our package should be seen as an external patch to Python on Windows rather than a feature package for other packages not directly related to fixing Unicode issues.
Different ways of how win_unicode_console can be used to fix a Python environment on Windows follow.
- Python patch (recommended). Just call win_unicode_console.enable() in your sitecustomize or usercustomize module (see https://docs.python.org/3/tutorial/appendix.html#the-customization-modules for more information). This will enable win_unicode_console on every run of the Python interpreter (unless site is disabled). Doing so should not break executed scripts in any way. Otherwise, it is a bug of win_unicode_console that should be fixed.
- Opt-in runner. You may easily run a script with win_unicode_console enabled by using our runner module and its helper run script. To do so, execute py -i -m run script.py instead of py -i script.py for interactive mode, and similarly py -m run script.py instead of py script.py for non-interactive mode. Of course you may provide arguments to your script: py -i -m run script.py arg1 arg2. To run the bare interactive interpreter with win_unicode_console enabled, execute py -i -m run.
- Opt-out runner. In case you are using win_unicode_console as Python patch, but you want to run a particular script with win_unicode_console disabled, you can also use the runner. To do so, execute py -i -m run --init-disable script.py.
- Customized runner. To move arbitrary initialization (e.g. enabling win_unicode_console with non-default arguments) from sitecustomize to opt-in runner, move it to a separate module and use py -i -m run --init-module module script.py. That will import a module module on startup instead of enabling win_unicode_console with default arguments.
win_unicode_console package was tested on Python 3.4, Python 3.5, and Python 2.7. 32-bit or 64-bit shouldn’t matter. It also interacts well with the following packages:
colorama package (https://pypi.python.org/pypi/colorama) makes ANSI escape character sequences (for producing colored terminal text and cursor positioning) work under MS Windows. It does so by wrapping sys.stdout and sys.stderr streams. Since win_unicode_console replaces the streams in order to support Unicode, win_unicode_console.enable has to be called before colorama.init so everything works as expected.
As of colorama v0.3.3, there is an early binding issue (https://github.com/tartley/colorama/issues/32), so win_unicode_console.enable has to be called even before importing colorama. Note that is already the case when win_unicode_console is used as Python patch or as opt-in runner. The issue is already fixed, but there is no new release yet.
pyreadline package (https://pypi.python.org/pypi/pyreadline/2.0) implements GNU readline features on Windows. It provides its own readline hook, which actually supports Unicode input. win_unicode_console.readline_hook detects when pyreadline is active, and in that case, by default, reuses its readline hook rather than installing its own, so GNU readline features are preserved on top of our Unicode streams.
IPython (https://pypi.python.org/pypi/ipython/3.2.1) can be also used with win_unicode_console.
As of IPython 3.2.1, there is an early binding issue (https://github.com/ipython/ipython/issues/8669), so win_unicode_console.enable has to be called even before importing IPython. That is the case when win_unicode_console is used as Python patch.
There was also an issue such that IPython is not compatible with builtin function raw_input returning unicode on Python 2 (https://github.com/ipython/ipython/issues/8670). If you hit the issue, you can make win_unicode_console.raw_input.raw_input return bytes by enabling it as win_unicode_console.enable(raw_input__return_unicode=False). The issue is already fixed, but there is no new release yet.
Since version 0.4, the signature of streams.enable has been changed because there are now more options for the stream objects to be used. It now accepts a keyword argument for each stdin, stdout, stderr, setting the corresponding stream. None means “do not set”, Ellipsis means “use the default value”.
A function streams.enable_only was added. It works the same way as streams.enable, but the default value for each parameter is None.
Functions streams.enable_reader, streams.enable_writer, and streams.enable_error_writer have been removed. Example: instead of streams.enable_reader(transcode=True) use streams.enable_only(stdin=streams.stdin_text_transcoding).
There are also corresponding changes in top-level enable function.
Since version 0.3, the custom stream objects have the standard filenos, so calling input doesn’t handle Unicode without custom readline hook.