Debian Codegraph @

Doc/howto/regex.rst

raw
1 .. _regex-howto:
2
3 ****************************
4 Regular expression HOWTO
5 ****************************
6
7 :Author: A.M. Kuchling <[email protected]>
8
9 .. TODO:
10 Document lookbehind assertions
11 Better way of displaying a RE, a string, and what it matches
12 Mention optional argument to match.groups()
13 Unicode (at least a reference)
14
15
16 .. topic:: Abstract
17
18 This document is an introductory tutorial to using regular expressions in Python
19 with the :mod:`re` module. It provides a gentler introduction than the
20 corresponding section in the Library Reference.
21
22
23 Introduction
24 ============
25
26 Regular expressions (called REs, or regexes, or regex patterns) are essentially
27 a tiny, highly specialized programming language embedded inside Python and made
28 available through the :mod:`re` module. Using this little language, you specify
29 the rules for the set of possible strings that you want to match; this set might
30 contain English sentences, or e-mail addresses, or TeX commands, or anything you
31 like. You can then ask questions such as "Does this string match the pattern?",
32 or "Is there a match for the pattern anywhere in this string?". You can also
33 use REs to modify a string or to split it apart in various ways.
34
35 Regular expression patterns are compiled into a series of bytecodes which are
36 then executed by a matching engine written in C. For advanced use, it may be
37 necessary to pay careful attention to how the engine will execute a given RE,
38 and write the RE in a certain way in order to produce bytecode that runs faster.
39 Optimization isn't covered in this document, because it requires that you have a
40 good understanding of the matching engine's internals.
41
42 The regular expression language is relatively small and restricted, so not all
43 possible string processing tasks can be done using regular expressions. There
44 are also tasks that *can* be done with regular expressions, but the expressions
45 turn out to be very complicated. In these cases, you may be better off writing
46 Python code to do the processing; while Python code will be slower than an
47 elaborate regular expression, it will also probably be more understandable.
48
49
50 Simple patterns
51 ===============
52
53 We'll start by learning about the simplest possible regular expressions. Since
54 regular expressions are used to operate on strings, we'll begin with the most
55 common task: matching characters.
56
57 For a detailed explanation of the computer science underlying regular
58 expressions (deterministic and non-deterministic finite automata), you can refer
59 to almost any textbook on writing compilers.
60
61
62 Matching characters
63 -------------------
64
65 Most letters and characters will simply match themselves. For example, the
66 regular expression ``test`` will match the string ``test`` exactly. (You can
67 enable a case-insensitive mode that would let this RE match ``Test`` or ``TEST``
68 as well; more about this later.)
69
70 There are exceptions to this rule; some characters are special
71 :dfn:`metacharacters`, and don't match themselves. Instead, they signal that
72 some out-of-the-ordinary thing should be matched, or they affect other portions
73 of the RE by repeating them or changing their meaning. Much of this document is
74 devoted to discussing various metacharacters and what they do.
75
76 Here's a complete list of the metacharacters; their meanings will be discussed
77 in the rest of this HOWTO.
78
79 .. code-block:: none
80
81 . ^ $ * + ? { } [ ] \ | ( )
82
83 The first metacharacters we'll look at are ``[`` and ``]``. They're used for
84 specifying a character class, which is a set of characters that you wish to
85 match. Characters can be listed individually, or a range of characters can be
86 indicated by giving two characters and separating them by a ``'-'``. For
87 example, ``[abc]`` will match any of the characters ``a``, ``b``, or ``c``; this
88 is the same as ``[a-c]``, which uses a range to express the same set of
89 characters. If you wanted to match only lowercase letters, your RE would be
90 ``[a-z]``.
91
92 Metacharacters (except ``\``) are not active inside classes. For example, ``[akm$]`` will
93 match any of the characters ``'a'``, ``'k'``, ``'m'``, or ``'$'``; ``'$'`` is
94 usually a metacharacter, but inside a character class it's stripped of its
95 special nature.
96
97 You can match the characters not listed within the class by :dfn:`complementing`
98 the set. This is indicated by including a ``'^'`` as the first character of the
99 class. For example, ``[^5]`` will match any character except ``'5'``. If the
100 caret appears elsewhere in a character class, it does not have special meaning.
101 For example: ``[5^]`` will match either a ``'5'`` or a ``'^'``.
102
103 Perhaps the most important metacharacter is the backslash, ``\``. As in Python
104 string literals, the backslash can be followed by various characters to signal
105 various special sequences. It's also used to escape all the metacharacters so
106 you can still match them in patterns; for example, if you need to match a ``[``
107 or ``\``, you can precede them with a backslash to remove their special
108 meaning: ``\[`` or ``\\``.
109
110 Some of the special sequences beginning with ``'\'`` represent
111 predefined sets of characters that are often useful, such as the set
112 of digits, the set of letters, or the set of anything that isn't
113 whitespace.
114
115 Let's take an example: ``\w`` matches any alphanumeric character. If
116 the regex pattern is expressed in bytes, this is equivalent to the
117 class ``[a-zA-Z0-9_]``. If the regex pattern is a string, ``\w`` will
118 match all the characters marked as letters in the Unicode database
119 provided by the :mod:`unicodedata` module. You can use the more
120 restricted definition of ``\w`` in a string pattern by supplying the
121 :const:`re.ASCII` flag when compiling the regular expression.
122
123 The following list of special sequences isn't complete. For a complete
124 list of sequences and expanded class definitions for Unicode string
125 patterns, see the last part of :ref:`Regular Expression Syntax
126 <re-syntax>` in the Standard Library reference. In general, the
127 Unicode versions match any character that's in the appropriate
128 category in the Unicode database.
129
130 ``\d``
131 Matches any decimal digit; this is equivalent to the class ``[0-9]``.
132
133 ``\D``
134 Matches any non-digit character; this is equivalent to the class ``[^0-9]``.
135
136 ``\s``
137 Matches any whitespace character; this is equivalent to the class ``[
138 \t\n\r\f\v]``.
139
140 ``\S``
141 Matches any non-whitespace character; this is equivalent to the class ``[^
142 \t\n\r\f\v]``.
143
144 ``\w``
145 Matches any alphanumeric character; this is equivalent to the class
146 ``[a-zA-Z0-9_]``.
147
148 ``\W``
149 Matches any non-alphanumeric character; this is equivalent to the class
150 ``[^a-zA-Z0-9_]``.
151
152 These sequences can be included inside a character class. For example,
153 ``[\s,.]`` is a character class that will match any whitespace character, or
154 ``','`` or ``'.'``.
155
156 The final metacharacter in this section is ``.``. It matches anything except a
157 newline character, and there's an alternate mode (:const:`re.DOTALL`) where it will
158 match even a newline. ``.`` is often used where you want to match "any
159 character".
160
161
162 Repeating things
163 ----------------
164
165 Being able to match varying sets of characters is the first thing regular
166 expressions can do that isn't already possible with the methods available on
167 strings. However, if that was the only additional capability of regexes, they
168 wouldn't be much of an advance. Another capability is that you can specify that
169 portions of the RE must be repeated a certain number of times.
170
171 The first metacharacter for repeating things that we'll look at is ``*``. ``*``
172 doesn't match the literal character ``'*'``; instead, it specifies that the
173 previous character can be matched zero or more times, instead of exactly once.
174
175 For example, ``ca*t`` will match ``'ct'`` (0 ``'a'`` characters), ``'cat'`` (1 ``'a'``),
176 ``'caaat'`` (3 ``'a'`` characters), and so forth.
177
178 Repetitions such as ``*`` are :dfn:`greedy`; when repeating a RE, the matching
179 engine will try to repeat it as many times as possible. If later portions of the
180 pattern don't match, the matching engine will then back up and try again with
181 fewer repetitions.
182
183 A step-by-step example will make this more obvious. Let's consider the
184 expression ``a[bcd]*b``. This matches the letter ``'a'``, zero or more letters
185 from the class ``[bcd]``, and finally ends with a ``'b'``. Now imagine matching
186 this RE against the string ``'abcbd'``.
187
188 +------+-----------+---------------------------------+
189 | Step | Matched | Explanation |
190 +======+===========+=================================+
191 | 1 | ``a`` | The ``a`` in the RE matches. |
192 +------+-----------+---------------------------------+
193 | 2 | ``abcbd`` | The engine matches ``[bcd]*``, |
194 | | | going as far as it can, which |
195 | | | is to the end of the string. |
196 +------+-----------+---------------------------------+
197 | 3 | *Failure* | The engine tries to match |
198 | | | ``b``, but the current position |
199 | | | is at the end of the string, so |
200 | | | it fails. |
201 +------+-----------+---------------------------------+
202 | 4 | ``abcb`` | Back up, so that ``[bcd]*`` |
203 | | | matches one less character. |
204 +------+-----------+---------------------------------+
205 | 5 | *Failure* | Try ``b`` again, but the |
206 | | | current position is at the last |
207 | | | character, which is a ``'d'``. |
208 +------+-----------+---------------------------------+
209 | 6 | ``abc`` | Back up again, so that |
210 | | | ``[bcd]*`` is only matching |
211 | | | ``bc``. |
212 +------+-----------+---------------------------------+
213 | 7 | ``abcb`` | Try ``b`` again. This time |
214 | | | the character at the |
215 | | | current position is ``'b'``, so |
216 | | | it succeeds. |
217 +------+-----------+---------------------------------+
218
219 The end of the RE has now been reached, and it has matched ``'abcb'``. This
220 demonstrates how the matching engine goes as far as it can at first, and if no
221 match is found it will then progressively back up and retry the rest of the RE
222 again and again. It will back up until it has tried zero matches for
223 ``[bcd]*``, and if that subsequently fails, the engine will conclude that the
224 string doesn't match the RE at all.
225
226 Another repeating metacharacter is ``+``, which matches one or more times. Pay
227 careful attention to the difference between ``*`` and ``+``; ``*`` matches
228 *zero* or more times, so whatever's being repeated may not be present at all,
229 while ``+`` requires at least *one* occurrence. To use a similar example,
230 ``ca+t`` will match ``'cat'`` (1 ``'a'``), ``'caaat'`` (3 ``'a'``\ s), but won't
231 match ``'ct'``.
232
233 There are two more repeating operators or quantifiers. The question mark character, ``?``,
234 matches either once or zero times; you can think of it as marking something as
235 being optional. For example, ``home-?brew`` matches either ``'homebrew'`` or
236 ``'home-brew'``.
237
238 The most complicated quantifier is ``{m,n}``, where *m* and *n* are
239 decimal integers. This quantifier means there must be at least *m* repetitions,
240 and at most *n*. For example, ``a/{1,3}b`` will match ``'a/b'``, ``'a//b'``, and
241 ``'a///b'``. It won't match ``'ab'``, which has no slashes, or ``'a////b'``, which
242 has four.
243
244 You can omit either *m* or *n*; in that case, a reasonable value is assumed for
245 the missing value. Omitting *m* is interpreted as a lower limit of 0, while
246 omitting *n* results in an upper bound of infinity.
247
248 The simplest case ``{m}`` matches the preceding item exactly *m* times.
249 For example, ``a/{2}b`` will only match ``'a//b'``.
250
251 Readers of a reductionist bent may notice that the three other quantifiers can
252 all be expressed using this notation. ``{0,}`` is the same as ``*``, ``{1,}``
253 is equivalent to ``+``, and ``{0,1}`` is the same as ``?``. It's better to use
254 ``*``, ``+``, or ``?`` when you can, simply because they're shorter and easier
255 to read.
256
257
258 Using regular expressions
259 =========================
260
261 Now that we've looked at some simple regular expressions, how do we actually use
262 them in Python? The :mod:`re` module provides an interface to the regular
263 expression engine, allowing you to compile REs into objects and then perform
264 matches with them.
265
266
267 Compiling regular expressions
268 -----------------------------
269
270 Regular expressions are compiled into pattern objects, which have
271 methods for various operations such as searching for pattern matches or
272 performing string substitutions. ::
273
274 >>> import re
275 >>> p = re.compile('ab*')
276 >>> p
277 re.compile('ab*')
278
279 :func:`re.compile` also accepts an optional *flags* argument, used to enable
280 various special features and syntax variations. We'll go over the available
281 settings later, but for now a single example will do::
282
283 >>> p = re.compile('ab*', re.IGNORECASE)
284
285 The RE is passed to :func:`re.compile` as a string. REs are handled as strings
286 because regular expressions aren't part of the core Python language, and no
287 special syntax was created for expressing them. (There are applications that
288 don't need REs at all, so there's no need to bloat the language specification by
289 including them.) Instead, the :mod:`re` module is simply a C extension module
290 included with Python, just like the :mod:`socket` or :mod:`zlib` modules.
291
292 Putting REs in strings keeps the Python language simpler, but has one
293 disadvantage which is the topic of the next section.
294
295
296 .. _the-backslash-plague:
297
298 The backslash plague
299 --------------------
300
301 As stated earlier, regular expressions use the backslash character (``'\'``) to
302 indicate special forms or to allow special characters to be used without
303 invoking their special meaning. This conflicts with Python's usage of the same
304 character for the same purpose in string literals.
305
306 Let's say you want to write a RE that matches the string ``\section``, which
307 might be found in a LaTeX file. To figure out what to write in the program
308 code, start with the desired string to be matched. Next, you must escape any
309 backslashes and other metacharacters by preceding them with a backslash,
310 resulting in the string ``\\section``. The resulting string that must be passed
311 to :func:`re.compile` must be ``\\section``. However, to express this as a
312 Python string literal, both backslashes must be escaped *again*.
313
314 +-------------------+------------------------------------------+
315 | Characters | Stage |
316 +===================+==========================================+
317 | ``\section`` | Text string to be matched |
318 +-------------------+------------------------------------------+
319 | ``\\section`` | Escaped backslash for :func:`re.compile` |
320 +-------------------+------------------------------------------+
321 | ``"\\\\section"`` | Escaped backslashes for a string literal |
322 +-------------------+------------------------------------------+
323
324 In short, to match a literal backslash, one has to write ``'\\\\'`` as the RE
325 string, because the regular expression must be ``\\``, and each backslash must
326 be expressed as ``\\`` inside a regular Python string literal. In REs that
327 feature backslashes repeatedly, this leads to lots of repeated backslashes and
328 makes the resulting strings difficult to understand.
329
330 The solution is to use Python's raw string notation for regular expressions;
331 backslashes are not handled in any special way in a string literal prefixed with
332 ``'r'``, so ``r"\n"`` is a two-character string containing ``'\'`` and ``'n'``,
333 while ``"\n"`` is a one-character string containing a newline. Regular
334 expressions will often be written in Python code using this raw string notation.
335
336 In addition, special escape sequences that are valid in regular expressions,
337 but not valid as Python string literals, now result in a
338 :exc:`SyntaxWarning` and will eventually become a :exc:`SyntaxError`,
339 which means the sequences will be invalid if raw string notation or escaping
340 the backslashes isn't used.
341
342
343 +-------------------+------------------+
344 | Regular String | Raw string |
345 +===================+==================+
346 | ``"ab*"`` | ``r"ab*"`` |
347 +-------------------+------------------+
348 | ``"\\\\section"`` | ``r"\\section"`` |
349 +-------------------+------------------+
350 | ``"\\w+\\s+\\1"`` | ``r"\w+\s+\1"`` |
351 +-------------------+------------------+
352
353
354 Performing matches
355 ------------------
356
357 Once you have an object representing a compiled regular expression, what do you
358 do with it? Pattern objects have several methods and attributes.
359 Only the most significant ones will be covered here; consult the :mod:`re` docs
360 for a complete listing.
361
362 +------------------+-----------------------------------------------+
363 | Method/Attribute | Purpose |
364 +==================+===============================================+
365 | ``match()`` | Determine if the RE matches at the beginning |
366 | | of the string. |
367 +------------------+-----------------------------------------------+
368 | ``search()`` | Scan through a string, looking for any |
369 | | location where this RE matches. |
370 +------------------+-----------------------------------------------+
371 | ``findall()`` | Find all substrings where the RE matches, and |
372 | | return them as a list. |
373 +------------------+-----------------------------------------------+
374 | ``finditer()`` | Find all substrings where the RE matches, and |
375 | | return them as an :term:`iterator`. |
376 +------------------+-----------------------------------------------+
377
378 :meth:`~re.Pattern.match` and :meth:`~re.Pattern.search` return ``None`` if no match can be found. If
379 they're successful, a :ref:`match object <match-objects>` instance is returned,
380 containing information about the match: where it starts and ends, the substring
381 it matched, and more.
382
383 You can learn about this by interactively experimenting with the :mod:`re`
384 module.
385
386 This HOWTO uses the standard Python interpreter for its examples. First, run the
387 Python interpreter, import the :mod:`re` module, and compile a RE::
388
389 >>> import re
390 >>> p = re.compile('[a-z]+')
391 >>> p
392 re.compile('[a-z]+')
393
394 Now, you can try matching various strings against the RE ``[a-z]+``. An empty
395 string shouldn't match at all, since ``+`` means 'one or more repetitions'.
396 :meth:`~re.Pattern.match` should return ``None`` in this case, which will cause the
397 interpreter to print no output. You can explicitly print the result of
398 :meth:`!match` to make this clear. ::
399
400 >>> p.match("")
401 >>> print(p.match(""))
402 None
403
404 Now, let's try it on a string that it should match, such as ``tempo``. In this
405 case, :meth:`~re.Pattern.match` will return a :ref:`match object <match-objects>`, so you
406 should store the result in a variable for later use. ::
407
408 >>> m = p.match('tempo')
409 >>> m
410 <re.Match object; span=(0, 5), match='tempo'>
411
412 Now you can query the :ref:`match object <match-objects>` for information
413 about the matching string. Match object instances
414 also have several methods and attributes; the most important ones are:
415
416 +------------------+--------------------------------------------+
417 | Method/Attribute | Purpose |
418 +==================+============================================+
419 | ``group()`` | Return the string matched by the RE |
420 +------------------+--------------------------------------------+
421 | ``start()`` | Return the starting position of the match |
422 +------------------+--------------------------------------------+
423 | ``end()`` | Return the ending position of the match |
424 +------------------+--------------------------------------------+
425 | ``span()`` | Return a tuple containing the (start, end) |
426 | | positions of the match |
427 +------------------+--------------------------------------------+
428
429 Trying these methods will soon clarify their meaning::
430
431 >>> m.group()
432 'tempo'
433 >>> m.start(), m.end()
434 (0, 5)
435 >>> m.span()
436 (0, 5)
437
438 :meth:`~re.Match.group` returns the substring that was matched by the RE. :meth:`~re.Match.start`
439 and :meth:`~re.Match.end` return the starting and ending index of the match. :meth:`~re.Match.span`
440 returns both start and end indexes in a single tuple. Since the :meth:`~re.Pattern.match`
441 method only checks if the RE matches at the start of a string, :meth:`!start`
442 will always be zero. However, the :meth:`~re.Pattern.search` method of patterns
443 scans through the string, so the match may not start at zero in that
444 case. ::
445
446 >>> print(p.match('::: message'))
447 None
448 >>> m = p.search('::: message'); print(m)
449 <re.Match object; span=(4, 11), match='message'>
450 >>> m.group()
451 'message'
452 >>> m.span()
453 (4, 11)
454
455 In actual programs, the most common style is to store the
456 :ref:`match object <match-objects>` in a variable, and then check if it was
457 ``None``. This usually looks like::
458
459 p = re.compile( ... )
460 m = p.match( 'string goes here' )
461 if m:
462 print('Match found: ', m.group())
463 else:
464 print('No match')
465
466 Two pattern methods return all of the matches for a pattern.
467 :meth:`~re.Pattern.findall` returns a list of matching strings::
468
469 >>> p = re.compile(r'\d+')
470 >>> p.findall('12 drummers drumming, 11 pipers piping, 10 lords a-leaping')
471 ['12', '11', '10']
472
473 The ``r`` prefix, making the literal a raw string literal, is needed in this
474 example because escape sequences in a normal "cooked" string literal that are
475 not recognized by Python, as opposed to regular expressions, now result in a
476 :exc:`SyntaxWarning` and will eventually become a :exc:`SyntaxError`. See
477 :ref:`the-backslash-plague`.
478
479 :meth:`~re.Pattern.findall` has to create the entire list before it can be returned as the
480 result. The :meth:`~re.Pattern.finditer` method returns a sequence of
481 :ref:`match object <match-objects>` instances as an :term:`iterator`::
482
483 >>> iterator = p.finditer('12 drummers drumming, 11 ... 10 ...')
484 >>> iterator #doctest: +ELLIPSIS
485 <callable_iterator object at 0x...>
486 >>> for match in iterator:
487 ... print(match.span())
488 ...
489 (0, 2)
490 (22, 24)
491 (29, 31)
492
493
494 Module-level functions
495 ----------------------
496
497 You don't have to create a pattern object and call its methods; the
498 :mod:`re` module also provides top-level functions called :func:`~re.match`,
499 :func:`~re.search`, :func:`~re.findall`, :func:`~re.sub`, and so forth. These functions
500 take the same arguments as the corresponding pattern method with
501 the RE string added as the first argument, and still return either ``None`` or a
502 :ref:`match object <match-objects>` instance. ::
503
504 >>> print(re.match(r'From\s+', 'Fromage amk'))
505 None
506 >>> re.match(r'From\s+', 'From amk Thu May 14 19:12:10 1998') #doctest: +ELLIPSIS
507 <re.Match object; span=(0, 5), match='From '>
508
509 Under the hood, these functions simply create a pattern object for you
510 and call the appropriate method on it. They also store the compiled
511 object in a cache, so future calls using the same RE won't need to
512 parse the pattern again and again.
513
514 Should you use these module-level functions, or should you get the
515 pattern and call its methods yourself? If you're accessing a regex
516 within a loop, pre-compiling it will save a few function calls.
517 Outside of loops, there's not much difference thanks to the internal
518 cache.
519
520
521 Compilation flags
522 -----------------
523
524 .. currentmodule:: re
525
526 Compilation flags let you modify some aspects of how regular expressions work.
527 Flags are available in the :mod:`re` module under two names, a long name such as
528 :const:`IGNORECASE` and a short, one-letter form such as :const:`I`. (If you're
529 familiar with Perl's pattern modifiers, the one-letter forms use the same
530 letters; the short form of :const:`re.VERBOSE` is :const:`re.X`, for example.)
531 Multiple flags can be specified by bitwise OR-ing them; ``re.I | re.M`` sets
532 both the :const:`I` and :const:`M` flags, for example.
533
534 Here's a table of the available flags, followed by a more detailed explanation
535 of each one.
536
537 +---------------------------------+--------------------------------------------+
538 | Flag | Meaning |
539 +=================================+============================================+
540 | :const:`ASCII`, :const:`A` | Makes several escapes like ``\w``, ``\b``, |
541 | | ``\s`` and ``\d`` match only on ASCII |
542 | | characters with the respective property. |
543 +---------------------------------+--------------------------------------------+
544 | :const:`DOTALL`, :const:`S` | Make ``.`` match any character, including |
545 | | newlines. |
546 +---------------------------------+--------------------------------------------+
547 | :const:`IGNORECASE`, :const:`I` | Do case-insensitive matches. |
548 +---------------------------------+--------------------------------------------+
549 | :const:`LOCALE`, :const:`L` | Do a locale-aware match. |
550 +---------------------------------+--------------------------------------------+
551 | :const:`MULTILINE`, :const:`M` | Multi-line matching, affecting ``^`` and |
552 | | ``$``. |
553 +---------------------------------+--------------------------------------------+
554 | :const:`VERBOSE`, :const:`X` | Enable verbose REs, which can be organized |
555 | (for 'extended') | more cleanly and understandably. |
556 +---------------------------------+--------------------------------------------+
557
558
559 .. data:: I
560 IGNORECASE
561 :noindex:
562
563 Perform case-insensitive matching; character class and literal strings will
564 match letters by ignoring case. For example, ``[A-Z]`` will match lowercase
565 letters, too. Full Unicode matching also works unless the :const:`ASCII`
566 flag is used to disable non-ASCII matches. When the Unicode patterns
567 ``[a-z]`` or ``[A-Z]`` are used in combination with the :const:`IGNORECASE`
568 flag, they will match the 52 ASCII letters and 4 additional non-ASCII
569 letters: 'İ' (U+0130, Latin capital letter I with dot above), 'ı' (U+0131,
570 Latin small letter dotless i), 'Å¿' (U+017F, Latin small letter long s) and
571 'K' (U+212A, Kelvin sign). ``Spam`` will match ``'Spam'``, ``'spam'``,
572 ``'spAM'``, or ``'ſpam'`` (the latter is matched only in Unicode mode).
573 This lowercasing doesn't take the current locale into account;
574 it will if you also set the :const:`LOCALE` flag.
575
576
577 .. data:: L
578 LOCALE
579 :noindex:
580
581 Make ``\w``, ``\W``, ``\b``, ``\B`` and case-insensitive matching dependent
582 on the current locale instead of the Unicode database.
583
584 Locales are a feature of the C library intended to help in writing programs
585 that take account of language differences. For example, if you're
586 processing encoded French text, you'd want to be able to write ``\w+`` to
587 match words, but ``\w`` only matches the character class ``[A-Za-z]`` in
588 bytes patterns; it won't match bytes corresponding to ``é`` or ``ç``.
589 If your system is configured properly and a French locale is selected,
590 certain C functions will tell the program that the byte corresponding to
591 ``é`` should also be considered a letter.
592 Setting the :const:`LOCALE` flag when compiling a regular expression will cause
593 the resulting compiled object to use these C functions for ``\w``; this is
594 slower, but also enables ``\w+`` to match French words as you'd expect.
595 The use of this flag is discouraged in Python 3 as the locale mechanism
596 is very unreliable, it only handles one "culture" at a time, and it only
597 works with 8-bit locales. Unicode matching is already enabled by default
598 in Python 3 for Unicode (str) patterns, and it is able to handle different
599 locales/languages.
600
601
602 .. data:: M
603 MULTILINE
604 :noindex:
605
606 (``^`` and ``$`` haven't been explained yet; they'll be introduced in section
607 :ref:`more-metacharacters`.)
608
609 Usually ``^`` matches only at the beginning of the string, and ``$`` matches
610 only at the end of the string and immediately before the newline (if any) at the
611 end of the string. When this flag is specified, ``^`` matches at the beginning
612 of the string and at the beginning of each line within the string, immediately
613 following each newline. Similarly, the ``$`` metacharacter matches either at
614 the end of the string and at the end of each line (immediately preceding each
615 newline).
616
617
618 .. data:: S
619 DOTALL
620 :noindex:
621
622 Makes the ``'.'`` special character match any character at all, including a
623 newline; without this flag, ``'.'`` will match anything *except* a newline.
624
625
626 .. data:: A
627 ASCII
628 :noindex:
629
630 Make ``\w``, ``\W``, ``\b``, ``\B``, ``\s`` and ``\S`` perform ASCII-only
631 matching instead of full Unicode matching. This is only meaningful for
632 Unicode patterns, and is ignored for byte patterns.
633
634
635 .. data:: X
636 VERBOSE
637 :noindex:
638
639 This flag allows you to write regular expressions that are more readable by
640 granting you more flexibility in how you can format them. When this flag has
641 been specified, whitespace within the RE string is ignored, except when the
642 whitespace is in a character class or preceded by an unescaped backslash; this
643 lets you organize and indent the RE more clearly. This flag also lets you put
644 comments within a RE that will be ignored by the engine; comments are marked by
645 a ``'#'`` that's neither in a character class nor preceded by an unescaped
646 backslash.
647
648 For example, here's a RE that uses :const:`re.VERBOSE`; see how much easier it
649 is to read? ::
650
651 charref = re.compile(r"""
652 &[#] # Start of a numeric entity reference
653 (
654 0[0-7]+ # Octal form
655 | [0-9]+ # Decimal form
656 | x[0-9a-fA-F]+ # Hexadecimal form
657 )
658 ; # Trailing semicolon
659 """, re.VERBOSE)
660
661 Without the verbose setting, the RE would look like this::
662
663 charref = re.compile("&#(0[0-7]+"
664 "|[0-9]+"
665 "|x[0-9a-fA-F]+);")
666
667 In the above example, Python's automatic concatenation of string literals has
668 been used to break up the RE into smaller pieces, but it's still more difficult
669 to understand than the version using :const:`re.VERBOSE`.
670
671
672 More pattern power
673 ==================
674
675 So far we've only covered a part of the features of regular expressions. In
676 this section, we'll cover some new metacharacters, and how to use groups to
677 retrieve portions of the text that was matched.
678
679
680 .. _more-metacharacters:
681
682 More metacharacters
683 -------------------
684
685 There are some metacharacters that we haven't covered yet. Most of them will be
686 covered in this section.
687
688 Some of the remaining metacharacters to be discussed are :dfn:`zero-width
689 assertions`. They don't cause the engine to advance through the string;
690 instead, they consume no characters at all, and simply succeed or fail. For
691 example, ``\b`` is an assertion that the current position is located at a word
692 boundary; the position isn't changed by the ``\b`` at all. This means that
693 zero-width assertions should never be repeated, because if they match once at a
694 given location, they can obviously be matched an infinite number of times.
695
696 ``|``
697 Alternation, or the "or" operator. If *A* and *B* are regular expressions,
698 ``A|B`` will match any string that matches either *A* or *B*. ``|`` has very
699 low precedence in order to make it work reasonably when you're alternating
700 multi-character strings. ``Crow|Servo`` will match either ``'Crow'`` or ``'Servo'``,
701 not ``'Cro'``, a ``'w'`` or an ``'S'``, and ``'ervo'``.
702
703 To match a literal ``'|'``, use ``\|``, or enclose it inside a character class,
704 as in ``[|]``.
705
706 ``^``
707 Matches at the beginning of lines. Unless the :const:`MULTILINE` flag has been
708 set, this will only match at the beginning of the string. In :const:`MULTILINE`
709 mode, this also matches immediately after each newline within the string.
710
711 For example, if you wish to match the word ``From`` only at the beginning of a
712 line, the RE to use is ``^From``. ::
713
714 >>> print(re.search('^From', 'From Here to Eternity')) #doctest: +ELLIPSIS
715 <re.Match object; span=(0, 4), match='From'>
716 >>> print(re.search('^From', 'Reciting From Memory'))
717 None
718
719 To match a literal ``'^'``, use ``\^``.
720
721 ``$``
722 Matches at the end of a line, which is defined as either the end of the string,
723 or any location followed by a newline character. ::
724
725 >>> print(re.search('}$', '{block}')) #doctest: +ELLIPSIS
726 <re.Match object; span=(6, 7), match='}'>
727 >>> print(re.search('}$', '{block} '))
728 None
729 >>> print(re.search('}$', '{block}\n')) #doctest: +ELLIPSIS
730 <re.Match object; span=(6, 7), match='}'>
731
732 To match a literal ``'$'``, use ``\$`` or enclose it inside a character class,
733 as in ``[$]``.
734
735 ``\A``
736 Matches only at the start of the string. When not in :const:`MULTILINE` mode,
737 ``\A`` and ``^`` are effectively the same. In :const:`MULTILINE` mode, they're
738 different: ``\A`` still matches only at the beginning of the string, but ``^``
739 may match at any location inside the string that follows a newline character.
740
741 ``\z``
742 Matches only at the end of the string.
743
744 ``\Z``
745 The same as ``\z``. For compatibility with old Python versions.
746
747 ``\b``
748 Word boundary. This is a zero-width assertion that matches only at the
749 beginning or end of a word. A word is defined as a sequence of alphanumeric
750 characters, so the end of a word is indicated by whitespace or a
751 non-alphanumeric character.
752
753 The following example matches ``class`` only when it's a complete word; it won't
754 match when it's contained inside another word. ::
755
756 >>> p = re.compile(r'\bclass\b')
757 >>> print(p.search('no class at all'))
758 <re.Match object; span=(3, 8), match='class'>
759 >>> print(p.search('the declassified algorithm'))
760 None
761 >>> print(p.search('one subclass is'))
762 None
763
764 There are two subtleties you should remember when using this special sequence.
765 First, this is the worst collision between Python's string literals and regular
766 expression sequences. In Python's string literals, ``\b`` is the backspace
767 character, ASCII value 8. If you're not using raw strings, then Python will
768 convert the ``\b`` to a backspace, and your RE won't match as you expect it to.
769 The following example looks the same as our previous RE, but omits the ``'r'``
770 in front of the RE string. ::
771
772 >>> p = re.compile('\bclass\b')
773 >>> print(p.search('no class at all'))
774 None
775 >>> print(p.search('\b' + 'class' + '\b'))
776 <re.Match object; span=(0, 7), match='\x08class\x08'>
777
778 Second, inside a character class, where there's no use for this assertion,
779 ``\b`` represents the backspace character, for compatibility with Python's
780 string literals.
781
782 ``\B``
783 Another zero-width assertion, this is the opposite of ``\b``, only matching when
784 the current position is not at a word boundary.
785
786
787 Grouping
788 --------
789
790 Frequently you need to obtain more information than just whether the RE matched
791 or not. Regular expressions are often used to dissect strings by writing a RE
792 divided into several subgroups which match different components of interest.
793 For example, an RFC-822 header line is divided into a header name and a value,
794 separated by a ``':'``, like this:
795
796 .. code-block:: none
797
798 From: [email protected]
799 User-Agent: Thunderbird 1.5.0.9 (X11/20061227)
800 MIME-Version: 1.0
801 To: [email protected]
802
803 This can be handled by writing a regular expression which matches an entire
804 header line, and has one group which matches the header name, and another group
805 which matches the header's value.
806
807 Groups are marked by the ``'('``, ``')'`` metacharacters. ``'('`` and ``')'``
808 have much the same meaning as they do in mathematical expressions; they group
809 together the expressions contained inside them, and you can repeat the contents
810 of a group with a quantifier, such as ``*``, ``+``, ``?``, or
811 ``{m,n}``. For example, ``(ab)*`` will match zero or more repetitions of
812 ``ab``. ::
813
814 >>> p = re.compile('(ab)*')
815 >>> print(p.match('ababababab').span())
816 (0, 10)
817
818 Groups indicated with ``'('``, ``')'`` also capture the starting and ending
819 index of the text that they match; this can be retrieved by passing an argument
820 to :meth:`~re.Match.group`, :meth:`~re.Match.start`, :meth:`~re.Match.end`, and
821 :meth:`~re.Match.span`. Groups are
822 numbered starting with 0. Group 0 is always present; it's the whole RE, so
823 :ref:`match object <match-objects>` methods all have group 0 as their default
824 argument. Later we'll see how to express groups that don't capture the span
825 of text that they match. ::
826
827 >>> p = re.compile('(a)b')
828 >>> m = p.match('ab')
829 >>> m.group()
830 'ab'
831 >>> m.group(0)
832 'ab'
833
834 Subgroups are numbered from left to right, from 1 upward. Groups can be nested;
835 to determine the number, just count the opening parenthesis characters, going
836 from left to right. ::
837
838 >>> p = re.compile('(a(b)c)d')
839 >>> m = p.match('abcd')
840 >>> m.group(0)
841 'abcd'
842 >>> m.group(1)
843 'abc'
844 >>> m.group(2)
845 'b'
846
847 :meth:`~re.Match.group` can be passed multiple group numbers at a time, in which case it
848 will return a tuple containing the corresponding values for those groups. ::
849
850 >>> m.group(2,1,2)
851 ('b', 'abc', 'b')
852
853 The :meth:`~re.Match.groups` method returns a tuple containing the strings for all the
854 subgroups, from 1 up to however many there are. ::
855
856 >>> m.groups()
857 ('abc', 'b')
858
859 Backreferences in a pattern allow you to specify that the contents of an earlier
860 capturing group must also be found at the current location in the string. For
861 example, ``\1`` will succeed if the exact contents of group 1 can be found at
862 the current position, and fails otherwise. Remember that Python's string
863 literals also use a backslash followed by numbers to allow including arbitrary
864 characters in a string, so be sure to use a raw string when incorporating
865 backreferences in a RE.
866
867 For example, the following RE detects doubled words in a string. ::
868
869 >>> p = re.compile(r'\b(\w+)\s+\1\b')
870 >>> p.search('Paris in the the spring').group()
871 'the the'
872
873 Backreferences like this aren't often useful for just searching through a string
874 --- there are few text formats which repeat data in this way --- but you'll soon
875 find out that they're *very* useful when performing string substitutions.
876
877
878 Non-capturing and named groups
879 ------------------------------
880
881 Elaborate REs may use many groups, both to capture substrings of interest, and
882 to group and structure the RE itself. In complex REs, it becomes difficult to
883 keep track of the group numbers. There are two features which help with this
884 problem. Both of them use a common syntax for regular expression extensions, so
885 we'll look at that first.
886
887 Perl 5 is well known for its powerful additions to standard regular expressions.
888 For these new features the Perl developers couldn't choose new single-keystroke metacharacters
889 or new special sequences beginning with ``\`` without making Perl's regular
890 expressions confusingly different from standard REs. If they chose ``&`` as a
891 new metacharacter, for example, old expressions would be assuming that ``&`` was
892 a regular character and wouldn't have escaped it by writing ``\&`` or ``[&]``.
893
894 The solution chosen by the Perl developers was to use ``(?...)`` as the
895 extension syntax. ``?`` immediately after a parenthesis was a syntax error
896 because the ``?`` would have nothing to repeat, so this didn't introduce any
897 compatibility problems. The characters immediately after the ``?`` indicate
898 what extension is being used, so ``(?=foo)`` is one thing (a positive lookahead
899 assertion) and ``(?:foo)`` is something else (a non-capturing group containing
900 the subexpression ``foo``).
901
902 Python supports several of Perl's extensions and adds an extension
903 syntax to Perl's extension syntax. If the first character after the
904 question mark is a ``P``, you know that it's an extension that's
905 specific to Python.
906
907 Now that we've looked at the general extension syntax, we can return
908 to the features that simplify working with groups in complex REs.
909
910 Sometimes you'll want to use a group to denote a part of a regular expression,
911 but aren't interested in retrieving the group's contents. You can make this fact
912 explicit by using a non-capturing group: ``(?:...)``, where you can replace the
913 ``...`` with any other regular expression. ::
914
915 >>> m = re.match("([abc])+", "abc")
916 >>> m.groups()
917 ('c',)
918 >>> m = re.match("(?:[abc])+", "abc")
919 >>> m.groups()
920 ()
921
922 Except for the fact that you can't retrieve the contents of what the group
923 matched, a non-capturing group behaves exactly the same as a capturing group;
924 you can put anything inside it, repeat it with a repetition metacharacter such
925 as ``*``, and nest it within other groups (capturing or non-capturing).
926 ``(?:...)`` is particularly useful when modifying an existing pattern, since you
927 can add new groups without changing how all the other groups are numbered. It
928 should be mentioned that there's no performance difference in searching between
929 capturing and non-capturing groups; neither form is any faster than the other.
930
931 A more significant feature is named groups: instead of referring to them by
932 numbers, groups can be referenced by a name.
933
934 The syntax for a named group is one of the Python-specific extensions:
935 ``(?P<name>...)``. *name* is, obviously, the name of the group. Named groups
936 behave exactly like capturing groups, and additionally associate a name
937 with a group. The :ref:`match object <match-objects>` methods that deal with
938 capturing groups all accept either integers that refer to the group by number
939 or strings that contain the desired group's name. Named groups are still
940 given numbers, so you can retrieve information about a group in two ways::
941
942 >>> p = re.compile(r'(?P<word>\b\w+\b)')
943 >>> m = p.search( '(((( Lots of punctuation )))' )
944 >>> m.group('word')
945 'Lots'
946 >>> m.group(1)
947 'Lots'
948
949 Additionally, you can retrieve named groups as a dictionary with
950 :meth:`~re.Match.groupdict`::
951
952 >>> m = re.match(r'(?P<first>\w+) (?P<last>\w+)', 'Jane Doe')
953 >>> m.groupdict()
954 {'first': 'Jane', 'last': 'Doe'}
955
956 Named groups are handy because they let you use easily remembered names, instead
957 of having to remember numbers. Here's an example RE from the :mod:`imaplib`
958 module::
959
960 InternalDate = re.compile(r'INTERNALDATE "'
961 r'(?P<day>[ 123][0-9])-(?P<mon>[A-Z][a-z][a-z])-'
962 r'(?P<year>[0-9][0-9][0-9][0-9])'
963 r' (?P<hour>[0-9][0-9]):(?P<min>[0-9][0-9]):(?P<sec>[0-9][0-9])'
964 r' (?P<zonen>[-+])(?P<zoneh>[0-9][0-9])(?P<zonem>[0-9][0-9])'
965 r'"')
966
967 It's obviously much easier to retrieve ``m.group('zonem')``, instead of having
968 to remember to retrieve group 9.
969
970 The syntax for backreferences in an expression such as ``(...)\1`` refers to the
971 number of the group. There's naturally a variant that uses the group name
972 instead of the number. This is another Python extension: ``(?P=name)`` indicates
973 that the contents of the group called *name* should again be matched at the
974 current point. The regular expression for finding doubled words,
975 ``\b(\w+)\s+\1\b`` can also be written as ``\b(?P<word>\w+)\s+(?P=word)\b``::
976
977 >>> p = re.compile(r'\b(?P<word>\w+)\s+(?P=word)\b')
978 >>> p.search('Paris in the the spring').group()
979 'the the'
980
981
982 Lookahead assertions
983 --------------------
984
985 Another zero-width assertion is the lookahead assertion. Lookahead assertions
986 are available in both positive and negative form, and look like this:
987
988 ``(?=...)``
989 Positive lookahead assertion. This succeeds if the contained regular
990 expression, represented here by ``...``, successfully matches at the current
991 location, and fails otherwise. But, once the contained expression has been
992 tried, the matching engine doesn't advance at all; the rest of the pattern is
993 tried right where the assertion started.
994
995 ``(?!...)``
996 Negative lookahead assertion. This is the opposite of the positive assertion;
997 it succeeds if the contained expression *doesn't* match at the current position
998 in the string.
999
1000 To make this concrete, let's look at a case where a lookahead is useful.
1001 Consider a simple pattern to match a filename and split it apart into a base
1002 name and an extension, separated by a ``.``. For example, in ``news.rc``,
1003 ``news`` is the base name, and ``rc`` is the filename's extension.
1004
1005 The pattern to match this is quite simple:
1006
1007 ``.*[.].*$``
1008
1009 Notice that the ``.`` needs to be treated specially because it's a
1010 metacharacter, so it's inside a character class to only match that
1011 specific character. Also notice the trailing ``$``; this is added to
1012 ensure that all the rest of the string must be included in the
1013 extension. This regular expression matches ``foo.bar`` and
1014 ``autoexec.bat`` and ``sendmail.cf`` and ``printers.conf``.
1015
1016 Now, consider complicating the problem a bit; what if you want to match
1017 filenames where the extension is not ``bat``? Some incorrect attempts:
1018
1019 ``.*[.][^b].*$``
1020
1021 The first attempt above tries to exclude ``bat`` by requiring
1022 that the first character of the extension is not a ``b``. This is wrong,
1023 because the pattern also doesn't match ``foo.bar``.
1024
1025 ``.*[.]([^b]..|.[^a].|..[^t])$``
1026
1027 The expression gets messier when you try to patch up the first solution by
1028 requiring one of the following cases to match: the first character of the
1029 extension isn't ``b``; the second character isn't ``a``; or the third character
1030 isn't ``t``. This accepts ``foo.bar`` and rejects ``autoexec.bat``, but it
1031 requires a three-letter extension and won't accept a filename with a two-letter
1032 extension such as ``sendmail.cf``. We'll complicate the pattern again in an
1033 effort to fix it.
1034
1035 ``.*[.]([^b].?.?|.[^a]?.?|..?[^t]?)$``
1036
1037 In the third attempt, the second and third letters are all made optional in
1038 order to allow matching extensions shorter than three characters, such as
1039 ``sendmail.cf``.
1040
1041 The pattern's getting really complicated now, which makes it hard to read and
1042 understand. Worse, if the problem changes and you want to exclude both ``bat``
1043 and ``exe`` as extensions, the pattern would get even more complicated and
1044 confusing.
1045
1046 A negative lookahead cuts through all this confusion:
1047
1048 ``.*[.](?!bat$)[^.]*$``
1049
1050 The negative lookahead means: if the expression ``bat``
1051 doesn't match at this point, try the rest of the pattern; if ``bat$`` does
1052 match, the whole pattern will fail. The trailing ``$`` is required to ensure
1053 that something like ``sample.batch``, where the extension only starts with
1054 ``bat``, will be allowed. The ``[^.]*`` makes sure that the pattern works
1055 when there are multiple dots in the filename.
1056
1057 Excluding another filename extension is now easy; simply add it as an
1058 alternative inside the assertion. The following pattern excludes filenames that
1059 end in either ``bat`` or ``exe``:
1060
1061 ``.*[.](?!bat$|exe$)[^.]*$``
1062
1063
1064 Modifying strings
1065 =================
1066
1067 Up to this point, we've simply performed searches against a static string.
1068 Regular expressions are also commonly used to modify strings in various ways,
1069 using the following pattern methods:
1070
1071 +------------------+-----------------------------------------------+
1072 | Method/Attribute | Purpose |
1073 +==================+===============================================+
1074 | ``split()`` | Split the string into a list, splitting it |
1075 | | wherever the RE matches |
1076 +------------------+-----------------------------------------------+
1077 | ``sub()`` | Find all substrings where the RE matches, and |
1078 | | replace them with a different string |
1079 +------------------+-----------------------------------------------+
1080 | ``subn()`` | Does the same thing as :meth:`!sub`, but |
1081 | | returns the new string and the number of |
1082 | | replacements |
1083 +------------------+-----------------------------------------------+
1084
1085
1086 Splitting strings
1087 -----------------
1088
1089 The :meth:`~re.Pattern.split` method of a pattern splits a string apart
1090 wherever the RE matches, returning a list of the pieces. It's similar to the
1091 :meth:`~str.split` method of strings but provides much more generality in the
1092 delimiters that you can split by; string :meth:`!split` only supports splitting by
1093 whitespace or by a fixed string. As you'd expect, there's a module-level
1094 :func:`re.split` function, too.
1095
1096
1097 .. method:: .split(string [, maxsplit=0])
1098 :noindex:
1099
1100 Split *string* by the matches of the regular expression. If capturing
1101 parentheses are used in the RE, then their contents will also be returned as
1102 part of the resulting list. If *maxsplit* is nonzero, at most *maxsplit* splits
1103 are performed.
1104
1105 You can limit the number of splits made, by passing a value for *maxsplit*.
1106 When *maxsplit* is nonzero, at most *maxsplit* splits will be made, and the
1107 remainder of the string is returned as the final element of the list. In the
1108 following example, the delimiter is any sequence of non-alphanumeric characters.
1109 ::
1110
1111 >>> p = re.compile(r'\W+')
1112 >>> p.split('This is a test, short and sweet, of split().')
1113 ['This', 'is', 'a', 'test', 'short', 'and', 'sweet', 'of', 'split', '']
1114 >>> p.split('This is a test, short and sweet, of split().', 3)
1115 ['This', 'is', 'a', 'test, short and sweet, of split().']
1116
1117 Sometimes you're not only interested in what the text between delimiters is, but
1118 also need to know what the delimiter was. If capturing parentheses are used in
1119 the RE, then their values are also returned as part of the list. Compare the
1120 following calls::
1121
1122 >>> p = re.compile(r'\W+')
1123 >>> p2 = re.compile(r'(\W+)')
1124 >>> p.split('This... is a test.')
1125 ['This', 'is', 'a', 'test', '']
1126 >>> p2.split('This... is a test.')
1127 ['This', '... ', 'is', ' ', 'a', ' ', 'test', '.', '']
1128
1129 The module-level function :func:`re.split` adds the RE to be used as the first
1130 argument, but is otherwise the same. ::
1131
1132 >>> re.split(r'[\W]+', 'Words, words, words.')
1133 ['Words', 'words', 'words', '']
1134 >>> re.split(r'([\W]+)', 'Words, words, words.')
1135 ['Words', ', ', 'words', ', ', 'words', '.', '']
1136 >>> re.split(r'[\W]+', 'Words, words, words.', 1)
1137 ['Words', 'words, words.']
1138
1139
1140 Search and replace
1141 ------------------
1142
1143 Another common task is to find all the matches for a pattern, and replace them
1144 with a different string. The :meth:`~re.Pattern.sub` method takes a replacement value,
1145 which can be either a string or a function, and the string to be processed.
1146
1147 .. method:: .sub(replacement, string[, count=0])
1148 :noindex:
1149
1150 Returns the string obtained by replacing the leftmost non-overlapping
1151 occurrences of the RE in *string* by the replacement *replacement*. If the
1152 pattern isn't found, *string* is returned unchanged.
1153
1154 The optional argument *count* is the maximum number of pattern occurrences to be
1155 replaced; *count* must be a non-negative integer. The default value of 0 means
1156 to replace all occurrences.
1157
1158 Here's a simple example of using the :meth:`~re.Pattern.sub` method. It replaces colour
1159 names with the word ``colour``::
1160
1161 >>> p = re.compile('(blue|white|red)')
1162 >>> p.sub('colour', 'blue socks and red shoes')
1163 'colour socks and colour shoes'
1164 >>> p.sub('colour', 'blue socks and red shoes', count=1)
1165 'colour socks and red shoes'
1166
1167 The :meth:`~re.Pattern.subn` method does the same work, but returns a 2-tuple containing the
1168 new string value and the number of replacements that were performed::
1169
1170 >>> p = re.compile('(blue|white|red)')
1171 >>> p.subn('colour', 'blue socks and red shoes')
1172 ('colour socks and colour shoes', 2)
1173 >>> p.subn('colour', 'no colours at all')
1174 ('no colours at all', 0)
1175
1176 Empty matches are replaced only when they're not adjacent to a previous empty match.
1177 ::
1178
1179 >>> p = re.compile('x*')
1180 >>> p.sub('-', 'abxd')
1181 '-a-b--d-'
1182
1183 If *replacement* is a string, any backslash escapes in it are processed. That
1184 is, ``\n`` is converted to a single newline character, ``\r`` is converted to a
1185 carriage return, and so forth. Unknown escapes such as ``\&`` are left alone.
1186 Backreferences, such as ``\6``, are replaced with the substring matched by the
1187 corresponding group in the RE. This lets you incorporate portions of the
1188 original text in the resulting replacement string.
1189
1190 This example matches the word ``section`` followed by a string enclosed in
1191 ``{``, ``}``, and changes ``section`` to ``subsection``::
1192
1193 >>> p = re.compile('section{ ( [^}]* ) }', re.VERBOSE)
1194 >>> p.sub(r'subsection{\1}','section{First} section{second}')
1195 'subsection{First} subsection{second}'
1196
1197 There's also a syntax for referring to named groups as defined by the
1198 ``(?P<name>...)`` syntax. ``\g<name>`` will use the substring matched by the
1199 group named ``name``, and ``\g<number>`` uses the corresponding group number.
1200 ``\g<2>`` is therefore equivalent to ``\2``, but isn't ambiguous in a
1201 replacement string such as ``\g<2>0``. (``\20`` would be interpreted as a
1202 reference to group 20, not a reference to group 2 followed by the literal
1203 character ``'0'``.) The following substitutions are all equivalent, but use all
1204 three variations of the replacement string. ::
1205
1206 >>> p = re.compile('section{ (?P<name> [^}]* ) }', re.VERBOSE)
1207 >>> p.sub(r'subsection{\1}','section{First}')
1208 'subsection{First}'
1209 >>> p.sub(r'subsection{\g<1>}','section{First}')
1210 'subsection{First}'
1211 >>> p.sub(r'subsection{\g<name>}','section{First}')
1212 'subsection{First}'
1213
1214 *replacement* can also be a function, which gives you even more control. If
1215 *replacement* is a function, the function is called for every non-overlapping
1216 occurrence of *pattern*. On each call, the function is passed a
1217 :ref:`match object <match-objects>` argument for the match and can use this
1218 information to compute the desired replacement string and return it.
1219
1220 In the following example, the replacement function translates decimals into
1221 hexadecimal::
1222
1223 >>> def hexrepl(match):
1224 ... "Return the hex string for a decimal number"
1225 ... value = int(match.group())
1226 ... return hex(value)
1227 ...
1228 >>> p = re.compile(r'\d+')
1229 >>> p.sub(hexrepl, 'Call 65490 for printing, 49152 for user code.')
1230 'Call 0xffd2 for printing, 0xc000 for user code.'
1231
1232 When using the module-level :func:`re.sub` function, the pattern is passed as
1233 the first argument. The pattern may be provided as an object or as a string; if
1234 you need to specify regular expression flags, you must either use a
1235 pattern object as the first parameter, or use embedded modifiers in the
1236 pattern string, e.g. ``sub("(?i)b+", "x", "bbbb BBBB")`` returns ``'x x'``.
1237
1238
1239 Common problems
1240 ===============
1241
1242 Regular expressions are a powerful tool for some applications, but in some ways
1243 their behaviour isn't intuitive and at times they don't behave the way you may
1244 expect them to. This section will point out some of the most common pitfalls.
1245
1246
1247 Use string methods
1248 ------------------
1249
1250 Sometimes using the :mod:`re` module is a mistake. If you're matching a fixed
1251 string, or a single character class, and you're not using any :mod:`re` features
1252 such as the :const:`~re.IGNORECASE` flag, then the full power of regular expressions
1253 may not be required. Strings have several methods for performing operations with
1254 fixed strings and they're usually much faster, because the implementation is a
1255 single small C loop that's been optimized for the purpose, instead of the large,
1256 more generalized regular expression engine.
1257
1258 One example might be replacing a single fixed string with another one; for
1259 example, you might replace ``word`` with ``deed``. :func:`re.sub` seems like the
1260 function to use for this, but consider the :meth:`~str.replace` method. Note that
1261 :meth:`!replace` will also replace ``word`` inside words, turning ``swordfish``
1262 into ``sdeedfish``, but the naive RE ``word`` would have done that, too. (To
1263 avoid performing the substitution on parts of words, the pattern would have to
1264 be ``\bword\b``, in order to require that ``word`` have a word boundary on
1265 either side. This takes the job beyond :meth:`!replace`'s abilities.)
1266
1267 Another common task is deleting every occurrence of a single character from a
1268 string or replacing it with another single character. You might do this with
1269 something like ``re.sub('\n', ' ', S)``, but :meth:`~str.translate` is capable of
1270 doing both tasks and will be faster than any regular expression operation can
1271 be.
1272
1273 In short, before turning to the :mod:`re` module, consider whether your problem
1274 can be solved with a faster and simpler string method.
1275
1276
1277 match() versus search()
1278 -----------------------
1279
1280 The :func:`~re.match` function only checks if the RE matches at the beginning of the
1281 string while :func:`~re.search` will scan forward through the string for a match.
1282 It's important to keep this distinction in mind. Remember, :func:`!match` will
1283 only report a successful match which will start at 0; if the match wouldn't
1284 start at zero, :func:`!match` will *not* report it. ::
1285
1286 >>> print(re.match('super', 'superstition').span())
1287 (0, 5)
1288 >>> print(re.match('super', 'insuperable'))
1289 None
1290
1291 On the other hand, :func:`~re.search` will scan forward through the string,
1292 reporting the first match it finds. ::
1293
1294 >>> print(re.search('super', 'superstition').span())
1295 (0, 5)
1296 >>> print(re.search('super', 'insuperable').span())
1297 (2, 7)
1298
1299 Sometimes you'll be tempted to keep using :func:`re.match`, and just add ``.*``
1300 to the front of your RE. Resist this temptation and use :func:`re.search`
1301 instead. The regular expression compiler does some analysis of REs in order to
1302 speed up the process of looking for a match. One such analysis figures out what
1303 the first character of a match must be; for example, a pattern starting with
1304 ``Crow`` must match starting with a ``'C'``. The analysis lets the engine
1305 quickly scan through the string looking for the starting character, only trying
1306 the full match if a ``'C'`` is found.
1307
1308 Adding ``.*`` defeats this optimization, requiring scanning to the end of the
1309 string and then backtracking to find a match for the rest of the RE. Use
1310 :func:`re.search` instead.
1311
1312
1313 Greedy versus non-greedy
1314 ------------------------
1315
1316 When repeating a regular expression, as in ``a*``, the resulting action is to
1317 consume as much of the pattern as possible. This fact often bites you when
1318 you're trying to match a pair of balanced delimiters, such as the angle brackets
1319 surrounding an HTML tag. The naive pattern for matching a single HTML tag
1320 doesn't work because of the greedy nature of ``.*``. ::
1321
1322 >>> s = '<html><head><title>Title</title>'
1323 >>> len(s)
1324 32
1325 >>> print(re.match('<.*>', s).span())
1326 (0, 32)
1327 >>> print(re.match('<.*>', s).group())
1328 <html><head><title>Title</title>
1329
1330 The RE matches the ``'<'`` in ``'<html>'``, and the ``.*`` consumes the rest of
1331 the string. There's still more left in the RE, though, and the ``>`` can't
1332 match at the end of the string, so the regular expression engine has to
1333 backtrack character by character until it finds a match for the ``>``. The
1334 final match extends from the ``'<'`` in ``'<html>'`` to the ``'>'`` in
1335 ``'</title>'``, which isn't what you want.
1336
1337 In this case, the solution is to use the non-greedy quantifiers ``*?``, ``+?``,
1338 ``??``, or ``{m,n}?``, which match as *little* text as possible. In the above
1339 example, the ``'>'`` is tried immediately after the first ``'<'`` matches, and
1340 when it fails, the engine advances a character at a time, retrying the ``'>'``
1341 at every step. This produces just the right result::
1342
1343 >>> print(re.match('<.*?>', s).group())
1344 <html>
1345
1346 (Note that parsing HTML or XML with regular expressions is painful.
1347 Quick-and-dirty patterns will handle common cases, but HTML and XML have special
1348 cases that will break the obvious regular expression; by the time you've written
1349 a regular expression that handles all of the possible cases, the patterns will
1350 be *very* complicated. Use an HTML or XML parser module for such tasks.)
1351
1352
1353 Using re.VERBOSE
1354 ----------------
1355
1356 By now you've probably noticed that regular expressions are a very compact
1357 notation, but they're not terribly readable. REs of moderate complexity can
1358 become lengthy collections of backslashes, parentheses, and metacharacters,
1359 making them difficult to read and understand.
1360
1361 For such REs, specifying the :const:`re.VERBOSE` flag when compiling the regular
1362 expression can be helpful, because it allows you to format the regular
1363 expression more clearly.
1364
1365 The ``re.VERBOSE`` flag has several effects. Whitespace in the regular
1366 expression that *isn't* inside a character class is ignored. This means that an
1367 expression such as ``dog | cat`` is equivalent to the less readable ``dog|cat``,
1368 but ``[a b]`` will still match the characters ``'a'``, ``'b'``, or a space. In
1369 addition, you can also put comments inside a RE; comments extend from a ``#``
1370 character to the next newline. When used with triple-quoted strings, this
1371 enables REs to be formatted more neatly::
1372
1373 pat = re.compile(r"""
1374 \s* # Skip leading whitespace
1375 (?P<header>[^:]+) # Header name
1376 \s* : # Whitespace, and a colon
1377 (?P<value>.*?) # The header's value -- *? used to
1378 # lose the following trailing whitespace
1379 \s*$ # Trailing whitespace to end-of-line
1380 """, re.VERBOSE)
1381
1382 This is far more readable than::
1383
1384 pat = re.compile(r"\s*(?P<header>[^:]+)\s*:(?P<value>.*?)\s*$")
1385
1386
1387 Feedback
1388 ========
1389
1390 Regular expressions are a complicated topic. Did this document help you
1391 understand them? Were there parts that were unclear, or problems you
1392 encountered that weren't covered here? If so, please send suggestions for
1393 improvements to the :ref:`issue tracker <using-the-tracker>`.
1394
1395 The most complete book on regular expressions is almost certainly Jeffrey
1396 Friedl's Mastering Regular Expressions, published by O'Reilly. Unfortunately,
1397 it exclusively concentrates on Perl and Java's flavours of regular expressions,
1398 and doesn't contain any Python material at all, so it won't be useful as a
1399 reference for programming in Python. (The first edition covered Python's
1400 now-removed :mod:`!regex` module, which won't help you much.) Consider checking
1401 it out from your library.