1
0
mirror of https://github.com/django/django.git synced 2024-12-26 02:56:25 +00:00
django/django/utils/regex_helper.py
2012-08-07 12:00:24 +02:00

343 lines
12 KiB
Python

"""
Functions for reversing a regular expression (used in reverse URL resolving).
Used internally by Django and not intended for external use.
This is not, and is not intended to be, a complete reg-exp decompiler. It
should be good enough for a large class of URLS, however.
"""
from __future__ import unicode_literals
from django.utils import six
from django.utils.six.moves import zip
# Mapping of an escape character to a representative of that class. So, e.g.,
# "\w" is replaced by "x" in a reverse URL. A value of None means to ignore
# this sequence. Any missing key is mapped to itself.
ESCAPE_MAPPINGS = {
"A": None,
"b": None,
"B": None,
"d": "0",
"D": "x",
"s": " ",
"S": "x",
"w": "x",
"W": "!",
"Z": None,
}
class Choice(list):
"""
Used to represent multiple possibilities at this point in a pattern string.
We use a distinguished type, rather than a list, so that the usage in the
code is clear.
"""
class Group(list):
"""
Used to represent a capturing group in the pattern string.
"""
class NonCapture(list):
"""
Used to represent a non-capturing group in the pattern string.
"""
def normalize(pattern):
"""
Given a reg-exp pattern, normalizes it to an iterable of forms that
suffice for reverse matching. This does the following:
(1) For any repeating sections, keeps the minimum number of occurrences
permitted (this means zero for optional groups).
(2) If an optional group includes parameters, include one occurrence of
that group (along with the zero occurrence case from step (1)).
(3) Select the first (essentially an arbitrary) element from any character
class. Select an arbitrary character for any unordered class (e.g. '.'
or '\w') in the pattern.
(5) Ignore comments and any of the reg-exp flags that won't change
what we construct ("iLmsu"). "(?x)" is an error, however.
(6) Raise an error on all other non-capturing (?...) forms (e.g.
look-ahead and look-behind matches) and any disjunctive ('|')
constructs.
Django's URLs for forward resolving are either all positional arguments or
all keyword arguments. That is assumed here, as well. Although reverse
resolving can be done using positional args when keyword args are
specified, the two cannot be mixed in the same reverse() call.
"""
# Do a linear scan to work out the special features of this pattern. The
# idea is that we scan once here and collect all the information we need to
# make future decisions.
result = []
non_capturing_groups = []
consume_next = True
pattern_iter = next_char(iter(pattern))
num_args = 0
# A "while" loop is used here because later on we need to be able to peek
# at the next character and possibly go around without consuming another
# one at the top of the loop.
try:
ch, escaped = next(pattern_iter)
except StopIteration:
return [('', [])]
try:
while True:
if escaped:
result.append(ch)
elif ch == '.':
# Replace "any character" with an arbitrary representative.
result.append(".")
elif ch == '|':
# FIXME: One day we'll should do this, but not in 1.0.
raise NotImplementedError
elif ch == "^":
pass
elif ch == '$':
break
elif ch == ')':
# This can only be the end of a non-capturing group, since all
# other unescaped parentheses are handled by the grouping
# section later (and the full group is handled there).
#
# We regroup everything inside the capturing group so that it
# can be quantified, if necessary.
start = non_capturing_groups.pop()
inner = NonCapture(result[start:])
result = result[:start] + [inner]
elif ch == '[':
# Replace ranges with the first character in the range.
ch, escaped = next(pattern_iter)
result.append(ch)
ch, escaped = next(pattern_iter)
while escaped or ch != ']':
ch, escaped = next(pattern_iter)
elif ch == '(':
# Some kind of group.
ch, escaped = next(pattern_iter)
if ch != '?' or escaped:
# A positional group
name = "_%d" % num_args
num_args += 1
result.append(Group((("%%(%s)s" % name), name)))
walk_to_end(ch, pattern_iter)
else:
ch, escaped = next(pattern_iter)
if ch in "iLmsu#":
# All of these are ignorable. Walk to the end of the
# group.
walk_to_end(ch, pattern_iter)
elif ch == ':':
# Non-capturing group
non_capturing_groups.append(len(result))
elif ch != 'P':
# Anything else, other than a named group, is something
# we cannot reverse.
raise ValueError("Non-reversible reg-exp portion: '(?%s'" % ch)
else:
ch, escaped = next(pattern_iter)
if ch not in ('<', '='):
raise ValueError("Non-reversible reg-exp portion: '(?P%s'" % ch)
# We are in a named capturing group. Extra the name and
# then skip to the end.
if ch == '<':
terminal_char = '>'
# We are in a named backreference.
else:
terminal_char = ')'
name = []
ch, escaped = next(pattern_iter)
while ch != terminal_char:
name.append(ch)
ch, escaped = next(pattern_iter)
param = ''.join(name)
# Named backreferences have already consumed the
# parenthesis.
if terminal_char != ')':
result.append(Group((("%%(%s)s" % param), param)))
walk_to_end(ch, pattern_iter)
else:
result.append(Group((("%%(%s)s" % param), None)))
elif ch in "*?+{":
# Quanitifers affect the previous item in the result list.
count, ch = get_quantifier(ch, pattern_iter)
if ch:
# We had to look ahead, but it wasn't need to compute the
# quanitifer, so use this character next time around the
# main loop.
consume_next = False
if count == 0:
if contains(result[-1], Group):
# If we are quantifying a capturing group (or
# something containing such a group) and the minimum is
# zero, we must also handle the case of one occurrence
# being present. All the quantifiers (except {0,0},
# which we conveniently ignore) that have a 0 minimum
# also allow a single occurrence.
result[-1] = Choice([None, result[-1]])
else:
result.pop()
elif count > 1:
result.extend([result[-1]] * (count - 1))
else:
# Anything else is a literal.
result.append(ch)
if consume_next:
ch, escaped = next(pattern_iter)
else:
consume_next = True
except StopIteration:
pass
except NotImplementedError:
# A case of using the disjunctive form. No results for you!
return [('', [])]
return list(zip(*flatten_result(result)))
def next_char(input_iter):
"""
An iterator that yields the next character from "pattern_iter", respecting
escape sequences. An escaped character is replaced by a representative of
its class (e.g. \w -> "x"). If the escaped character is one that is
skipped, it is not returned (the next character is returned instead).
Yields the next character, along with a boolean indicating whether it is a
raw (unescaped) character or not.
"""
for ch in input_iter:
if ch != '\\':
yield ch, False
continue
ch = next(input_iter)
representative = ESCAPE_MAPPINGS.get(ch, ch)
if representative is None:
continue
yield representative, True
def walk_to_end(ch, input_iter):
"""
The iterator is currently inside a capturing group. We want to walk to the
close of this group, skipping over any nested groups and handling escaped
parentheses correctly.
"""
if ch == '(':
nesting = 1
else:
nesting = 0
for ch, escaped in input_iter:
if escaped:
continue
elif ch == '(':
nesting += 1
elif ch == ')':
if not nesting:
return
nesting -= 1
def get_quantifier(ch, input_iter):
"""
Parse a quantifier from the input, where "ch" is the first character in the
quantifier.
Returns the minimum number of occurences permitted by the quantifier and
either None or the next character from the input_iter if the next character
is not part of the quantifier.
"""
if ch in '*?+':
try:
ch2, escaped = next(input_iter)
except StopIteration:
ch2 = None
if ch2 == '?':
ch2 = None
if ch == '+':
return 1, ch2
return 0, ch2
quant = []
while ch != '}':
ch, escaped = next(input_iter)
quant.append(ch)
quant = quant[:-1]
values = ''.join(quant).split(',')
# Consume the trailing '?', if necessary.
try:
ch, escaped = next(input_iter)
except StopIteration:
ch = None
if ch == '?':
ch = None
return int(values[0]), ch
def contains(source, inst):
"""
Returns True if the "source" contains an instance of "inst". False,
otherwise.
"""
if isinstance(source, inst):
return True
if isinstance(source, NonCapture):
for elt in source:
if contains(elt, inst):
return True
return False
def flatten_result(source):
"""
Turns the given source sequence into a list of reg-exp possibilities and
their arguments. Returns a list of strings and a list of argument lists.
Each of the two lists will be of the same length.
"""
if source is None:
return [''], [[]]
if isinstance(source, Group):
if source[1] is None:
params = []
else:
params = [source[1]]
return [source[0]], [params]
result = ['']
result_args = [[]]
pos = last = 0
for pos, elt in enumerate(source):
if isinstance(elt, six.string_types):
continue
piece = ''.join(source[last:pos])
if isinstance(elt, Group):
piece += elt[0]
param = elt[1]
else:
param = None
last = pos + 1
for i in range(len(result)):
result[i] += piece
if param:
result_args[i].append(param)
if isinstance(elt, (Choice, NonCapture)):
if isinstance(elt, NonCapture):
elt = [elt]
inner_result, inner_args = [], []
for item in elt:
res, args = flatten_result(item)
inner_result.extend(res)
inner_args.extend(args)
new_result = []
new_args = []
for item, args in zip(result, result_args):
for i_item, i_args in zip(inner_result, inner_args):
new_result.append(item + i_item)
new_args.append(args[:] + i_args)
result = new_result
result_args = new_args
if pos >= last:
piece = ''.join(source[last:])
for i in range(len(result)):
result[i] += piece
return result, result_args