path: root/docs/src/tutorial/strings.rst

.. highlight:: cython

Unicode and passing strings
===========================

Similar to the string semantics in Python 3, Cython strictly separates
byte strings and unicode strings.  Above all, this means that by default
there is no automatic conversion between byte strings and unicode strings
(except for what Python 2 does in string operations).  All encoding and
decoding must pass through an explicit encoding/decoding step.  To ease
conversion between Python and C strings in simple cases, the module-level
``c_string_type`` and ``c_string_encoding`` directives can be used to
implicitly insert these encoding/decoding steps.


Python string types in Cython code
----------------------------------

Cython supports four Python string types: :obj:`bytes`, :obj:`str`,
:obj:`unicode` and :obj:`basestring`.  The :obj:`bytes` and :obj:`unicode` types
are the specific types known from normal Python 2.x (named :obj:`bytes`
and :obj:`str` in Python 3).  Additionally, Cython also supports the
:obj:`bytearray` type which behaves like the :obj:`bytes` type, except
that it is mutable.

The :obj:`str` type is special in that it is the byte string in Python 2
and the Unicode string in Python 3 (for Cython code compiled with
language level 2, i.e. the default).  Meaning, it always corresponds
exactly with the type that the Python runtime itself calls :obj:`str`.
Thus, in Python 2, both :obj:`bytes` and :obj:`str` represent the byte string
type, whereas in Python 3, both :obj:`str` and :obj:`unicode` represent the
Python Unicode string type.  The switch is made at C compile time, the
Python version that is used to run Cython is not relevant.

When compiling Cython code with language level 3, the :obj:`str` type is
identified with exactly the Unicode string type at Cython compile time,
i.e. it does not identify with :obj:`bytes` when running in Python 2.

Note that the :obj:`str` type is not compatible with the :obj:`unicode`
type in Python 2, i.e. you cannot assign a Unicode string to a variable
or argument that is typed :obj:`str`.  The attempt will result in either
a compile time error (if detectable) or a :obj:`TypeError` exception at
runtime.  You should therefore be careful when you statically type a
string variable in code that must be compatible with Python 2, as this
Python version allows a mix of byte strings and unicode strings for data
and users normally expect code to be able to work with both.  Code that
only targets Python 3 can safely type variables and arguments as either
:obj:`bytes` or :obj:`unicode`.

The :obj:`basestring` type represents both the types :obj:`str` and :obj:`unicode`,
i.e. all Python text string types in Python 2 and Python 3.  This can be
used for typing text variables that normally contain Unicode text (at
least in Python 3) but must additionally accept the :obj:`str` type in
Python 2 for backwards compatibility reasons.  It is not compatible with
the :obj:`bytes` type.  Its usage should be rare in normal Cython code as
the generic :obj:`object` type (i.e. untyped code) will normally be good
enough and has the additional advantage of supporting the assignment of
string subtypes.  Support for the :obj:`basestring` type was added in Cython
0.20.


String literals
---------------

Cython understands all Python string type prefixes:

* ``b'bytes'`` for byte strings
* ``u'text'`` for Unicode strings
* ``f'formatted {value}'`` for formatted Unicode string literals as defined by
  :PEP:`498` (added in Cython 0.24)

Unprefixed string literals become :obj:`str` objects when compiling
with language level 2 and :obj:`unicode` objects (i.e. Python 3
:obj:`str`) with language level 3.


General notes about C strings
-----------------------------

In many use cases, C strings (a.k.a. character pointers) are slow
and cumbersome.  For one, they usually require manual memory
management in one way or another, which makes it more likely to
introduce bugs into your code.

Then, Python string objects cache their length, so requesting it
(e.g. to validate the bounds of index access or when concatenating
two strings into one) is an efficient constant time operation.
In contrast, calling :c:func:`strlen()` to get this information
from a C string takes linear time, which makes many operations on
C strings rather costly.

Regarding text processing, Python has built-in support for Unicode,
which C lacks completely.  If you are dealing with Unicode text,
you are usually better off using Python Unicode string objects than
trying to work with encoded data in C strings.  Cython makes this
quite easy and efficient.

Generally speaking: unless you know what you are doing, avoid
using C strings where possible and use Python string objects instead.
The obvious exception to this is when passing them back and forth
from and to external C code.  Also, C++ strings remember their length
as well, so they can provide a suitable alternative to Python bytes
objects in some cases, e.g. when reference counting is not needed
within a well defined context.


Passing byte strings
--------------------

we have dummy C functions declared in
a file called :file:`c_func.pyx` that we are going to reuse throughout this tutorial:

.. literalinclude:: ../../examples/tutorial/string/c_func.pyx

We make a corresponding :file:`c_func.pxd` to be able to cimport those functions:

.. literalinclude:: ../../examples/tutorial/string/c_func.pxd

It is very easy to pass byte strings between C code and Python.
When receiving a byte string from a C library, you can let Cython
convert it into a Python byte string by simply assigning it to a
Python variable::

    from c_func cimport c_call_returning_a_c_string

    cdef char* c_string = c_call_returning_a_c_string()
    cdef bytes py_string = c_string

A type cast to :obj:`object` or :obj:`bytes` will do the same thing::

    py_string = <bytes> c_string

This creates a Python byte string object that holds a copy of the
original C string.  It can be safely passed around in Python code, and
will be garbage collected when the last reference to it goes out of
scope.  It is important to remember that null bytes in the string act
as terminator character, as generally known from C.  The above will
therefore only work correctly for C strings that do not contain null
bytes.

Besides not working for null bytes, the above is also very inefficient
for long strings, since Cython has to call :c:func:`strlen()` on the
C string first to find out the length by counting the bytes up to the
terminating null byte.  In many cases, the user code will know the
length already, e.g. because a C function returned it.  In this case,
it is much more efficient to tell Cython the exact number of bytes by
slicing the C string. Here is an example:

.. literalinclude:: ../../examples/tutorial/string/slicing_c_string.pyx

Here, no additional byte counting is required and ``length`` bytes from
the ``c_string`` will be copied into the Python bytes object, including
any null bytes.  Keep in mind that the slice indices are assumed to be
accurate in this case and no bounds checking is done, so incorrect
slice indices will lead to data corruption and crashes.

Note that the creation of the Python bytes string can fail with an
exception, e.g. due to insufficient memory.  If you need to
:c:func:`free()` the string after the conversion, you should wrap
the assignment in a try-finally construct:

.. literalinclude:: ../../examples/tutorial/string/try_finally.pyx

To convert the byte string back into a C :c:type:`char*`, use the
opposite assignment::

    cdef char* other_c_string = py_string  # other_c_string is a 0-terminated string.

This is a very fast operation after which ``other_c_string`` points to
the byte string buffer of the Python string itself.  It is tied to the
life time of the Python string.  When the Python string is garbage
collected, the pointer becomes invalid.  It is therefore important to
keep a reference to the Python string as long as the :c:type:`char*`
is in use.  Often enough, this only spans the call to a C function that
receives the pointer as parameter.  Special care must be taken,
however, when the C function stores the pointer for later use.  Apart
from keeping a Python reference to the string object, no manual memory
management is required.

Starting with Cython 0.20, the :obj:`bytearray` type is supported and
coerces in the same way as the :obj:`bytes` type.  However, when using it
in a C context, special care must be taken not to grow or shrink the
object buffer after converting it to a C string pointer.  These
modifications can change the internal buffer address, which will make
the pointer invalid.


Accepting strings from Python code
----------------------------------

The other side, receiving input from Python code, may appear simple
at first sight, as it only deals with objects.  However, getting this
right without making the API too narrow or too unsafe may not be
entirely obvious.

In the case that the API only deals with byte strings, i.e. binary
data or encoded text, it is best not to type the input argument as
something like :obj:`bytes`, because that would restrict the allowed
input to exactly that type and exclude both subtypes and other kinds
of byte containers, e.g. :obj:`bytearray` objects or memory views.

Depending on how (and where) the data is being processed, it may be a
good idea to instead receive a 1-dimensional memory view, e.g.

.. literalinclude:: ../../examples/tutorial/string/arg_memview.pyx

Cython's memory views are described in more detail in
:doc:`../userguide/memoryviews`, but the above example already shows
most of the relevant functionality for 1-dimensional byte views.  They
allow for efficient processing of arrays and accept anything that can
unpack itself into a byte buffer, without intermediate copying.  The
processed content can finally be returned in the memory view itself
(or a slice of it), but it is often better to copy the data back into
a flat and simple :obj:`bytes` or :obj:`bytearray` object, especially
when only a small slice is returned.  Since memoryviews do not copy the
data, they would otherwise keep the entire original buffer alive.  The
general idea here is to be liberal with input by accepting any kind of
byte buffer, but strict with output by returning a simple, well adapted
object.  This can simply be done as follows:

.. literalinclude:: ../../examples/tutorial/string/return_memview.pyx

For read-only buffers, like :obj:`bytes`, the memoryview item type should
be declared as ``const`` (see :ref:`readonly_views`). If the byte input is
actually encoded text, and the further processing should happen at the
Unicode level, then the right thing to do is to decode the input straight
away.  This is almost only a problem in Python 2.x, where Python code
expects that it can pass a byte string (:obj:`str`) with encoded text into
a text API.  Since this usually happens in more than one place in the
module's API, a helper function is almost always the way to go, since it
allows for easy adaptation of the input normalisation process later.

This kind of input normalisation function will commonly look similar to
the following:

.. literalinclude:: ../../examples/tutorial/string/to_unicode.pyx

And should then be used like this:

.. literalinclude:: ../../examples/tutorial/string/api_func.pyx

Similarly, if the further processing happens at the byte level, but Unicode
string input should be accepted, then the following might work, if you are
using memory views:

.. literalinclude:: ../../examples/tutorial/string/to_char.pyx

In this case, you might want to additionally ensure that byte string
input really uses the correct encoding, e.g. if you require pure ASCII
input data, you can run over the buffer in a loop and check the highest
bit of each byte.  This should then also be done in the input normalisation
function.


Dealing with "const"
--------------------

Many C libraries use the ``const`` modifier in their API to declare
that they will not modify a string, or to require that users must
not modify a string they return, for example:

.. literalinclude:: ../../examples/tutorial/string/someheader.h

Cython has support for the ``const`` modifier in
the language, so you can declare the above functions straight away as
follows:

.. literalinclude:: ../../examples/tutorial/string/const.pyx


Decoding bytes to text
----------------------

The initially presented way of passing and receiving C strings is
sufficient if your code only deals with binary data in the strings.
When we deal with encoded text, however, it is best practice to decode
the C byte strings to Python Unicode strings on reception, and to
encode Python Unicode strings to C byte strings on the way out.

With a Python byte string object, you would normally just call the
``bytes.decode()`` method to decode it into a Unicode string::

    ustring = byte_string.decode('UTF-8')

Cython allows you to do the same for a C string, as long as it
contains no null bytes:

.. literalinclude:: ../../examples/tutorial/string/naive_decode.pyx

And, more efficiently, for strings where the length is known:

.. literalinclude:: ../../examples/tutorial/string/decode.pyx

The same should be used when the string contains null bytes, e.g. when
it uses an encoding like UCS-4, where each character is encoded in four
bytes most of which tend to be 0.

Again, no bounds checking is done if slice indices are provided, so
incorrect indices lead to data corruption and crashes.  However, using
negative indices is possible and will inject a call
to :c:func:`strlen()` in order to determine the string length.
Obviously, this only works for 0-terminated strings without internal
null bytes.  Text encoded in UTF-8 or one of the ISO-8859 encodings is
usually a good candidate.  If in doubt, it's better to pass indices
that are 'obviously' correct than to rely on the data to be as expected.

It is common practice to wrap string conversions (and non-trivial type
conversions in general) in dedicated functions, as this needs to be
done in exactly the same way whenever receiving text from C.  This
could look as follows:

.. literalinclude:: ../../examples/tutorial/string/utf_eight.pyx

Most likely, you will prefer shorter function names in your code based
on the kind of string being handled.  Different types of content often
imply different ways of handling them on reception.  To make the code
more readable and to anticipate future changes, it is good practice to
use separate conversion functions for different types of strings.

Encoding text to bytes
----------------------

The reverse way, converting a Python unicode string to a C
:c:type:`char*`, is pretty efficient by itself, assuming that what
you actually want is a memory managed byte string::

    py_byte_string = py_unicode_string.encode('UTF-8')
    cdef char* c_string = py_byte_string

As noted before, this takes the pointer to the byte buffer of the
Python byte string.  Trying to do the same without keeping a reference
to the Python byte string will fail with a compile error::

    # this will not compile !
    cdef char* c_string = py_unicode_string.encode('UTF-8')

Here, the Cython compiler notices that the code takes a pointer to a
temporary string result that will be garbage collected after the
assignment.  Later access to the invalidated pointer will read invalid
memory and likely result in a segfault.  Cython will therefore refuse
to compile this code.


C++ strings
-----------

When wrapping a C++ library, strings will usually come in the form of
the :c:type:`std::string` class.  As with C strings, Python byte strings
automatically coerce from and to C++ strings:

.. literalinclude:: ../../examples/tutorial/string/cpp_string.pyx

The memory management situation is different than in C because the
creation of a C++ string makes an independent copy of the string
buffer which the string object then owns.  It is therefore possible
to convert temporarily created Python objects directly into C++
strings.  A common way to make use of this is when encoding a Python
unicode string into a C++ string::

    cdef string cpp_string = py_unicode_string.encode('UTF-8')

Note that this involves a bit of overhead because it first encodes
the Unicode string into a temporarily created Python bytes object
and then copies its buffer into a new C++ string.

For the other direction, efficient decoding support is available
in Cython 0.17 and later:

.. literalinclude:: ../../examples/tutorial/string/decode_cpp_string.pyx

For C++ strings, decoding slices will always take the proper length
of the string into account and apply Python slicing semantics (e.g.
return empty strings for out-of-bounds indices).


Auto encoding and decoding
--------------------------

Cython 0.19 comes with two new directives: ``c_string_type`` and
``c_string_encoding``.  They can be used to change the Python string
types that C/C++ strings coerce from and to.  By default, they only
coerce from and to the bytes type, and encoding or decoding must
be done explicitly, as described above.

There are two use cases where this is inconvenient.  First, if all
C strings that are being processed (or the large majority) contain
text, automatic encoding and decoding from and to Python unicode
objects can reduce the code overhead a little.  In this case, you
can set the ``c_string_type`` directive in your module to :obj:`unicode`
and the ``c_string_encoding`` to the encoding that your C code uses,
for example:

.. literalinclude:: ../../examples/tutorial/string/auto_conversion_1.pyx

The second use case is when all C strings that are being processed
only contain ASCII encodable characters (e.g. numbers) and you want
your code to use the native legacy string type in Python 2 for them,
instead of always using Unicode. In this case, you can set the
string type to :obj:`str`:

.. literalinclude:: ../../examples/tutorial/string/auto_conversion_2.pyx

The other direction, i.e. automatic encoding to C strings, is only
supported for ASCII and the "default encoding", which is usually UTF-8
in Python 3 and usually ASCII in Python 2.  CPython handles the memory
management in this case by keeping an encoded copy of the string alive
together with the original unicode string.  Otherwise, there would be no
way to limit the lifetime of the encoded string in any sensible way,
thus rendering any attempt to extract a C string pointer from it a
dangerous endeavour.  The following safely converts a Unicode string to
ASCII (change ``c_string_encoding`` to ``default`` to use the default
encoding instead):

.. literalinclude:: ../../examples/tutorial/string/auto_conversion_3.pyx

(This example uses a function context in order to safely control the
lifetime of the Unicode string.  Global Python variables can be
modified from the outside, which makes it dangerous to rely on the
lifetime of their values.)


Source code encoding
--------------------

When string literals appear in the code, the source code encoding is
important.  It determines the byte sequence that Cython will store in
the C code for bytes literals, and the Unicode code points that Cython
builds for unicode literals when parsing the byte encoded source file.
Following :PEP:`263`, Cython supports the explicit declaration of
source file encodings.  For example, putting the following comment at
the top of an ``ISO-8859-15`` (Latin-9) encoded source file (into the
first or second line) is required to enable ``ISO-8859-15`` decoding
in the parser::

    # -*- coding: ISO-8859-15 -*-

When no explicit encoding declaration is provided, the source code is
parsed as UTF-8 encoded text, as specified by :PEP:`3120`.  `UTF-8`_
is a very common encoding that can represent the entire Unicode set of
characters and is compatible with plain ASCII encoded text that it
encodes efficiently.  This makes it a very good choice for source code
files which usually consist mostly of ASCII characters.

.. _`UTF-8`: http://en.wikipedia.org/wiki/UTF-8

As an example, putting the following line into a UTF-8 encoded source
file will print ``5``, as UTF-8 encodes the letter ``'ö'`` in the two
byte sequence ``'\xc3\xb6'``::

    print( len(b'abcö') )

whereas the following ``ISO-8859-15`` encoded source file will print
``4``, as the encoding uses only 1 byte for this letter::

    # -*- coding: ISO-8859-15 -*-
    print( len(b'abcö') )

Note that the unicode literal ``u'abcö'`` is a correctly decoded four
character Unicode string in both cases, whereas the unprefixed Python
:obj:`str` literal ``'abcö'`` will become a byte string in Python 2 (thus
having length 4 or 5 in the examples above), and a 4 character Unicode
string in Python 3.  If you are not familiar with encodings, this may
not appear obvious at first read.  See `CEP 108`_ for details.

As a rule of thumb, it is best to avoid unprefixed non-ASCII :obj:`str`
literals and to use unicode string literals for all text.  Cython also
supports the ``__future__`` import ``unicode_literals`` that instructs
the parser to read all unprefixed :obj:`str` literals in a source file as
unicode string literals, just like Python 3.

.. _`CEP 108`: https://github.com/cython/cython/wiki/enhancements-stringliterals

Single bytes and characters
---------------------------

The Python C-API uses the normal C :c:type:`char` type to represent
a byte value, but it has two special integer types for a Unicode code
point value, i.e. a single Unicode character: :c:type:`Py_UNICODE`
and :c:type:`Py_UCS4`.  Cython supports the
first natively, support for :c:type:`Py_UCS4` is new in Cython 0.15.
:c:type:`Py_UNICODE` is either defined as an unsigned 2-byte or
4-byte integer, or as :c:type:`wchar_t`, depending on the platform.
The exact type is a compile time option in the build of the CPython
interpreter and extension modules inherit this definition at C
compile time.  The advantage of :c:type:`Py_UCS4` is that it is
guaranteed to be large enough for any Unicode code point value,
regardless of the platform.  It is defined as a 32bit unsigned int
or long.

In Cython, the :c:type:`char` type behaves differently from the
:c:type:`Py_UNICODE` and :c:type:`Py_UCS4` types when coercing
to Python objects.  Similar to the behaviour of the bytes type in
Python 3, the :c:type:`char` type coerces to a Python integer
value by default, so that the following prints 65 and not ``A``::

    # -*- coding: ASCII -*-

    cdef char char_val = 'A'
    assert char_val == 65   # ASCII encoded byte value of 'A'
    print( char_val )

If you want a Python bytes string instead, you have to request it
explicitly, and the following will print ``A`` (or ``b'A'`` in Python
3)::

    print( <bytes>char_val )

The explicit coercion works for any C integer type.  Values outside of
the range of a :c:type:`char` or :c:type:`unsigned char` will raise an
:obj:`OverflowError` at runtime.  Coercion will also happen automatically
when assigning to a typed variable, e.g.::

    cdef bytes py_byte_string
    py_byte_string = char_val

On the other hand, the :c:type:`Py_UNICODE` and :c:type:`Py_UCS4`
types are rarely used outside of the context of a Python unicode string,
so their default behaviour is to coerce to a Python unicode object.  The
following will therefore print the character ``A``, as would the same
code with the :c:type:`Py_UNICODE` type::

    cdef Py_UCS4 uchar_val = u'A'
    assert uchar_val == 65 # character point value of u'A'
    print( uchar_val )

Again, explicit casting will allow users to override this behaviour.
The following will print 65::

    cdef Py_UCS4 uchar_val = u'A'
    print( <long>uchar_val )

Note that casting to a C :c:type:`long` (or :c:type:`unsigned long`) will work
just fine, as the maximum code point value that a Unicode character
can have is 1114111 (``0x10FFFF``).  On platforms with 32bit or more,
:c:type:`int` is just as good.


Narrow Unicode builds
----------------------

In narrow Unicode builds of CPython before version 3.3, i.e. builds
where ``sys.maxunicode`` is 65535 (such as all Windows builds, as
opposed to 1114111 in wide builds), it is still possible to use
Unicode character code points that do not fit into the 16 bit wide
:c:type:`Py_UNICODE` type.  For example, such a CPython build will
accept the unicode literal ``u'\U00012345'``.  However, the
underlying system level encoding leaks into Python space in this
case, so that the length of this literal becomes 2 instead of 1.
This also shows when iterating over it or when indexing into it.
The visible substrings are ``u'\uD808'`` and ``u'\uDF45'`` in this
example.  They form a so-called surrogate pair that represents the
above character.

For more information on this topic, it is worth reading the `Wikipedia
article about the UTF-16 encoding`_.

.. _`Wikipedia article about the UTF-16 encoding`: http://en.wikipedia.org/wiki/UTF-16/UCS-2

The same properties apply to Cython code that gets compiled for a
narrow CPython runtime environment.  In most cases, e.g. when
searching for a substring, this difference can be ignored as both the
text and the substring will contain the surrogates.  So most Unicode
processing code will work correctly also on narrow builds.  Encoding,
decoding and printing will work as expected, so that the above literal
turns into exactly the same byte sequence on both narrow and wide
Unicode platforms.

However, programmers should be aware that a single :c:type:`Py_UNICODE`
value (or single 'character' unicode string in CPython) may not be
enough to represent a complete Unicode character on narrow platforms.
For example, if an independent search for ``u'\uD808'`` and
``u'\uDF45'`` in a unicode string succeeds, this does not necessarily
mean that the character ``u'\U00012345`` is part of that string.  It
may well be that two different characters are in the string that just
happen to share a code unit with the surrogate pair of the character
in question.  Looking for substrings works correctly because the two
code units in the surrogate pair use distinct value ranges, so the
pair is always identifiable in a sequence of code points.

As of version 0.15, Cython has extended support for surrogate pairs so
that you can safely use an ``in`` test to search character values from
the full :c:type:`Py_UCS4` range even on narrow platforms::

    cdef Py_UCS4 uchar = 0x12345
    print( uchar in some_unicode_string )

Similarly, it can coerce a one character string with a high Unicode
code point value to a Py_UCS4 value on both narrow and wide Unicode
platforms::

    cdef Py_UCS4 uchar = u'\U00012345'
    assert uchar == 0x12345

In CPython 3.3 and later, the :c:type:`Py_UNICODE` type is an alias
for the system specific :c:type:`wchar_t` type and is no longer tied
to the internal representation of the Unicode string.  Instead, any
Unicode character can be represented on all platforms without
resorting to surrogate pairs.  This implies that narrow builds no
longer exist from that version on, regardless of the size of
:c:type:`Py_UNICODE`.  See :PEP:`393` for details.

Cython 0.16 and later handles this change internally and does the right
thing also for single character values as long as either type inference
is applied to untyped variables or the portable :c:type:`Py_UCS4` type
is explicitly used in the source code instead of the platform specific
:c:type:`Py_UNICODE` type.  Optimisations that Cython applies to the
Python unicode type will automatically adapt to :PEP:`393` at C compile
time, as usual.

Iteration
---------

Cython 0.13 supports efficient iteration over :c:type:`char*`,
bytes and unicode strings, as long as the loop variable is
appropriately typed. So the following will generate the expected
C code:

.. literalinclude:: ../../examples/tutorial/string/for_char.pyx

The same applies to bytes objects:

.. literalinclude:: ../../examples/tutorial/string/for_bytes.pyx

For unicode objects, Cython will automatically infer the type of the
loop variable as :c:type:`Py_UCS4`:

.. literalinclude:: ../../examples/tutorial/string/for_unicode.pyx

The automatic type inference usually leads to much more efficient code
here.  However, note that some unicode operations still require the
value to be a Python object, so Cython may end up generating redundant
conversion code for the loop variable value inside of the loop.  If
this leads to a performance degradation for a specific piece of code,
you can either type the loop variable as a Python object explicitly,
or assign its value to a Python typed variable somewhere inside of the
loop to enforce one-time coercion before running Python operations on
it.

There are also optimisations for ``in`` tests, so that the following
code will run in plain C code, (actually using a switch statement):

.. literalinclude:: ../../examples/tutorial/string/if_char_in.pyx

Combined with the looping optimisation above, this can result in very
efficient character switching code, e.g. in unicode parsers.

Windows and wide character APIs
-------------------------------

Windows system APIs natively support Unicode in the form of
zero-terminated UTF-16 encoded :c:type:`wchar_t*` strings, so called
"wide strings".

By default, Windows builds of CPython define :c:type:`Py_UNICODE` as
a synonym for :c:type:`wchar_t`. This makes internal :obj:`unicode`
representation compatible with UTF-16 and allows for efficient zero-copy
conversions. This also means that Windows builds are always
`Narrow Unicode builds`_ with all the caveats.

To aid interoperation with Windows APIs, Cython 0.19 supports wide
strings (in the form of :c:type:`Py_UNICODE*`) and implicitly converts
them to and from :obj:`unicode` string objects.  These conversions behave the
same way as they do for :c:type:`char*` and :obj:`bytes` as described in
`Passing byte strings`_.

In addition to automatic conversion, unicode literals that appear
in C context become C-level wide string literals and :py:func:`len`
built-in function is specialized to compute the length of zero-terminated
:c:type:`Py_UNICODE*` string or array.

Here is an example of how one would call a Unicode API on Windows::

    cdef extern from "Windows.h":

        ctypedef Py_UNICODE WCHAR
        ctypedef const WCHAR* LPCWSTR
        ctypedef void* HWND

        int MessageBoxW(HWND hWnd, LPCWSTR lpText, LPCWSTR lpCaption, int uType)

    title = u"Windows Interop Demo - Python %d.%d.%d" % sys.version_info[:3]
    MessageBoxW(NULL, u"Hello Cython \u263a", title, 0)

.. Warning::

    The use of :c:type:`Py_UNICODE*` strings outside of Windows is
    strongly discouraged. :c:type:`Py_UNICODE` is inherently not
    portable between different platforms and Python versions.

    CPython 3.3 has moved to a flexible internal representation of
    unicode strings (:pep:`393`), making all :c:type:`Py_UNICODE` related
    APIs deprecated and inefficient.

One consequence of CPython 3.3 changes is that :py:func:`len` of
:obj:`unicode` strings is always measured in *code points* ("characters"),
while Windows API expect the number of UTF-16 *code units*
(where each surrogate is counted individually). To always get the number
of code units, call :c:func:`PyUnicode_GetSize` directly.