PageRenderTime 343ms CodeModel.GetById 319ms app.highlight 17ms RepoModel.GetById 1ms app.codeStats 0ms

/Doc/howto/unicode.rst

http://unladen-swallow.googlecode.com/
ReStructuredText | 728 lines | 573 code | 155 blank | 0 comment | 0 complexity | f4c3b6383c4b14b0445802f4381cf4c8 MD5 | raw file
  1*****************
  2  Unicode HOWTO
  3*****************
  4
  5:Release: 1.02
  6
  7This HOWTO discusses Python's support for Unicode, and explains various problems
  8that people commonly encounter when trying to work with Unicode.
  9
 10Introduction to Unicode
 11=======================
 12
 13History of Character Codes
 14--------------------------
 15
 16In 1968, the American Standard Code for Information Interchange, better known by
 17its acronym ASCII, was standardized.  ASCII defined numeric codes for various
 18characters, with the numeric values running from 0 to
 19127.  For example, the lowercase letter 'a' is assigned 97 as its code
 20value.
 21
 22ASCII was an American-developed standard, so it only defined unaccented
 23characters.  There was an 'e', but no 'é' or 'Í'.  This meant that languages
 24which required accented characters couldn't be faithfully represented in ASCII.
 25(Actually the missing accents matter for English, too, which contains words such
 26as 'naïve' and 'café', and some publications have house styles which require
 27spellings such as 'coöperate'.)
 28
 29For a while people just wrote programs that didn't display accents.  I remember
 30looking at Apple ][ BASIC programs, published in French-language publications in
 31the mid-1980s, that had lines like these::
 32
 33   PRINT "FICHIER EST COMPLETE."
 34   PRINT "CARACTERE NON ACCEPTE."
 35
 36Those messages should contain accents, and they just look wrong to someone who
 37can read French.
 38
 39In the 1980s, almost all personal computers were 8-bit, meaning that bytes could
 40hold values ranging from 0 to 255.  ASCII codes only went up to 127, so some
 41machines assigned values between 128 and 255 to accented characters.  Different
 42machines had different codes, however, which led to problems exchanging files.
 43Eventually various commonly used sets of values for the 128-255 range emerged.
 44Some were true standards, defined by the International Standards Organization,
 45and some were **de facto** conventions that were invented by one company or
 46another and managed to catch on.
 47
 48255 characters aren't very many.  For example, you can't fit both the accented
 49characters used in Western Europe and the Cyrillic alphabet used for Russian
 50into the 128-255 range because there are more than 127 such characters.
 51
 52You could write files using different codes (all your Russian files in a coding
 53system called KOI8, all your French files in a different coding system called
 54Latin1), but what if you wanted to write a French document that quotes some
 55Russian text?  In the 1980s people began to want to solve this problem, and the
 56Unicode standardization effort began.
 57
 58Unicode started out using 16-bit characters instead of 8-bit characters.  16
 59bits means you have 2^16 = 65,536 distinct values available, making it possible
 60to represent many different characters from many different alphabets; an initial
 61goal was to have Unicode contain the alphabets for every single human language.
 62It turns out that even 16 bits isn't enough to meet that goal, and the modern
 63Unicode specification uses a wider range of codes, 0-1,114,111 (0x10ffff in
 64base-16).
 65
 66There's a related ISO standard, ISO 10646.  Unicode and ISO 10646 were
 67originally separate efforts, but the specifications were merged with the 1.1
 68revision of Unicode.
 69
 70(This discussion of Unicode's history is highly simplified.  I don't think the
 71average Python programmer needs to worry about the historical details; consult
 72the Unicode consortium site listed in the References for more information.)
 73
 74
 75Definitions
 76-----------
 77
 78A **character** is the smallest possible component of a text.  'A', 'B', 'C',
 79etc., are all different characters.  So are 'È' and 'Í'.  Characters are
 80abstractions, and vary depending on the language or context you're talking
 81about.  For example, the symbol for ohms (Ω) is usually drawn much like the
 82capital letter omega (Ω) in the Greek alphabet (they may even be the same in
 83some fonts), but these are two different characters that have different
 84meanings.
 85
 86The Unicode standard describes how characters are represented by **code
 87points**.  A code point is an integer value, usually denoted in base 16.  In the
 88standard, a code point is written using the notation U+12ca to mean the
 89character with value 0x12ca (4810 decimal).  The Unicode standard contains a lot
 90of tables listing characters and their corresponding code points::
 91
 92   0061    'a'; LATIN SMALL LETTER A
 93   0062    'b'; LATIN SMALL LETTER B
 94   0063    'c'; LATIN SMALL LETTER C
 95   ...
 96   007B    '{'; LEFT CURLY BRACKET
 97
 98Strictly, these definitions imply that it's meaningless to say 'this is
 99character U+12ca'.  U+12ca is a code point, which represents some particular
100character; in this case, it represents the character 'ETHIOPIC SYLLABLE WI'.  In
101informal contexts, this distinction between code points and characters will
102sometimes be forgotten.
103
104A character is represented on a screen or on paper by a set of graphical
105elements that's called a **glyph**.  The glyph for an uppercase A, for example,
106is two diagonal strokes and a horizontal stroke, though the exact details will
107depend on the font being used.  Most Python code doesn't need to worry about
108glyphs; figuring out the correct glyph to display is generally the job of a GUI
109toolkit or a terminal's font renderer.
110
111
112Encodings
113---------
114
115To summarize the previous section: a Unicode string is a sequence of code
116points, which are numbers from 0 to 0x10ffff.  This sequence needs to be
117represented as a set of bytes (meaning, values from 0-255) in memory.  The rules
118for translating a Unicode string into a sequence of bytes are called an
119**encoding**.
120
121The first encoding you might think of is an array of 32-bit integers.  In this
122representation, the string "Python" would look like this::
123
124       P           y           t           h           o           n
125    0x50 00 00 00 79 00 00 00 74 00 00 00 68 00 00 00 6f 00 00 00 6e 00 00 00
126       0  1  2  3  4  5  6  7  8  9 10 11 12 13 14 15 16 17 18 19 20 21 22 23
127
128This representation is straightforward but using it presents a number of
129problems.
130
1311. It's not portable; different processors order the bytes differently.
132
1332. It's very wasteful of space.  In most texts, the majority of the code points
134   are less than 127, or less than 255, so a lot of space is occupied by zero
135   bytes.  The above string takes 24 bytes compared to the 6 bytes needed for an
136   ASCII representation.  Increased RAM usage doesn't matter too much (desktop
137   computers have megabytes of RAM, and strings aren't usually that large), but
138   expanding our usage of disk and network bandwidth by a factor of 4 is
139   intolerable.
140
1413. It's not compatible with existing C functions such as ``strlen()``, so a new
142   family of wide string functions would need to be used.
143
1444. Many Internet standards are defined in terms of textual data, and can't
145   handle content with embedded zero bytes.
146
147Generally people don't use this encoding, instead choosing other encodings that
148are more efficient and convenient.
149
150Encodings don't have to handle every possible Unicode character, and most
151encodings don't.  For example, Python's default encoding is the 'ascii'
152encoding.  The rules for converting a Unicode string into the ASCII encoding are
153simple; for each code point:
154
1551. If the code point is < 128, each byte is the same as the value of the code
156   point.
157
1582. If the code point is 128 or greater, the Unicode string can't be represented
159   in this encoding.  (Python raises a :exc:`UnicodeEncodeError` exception in this
160   case.)
161
162Latin-1, also known as ISO-8859-1, is a similar encoding.  Unicode code points
1630-255 are identical to the Latin-1 values, so converting to this encoding simply
164requires converting code points to byte values; if a code point larger than 255
165is encountered, the string can't be encoded into Latin-1.
166
167Encodings don't have to be simple one-to-one mappings like Latin-1.  Consider
168IBM's EBCDIC, which was used on IBM mainframes.  Letter values weren't in one
169block: 'a' through 'i' had values from 129 to 137, but 'j' through 'r' were 145
170through 153.  If you wanted to use EBCDIC as an encoding, you'd probably use
171some sort of lookup table to perform the conversion, but this is largely an
172internal detail.
173
174UTF-8 is one of the most commonly used encodings.  UTF stands for "Unicode
175Transformation Format", and the '8' means that 8-bit numbers are used in the
176encoding.  (There's also a UTF-16 encoding, but it's less frequently used than
177UTF-8.)  UTF-8 uses the following rules:
178
1791. If the code point is <128, it's represented by the corresponding byte value.
1802. If the code point is between 128 and 0x7ff, it's turned into two byte values
181   between 128 and 255.
1823. Code points >0x7ff are turned into three- or four-byte sequences, where each
183   byte of the sequence is between 128 and 255.
184
185UTF-8 has several convenient properties:
186
1871. It can handle any Unicode code point.
1882. A Unicode string is turned into a string of bytes containing no embedded zero
189   bytes.  This avoids byte-ordering issues, and means UTF-8 strings can be
190   processed by C functions such as ``strcpy()`` and sent through protocols that
191   can't handle zero bytes.
1923. A string of ASCII text is also valid UTF-8 text.
1934. UTF-8 is fairly compact; the majority of code points are turned into two
194   bytes, and values less than 128 occupy only a single byte.
1955. If bytes are corrupted or lost, it's possible to determine the start of the
196   next UTF-8-encoded code point and resynchronize.  It's also unlikely that
197   random 8-bit data will look like valid UTF-8.
198
199
200
201References
202----------
203
204The Unicode Consortium site at <http://www.unicode.org> has character charts, a
205glossary, and PDF versions of the Unicode specification.  Be prepared for some
206difficult reading.  <http://www.unicode.org/history/> is a chronology of the
207origin and development of Unicode.
208
209To help understand the standard, Jukka Korpela has written an introductory guide
210to reading the Unicode character tables, available at
211<http://www.cs.tut.fi/~jkorpela/unicode/guide.html>.
212
213Two other good introductory articles were written by Joel Spolsky
214<http://www.joelonsoftware.com/articles/Unicode.html> and Jason Orendorff
215<http://www.jorendorff.com/articles/unicode/>.  If this introduction didn't make
216things clear to you, you should try reading one of these alternate articles
217before continuing.
218
219Wikipedia entries are often helpful; see the entries for "character encoding"
220<http://en.wikipedia.org/wiki/Character_encoding> and UTF-8
221<http://en.wikipedia.org/wiki/UTF-8>, for example.
222
223
224Python's Unicode Support
225========================
226
227Now that you've learned the rudiments of Unicode, we can look at Python's
228Unicode features.
229
230
231The Unicode Type
232----------------
233
234Unicode strings are expressed as instances of the :class:`unicode` type, one of
235Python's repertoire of built-in types.  It derives from an abstract type called
236:class:`basestring`, which is also an ancestor of the :class:`str` type; you can
237therefore check if a value is a string type with ``isinstance(value,
238basestring)``.  Under the hood, Python represents Unicode strings as either 16-
239or 32-bit integers, depending on how the Python interpreter was compiled.
240
241The :func:`unicode` constructor has the signature ``unicode(string[, encoding,
242errors])``.  All of its arguments should be 8-bit strings.  The first argument
243is converted to Unicode using the specified encoding; if you leave off the
244``encoding`` argument, the ASCII encoding is used for the conversion, so
245characters greater than 127 will be treated as errors::
246
247    >>> unicode('abcdef')
248    u'abcdef'
249    >>> s = unicode('abcdef')
250    >>> type(s)
251    <type 'unicode'>
252    >>> unicode('abcdef' + chr(255))
253    Traceback (most recent call last):
254      File "<stdin>", line 1, in ?
255    UnicodeDecodeError: 'ascii' codec can't decode byte 0xff in position 6:
256                        ordinal not in range(128)
257
258The ``errors`` argument specifies the response when the input string can't be
259converted according to the encoding's rules.  Legal values for this argument are
260'strict' (raise a ``UnicodeDecodeError`` exception), 'replace' (add U+FFFD,
261'REPLACEMENT CHARACTER'), or 'ignore' (just leave the character out of the
262Unicode result).  The following examples show the differences::
263
264    >>> unicode('\x80abc', errors='strict')
265    Traceback (most recent call last):
266      File "<stdin>", line 1, in ?
267    UnicodeDecodeError: 'ascii' codec can't decode byte 0x80 in position 0:
268                        ordinal not in range(128)
269    >>> unicode('\x80abc', errors='replace')
270    u'\ufffdabc'
271    >>> unicode('\x80abc', errors='ignore')
272    u'abc'
273
274Encodings are specified as strings containing the encoding's name.  Python 2.4
275comes with roughly 100 different encodings; see the Python Library Reference at
276:ref:`standard-encodings` for a list.  Some encodings
277have multiple names; for example, 'latin-1', 'iso_8859_1' and '8859' are all
278synonyms for the same encoding.
279
280One-character Unicode strings can also be created with the :func:`unichr`
281built-in function, which takes integers and returns a Unicode string of length 1
282that contains the corresponding code point.  The reverse operation is the
283built-in :func:`ord` function that takes a one-character Unicode string and
284returns the code point value::
285
286    >>> unichr(40960)
287    u'\ua000'
288    >>> ord(u'\ua000')
289    40960
290
291Instances of the :class:`unicode` type have many of the same methods as the
2928-bit string type for operations such as searching and formatting::
293
294    >>> s = u'Was ever feather so lightly blown to and fro as this multitude?'
295    >>> s.count('e')
296    5
297    >>> s.find('feather')
298    9
299    >>> s.find('bird')
300    -1
301    >>> s.replace('feather', 'sand')
302    u'Was ever sand so lightly blown to and fro as this multitude?'
303    >>> s.upper()
304    u'WAS EVER FEATHER SO LIGHTLY BLOWN TO AND FRO AS THIS MULTITUDE?'
305
306Note that the arguments to these methods can be Unicode strings or 8-bit
307strings.  8-bit strings will be converted to Unicode before carrying out the
308operation; Python's default ASCII encoding will be used, so characters greater
309than 127 will cause an exception::
310
311    >>> s.find('Was\x9f')
312    Traceback (most recent call last):
313      File "<stdin>", line 1, in ?
314    UnicodeDecodeError: 'ascii' codec can't decode byte 0x9f in position 3: ordinal not in range(128)
315    >>> s.find(u'Was\x9f')
316    -1
317
318Much Python code that operates on strings will therefore work with Unicode
319strings without requiring any changes to the code.  (Input and output code needs
320more updating for Unicode; more on this later.)
321
322Another important method is ``.encode([encoding], [errors='strict'])``, which
323returns an 8-bit string version of the Unicode string, encoded in the requested
324encoding.  The ``errors`` parameter is the same as the parameter of the
325``unicode()`` constructor, with one additional possibility; as well as 'strict',
326'ignore', and 'replace', you can also pass 'xmlcharrefreplace' which uses XML's
327character references.  The following example shows the different results::
328
329    >>> u = unichr(40960) + u'abcd' + unichr(1972)
330    >>> u.encode('utf-8')
331    '\xea\x80\x80abcd\xde\xb4'
332    >>> u.encode('ascii')
333    Traceback (most recent call last):
334      File "<stdin>", line 1, in ?
335    UnicodeEncodeError: 'ascii' codec can't encode character '\ua000' in position 0: ordinal not in range(128)
336    >>> u.encode('ascii', 'ignore')
337    'abcd'
338    >>> u.encode('ascii', 'replace')
339    '?abcd?'
340    >>> u.encode('ascii', 'xmlcharrefreplace')
341    '&#40960;abcd&#1972;'
342
343Python's 8-bit strings have a ``.decode([encoding], [errors])`` method that
344interprets the string using the given encoding::
345
346    >>> u = unichr(40960) + u'abcd' + unichr(1972)   # Assemble a string
347    >>> utf8_version = u.encode('utf-8')             # Encode as UTF-8
348    >>> type(utf8_version), utf8_version
349    (<type 'str'>, '\xea\x80\x80abcd\xde\xb4')
350    >>> u2 = utf8_version.decode('utf-8')            # Decode using UTF-8
351    >>> u == u2                                      # The two strings match
352    True
353
354The low-level routines for registering and accessing the available encodings are
355found in the :mod:`codecs` module.  However, the encoding and decoding functions
356returned by this module are usually more low-level than is comfortable, so I'm
357not going to describe the :mod:`codecs` module here.  If you need to implement a
358completely new encoding, you'll need to learn about the :mod:`codecs` module
359interfaces, but implementing encodings is a specialized task that also won't be
360covered here.  Consult the Python documentation to learn more about this module.
361
362The most commonly used part of the :mod:`codecs` module is the
363:func:`codecs.open` function which will be discussed in the section on input and
364output.
365
366
367Unicode Literals in Python Source Code
368--------------------------------------
369
370In Python source code, Unicode literals are written as strings prefixed with the
371'u' or 'U' character: ``u'abcdefghijk'``.  Specific code points can be written
372using the ``\u`` escape sequence, which is followed by four hex digits giving
373the code point.  The ``\U`` escape sequence is similar, but expects 8 hex
374digits, not 4.
375
376Unicode literals can also use the same escape sequences as 8-bit strings,
377including ``\x``, but ``\x`` only takes two hex digits so it can't express an
378arbitrary code point.  Octal escapes can go up to U+01ff, which is octal 777.
379
380::
381
382    >>> s = u"a\xac\u1234\u20ac\U00008000"
383               ^^^^ two-digit hex escape
384                   ^^^^^^ four-digit Unicode escape
385                               ^^^^^^^^^^ eight-digit Unicode escape
386    >>> for c in s:  print ord(c),
387    ...
388    97 172 4660 8364 32768
389
390Using escape sequences for code points greater than 127 is fine in small doses,
391but becomes an annoyance if you're using many accented characters, as you would
392in a program with messages in French or some other accent-using language.  You
393can also assemble strings using the :func:`unichr` built-in function, but this is
394even more tedious.
395
396Ideally, you'd want to be able to write literals in your language's natural
397encoding.  You could then edit Python source code with your favorite editor
398which would display the accented characters naturally, and have the right
399characters used at runtime.
400
401Python supports writing Unicode literals in any encoding, but you have to
402declare the encoding being used.  This is done by including a special comment as
403either the first or second line of the source file::
404
405    #!/usr/bin/env python
406    # -*- coding: latin-1 -*-
407
408    u = u'abcdé'
409    print ord(u[-1])
410
411The syntax is inspired by Emacs's notation for specifying variables local to a
412file.  Emacs supports many different variables, but Python only supports
413'coding'.  The ``-*-`` symbols indicate to Emacs that the comment is special;
414they have no significance to Python but are a convention.  Python looks for
415``coding: name`` or ``coding=name`` in the comment.
416
417If you don't include such a comment, the default encoding used will be ASCII.
418Versions of Python before 2.4 were Euro-centric and assumed Latin-1 as a default
419encoding for string literals; in Python 2.4, characters greater than 127 still
420work but result in a warning.  For example, the following program has no
421encoding declaration::
422
423    #!/usr/bin/env python
424    u = u'abcdé'
425    print ord(u[-1])
426
427When you run it with Python 2.4, it will output the following warning::
428
429    amk:~$ python p263.py
430    sys:1: DeprecationWarning: Non-ASCII character '\xe9'
431         in file p263.py on line 2, but no encoding declared;
432         see http://www.python.org/peps/pep-0263.html for details
433
434
435Unicode Properties
436------------------
437
438The Unicode specification includes a database of information about code points.
439For each code point that's defined, the information includes the character's
440name, its category, the numeric value if applicable (Unicode has characters
441representing the Roman numerals and fractions such as one-third and
442four-fifths).  There are also properties related to the code point's use in
443bidirectional text and other display-related properties.
444
445The following program displays some information about several characters, and
446prints the numeric value of one particular character::
447
448    import unicodedata
449
450    u = unichr(233) + unichr(0x0bf2) + unichr(3972) + unichr(6000) + unichr(13231)
451
452    for i, c in enumerate(u):
453        print i, '%04x' % ord(c), unicodedata.category(c),
454        print unicodedata.name(c)
455
456    # Get numeric value of second character
457    print unicodedata.numeric(u[1])
458
459When run, this prints::
460
461    0 00e9 Ll LATIN SMALL LETTER E WITH ACUTE
462    1 0bf2 No TAMIL NUMBER ONE THOUSAND
463    2 0f84 Mn TIBETAN MARK HALANTA
464    3 1770 Lo TAGBANWA LETTER SA
465    4 33af So SQUARE RAD OVER S SQUARED
466    1000.0
467
468The category codes are abbreviations describing the nature of the character.
469These are grouped into categories such as "Letter", "Number", "Punctuation", or
470"Symbol", which in turn are broken up into subcategories.  To take the codes
471from the above output, ``'Ll'`` means 'Letter, lowercase', ``'No'`` means
472"Number, other", ``'Mn'`` is "Mark, nonspacing", and ``'So'`` is "Symbol,
473other".  See
474<http://www.unicode.org/Public/UNIDATA/UCD.html#General_Category_Values> for a
475list of category codes.
476
477References
478----------
479
480The Unicode and 8-bit string types are described in the Python library reference
481at :ref:`typesseq`.
482
483The documentation for the :mod:`unicodedata` module.
484
485The documentation for the :mod:`codecs` module.
486
487Marc-André Lemburg gave a presentation at EuroPython 2002 titled "Python and
488Unicode".  A PDF version of his slides is available at
489<http://downloads.egenix.com/python/Unicode-EPC2002-Talk.pdf>, and is an
490excellent overview of the design of Python's Unicode features.
491
492
493Reading and Writing Unicode Data
494================================
495
496Once you've written some code that works with Unicode data, the next problem is
497input/output.  How do you get Unicode strings into your program, and how do you
498convert Unicode into a form suitable for storage or transmission?
499
500It's possible that you may not need to do anything depending on your input
501sources and output destinations; you should check whether the libraries used in
502your application support Unicode natively.  XML parsers often return Unicode
503data, for example.  Many relational databases also support Unicode-valued
504columns and can return Unicode values from an SQL query.
505
506Unicode data is usually converted to a particular encoding before it gets
507written to disk or sent over a socket.  It's possible to do all the work
508yourself: open a file, read an 8-bit string from it, and convert the string with
509``unicode(str, encoding)``.  However, the manual approach is not recommended.
510
511One problem is the multi-byte nature of encodings; one Unicode character can be
512represented by several bytes.  If you want to read the file in arbitrary-sized
513chunks (say, 1K or 4K), you need to write error-handling code to catch the case
514where only part of the bytes encoding a single Unicode character are read at the
515end of a chunk.  One solution would be to read the entire file into memory and
516then perform the decoding, but that prevents you from working with files that
517are extremely large; if you need to read a 2Gb file, you need 2Gb of RAM.
518(More, really, since for at least a moment you'd need to have both the encoded
519string and its Unicode version in memory.)
520
521The solution would be to use the low-level decoding interface to catch the case
522of partial coding sequences.  The work of implementing this has already been
523done for you: the :mod:`codecs` module includes a version of the :func:`open`
524function that returns a file-like object that assumes the file's contents are in
525a specified encoding and accepts Unicode parameters for methods such as
526``.read()`` and ``.write()``.
527
528The function's parameters are ``open(filename, mode='rb', encoding=None,
529errors='strict', buffering=1)``.  ``mode`` can be ``'r'``, ``'w'``, or ``'a'``,
530just like the corresponding parameter to the regular built-in ``open()``
531function; add a ``'+'`` to update the file.  ``buffering`` is similarly parallel
532to the standard function's parameter.  ``encoding`` is a string giving the
533encoding to use; if it's left as ``None``, a regular Python file object that
534accepts 8-bit strings is returned.  Otherwise, a wrapper object is returned, and
535data written to or read from the wrapper object will be converted as needed.
536``errors`` specifies the action for encoding errors and can be one of the usual
537values of 'strict', 'ignore', and 'replace'.
538
539Reading Unicode from a file is therefore simple::
540
541    import codecs
542    f = codecs.open('unicode.rst', encoding='utf-8')
543    for line in f:
544        print repr(line)
545
546It's also possible to open files in update mode, allowing both reading and
547writing::
548
549    f = codecs.open('test', encoding='utf-8', mode='w+')
550    f.write(u'\u4500 blah blah blah\n')
551    f.seek(0)
552    print repr(f.readline()[:1])
553    f.close()
554
555Unicode character U+FEFF is used as a byte-order mark (BOM), and is often
556written as the first character of a file in order to assist with autodetection
557of the file's byte ordering.  Some encodings, such as UTF-16, expect a BOM to be
558present at the start of a file; when such an encoding is used, the BOM will be
559automatically written as the first character and will be silently dropped when
560the file is read.  There are variants of these encodings, such as 'utf-16-le'
561and 'utf-16-be' for little-endian and big-endian encodings, that specify one
562particular byte ordering and don't skip the BOM.
563
564
565Unicode filenames
566-----------------
567
568Most of the operating systems in common use today support filenames that contain
569arbitrary Unicode characters.  Usually this is implemented by converting the
570Unicode string into some encoding that varies depending on the system.  For
571example, Mac OS X uses UTF-8 while Windows uses a configurable encoding; on
572Windows, Python uses the name "mbcs" to refer to whatever the currently
573configured encoding is.  On Unix systems, there will only be a filesystem
574encoding if you've set the ``LANG`` or ``LC_CTYPE`` environment variables; if
575you haven't, the default encoding is ASCII.
576
577The :func:`sys.getfilesystemencoding` function returns the encoding to use on
578your current system, in case you want to do the encoding manually, but there's
579not much reason to bother.  When opening a file for reading or writing, you can
580usually just provide the Unicode string as the filename, and it will be
581automatically converted to the right encoding for you::
582
583    filename = u'filename\u4500abc'
584    f = open(filename, 'w')
585    f.write('blah\n')
586    f.close()
587
588Functions in the :mod:`os` module such as :func:`os.stat` will also accept Unicode
589filenames.
590
591:func:`os.listdir`, which returns filenames, raises an issue: should it return
592the Unicode version of filenames, or should it return 8-bit strings containing
593the encoded versions?  :func:`os.listdir` will do both, depending on whether you
594provided the directory path as an 8-bit string or a Unicode string.  If you pass
595a Unicode string as the path, filenames will be decoded using the filesystem's
596encoding and a list of Unicode strings will be returned, while passing an 8-bit
597path will return the 8-bit versions of the filenames.  For example, assuming the
598default filesystem encoding is UTF-8, running the following program::
599
600   fn = u'filename\u4500abc'
601   f = open(fn, 'w')
602   f.close()
603
604   import os
605   print os.listdir('.')
606   print os.listdir(u'.')
607
608will produce the following output::
609
610   amk:~$ python t.py
611   ['.svn', 'filename\xe4\x94\x80abc', ...]
612   [u'.svn', u'filename\u4500abc', ...]
613
614The first list contains UTF-8-encoded filenames, and the second list contains
615the Unicode versions.
616
617
618
619Tips for Writing Unicode-aware Programs
620---------------------------------------
621
622This section provides some suggestions on writing software that deals with
623Unicode.
624
625The most important tip is:
626
627    Software should only work with Unicode strings internally, converting to a
628    particular encoding on output.
629
630If you attempt to write processing functions that accept both Unicode and 8-bit
631strings, you will find your program vulnerable to bugs wherever you combine the
632two different kinds of strings.  Python's default encoding is ASCII, so whenever
633a character with an ASCII value > 127 is in the input data, you'll get a
634:exc:`UnicodeDecodeError` because that character can't be handled by the ASCII
635encoding.
636
637It's easy to miss such problems if you only test your software with data that
638doesn't contain any accents; everything will seem to work, but there's actually
639a bug in your program waiting for the first user who attempts to use characters
640> 127.  A second tip, therefore, is:
641
642    Include characters > 127 and, even better, characters > 255 in your test
643    data.
644
645When using data coming from a web browser or some other untrusted source, a
646common technique is to check for illegal characters in a string before using the
647string in a generated command line or storing it in a database.  If you're doing
648this, be careful to check the string once it's in the form that will be used or
649stored; it's possible for encodings to be used to disguise characters.  This is
650especially true if the input data also specifies the encoding; many encodings
651leave the commonly checked-for characters alone, but Python includes some
652encodings such as ``'base64'`` that modify every single character.
653
654For example, let's say you have a content management system that takes a Unicode
655filename, and you want to disallow paths with a '/' character.  You might write
656this code::
657
658    def read_file (filename, encoding):
659        if '/' in filename:
660            raise ValueError("'/' not allowed in filenames")
661        unicode_name = filename.decode(encoding)
662        f = open(unicode_name, 'r')
663        # ... return contents of file ...
664
665However, if an attacker could specify the ``'base64'`` encoding, they could pass
666``'L2V0Yy9wYXNzd2Q='``, which is the base-64 encoded form of the string
667``'/etc/passwd'``, to read a system file.  The above code looks for ``'/'``
668characters in the encoded form and misses the dangerous character in the
669resulting decoded form.
670
671References
672----------
673
674The PDF slides for Marc-André Lemburg's presentation "Writing Unicode-aware
675Applications in Python" are available at
676<http://downloads.egenix.com/python/LSM2005-Developing-Unicode-aware-applications-in-Python.pdf>
677and discuss questions of character encodings as well as how to internationalize
678and localize an application.
679
680
681Revision History and Acknowledgements
682=====================================
683
684Thanks to the following people who have noted errors or offered suggestions on
685this article: Nicholas Bastin, Marius Gedminas, Kent Johnson, Ken Krugler,
686Marc-André Lemburg, Martin von Löwis, Chad Whitacre.
687
688Version 1.0: posted August 5 2005.
689
690Version 1.01: posted August 7 2005.  Corrects factual and markup errors; adds
691several links.
692
693Version 1.02: posted August 16 2005.  Corrects factual errors.
694
695
696.. comment Additional topic: building Python w/ UCS2 or UCS4 support
697.. comment Describe obscure -U switch somewhere?
698.. comment Describe use of codecs.StreamRecoder and StreamReaderWriter
699
700.. comment
701   Original outline:
702
703   - [ ] Unicode introduction
704       - [ ] ASCII
705       - [ ] Terms
706           - [ ] Character
707           - [ ] Code point
708         - [ ] Encodings
709            - [ ] Common encodings: ASCII, Latin-1, UTF-8
710       - [ ] Unicode Python type
711           - [ ] Writing unicode literals
712               - [ ] Obscurity: -U switch
713           - [ ] Built-ins
714               - [ ] unichr()
715               - [ ] ord()
716               - [ ] unicode() constructor
717           - [ ] Unicode type
718               - [ ] encode(), decode() methods
719       - [ ] Unicodedata module for character properties
720       - [ ] I/O
721           - [ ] Reading/writing Unicode data into files
722               - [ ] Byte-order marks
723           - [ ] Unicode filenames
724       - [ ] Writing Unicode programs
725           - [ ] Do everything in Unicode
726           - [ ] Declaring source code encodings (PEP 263)
727       - [ ] Other issues
728           - [ ] Building Python (UCS2, UCS4)