Started to work on the regex reverse-engineering phase.

git-svn-id: http://code.djangoproject.com/svn/django/trunk@7850 bcc190cf-cafb-0310-a4f2-bffc1f526a37

django/utils/regex_helper.py

@@ -0,0 +1,123 @@
+"""
+Functions for reversing a regular expression (used in reverse URL resolving).
+
+This is not, and is not intended to be, a complete reg-exp decompiler. It
+should be good enough for almost all sane URLs.
+"""
+
+import re
+from bisect import bisect
+
+GROUP_CLASS = re.compile(r'''\((?:
+        (?P<positional>[^?])|       # Unnamed (positional) capturing group.
+        \?(?:
+            P<(?P<named>[\w]+)>(?P<contents>.*)|    # Named capturing group.
+            P=(?P<repeat>.+)|       # Repeat of a previous named group.
+            (?P<grouping>:)|        # Non-capturing grouping parens.
+            (?P<comment>\#)|        # Comment group
+            (?P<illegal>.)          # Anything else (which will be an error)
+        )
+    ).*\)''', re.VERBOSE)
+
+def normalize(pattern):
+    """
+    Given a reg-exp pattern, normalizes it to a list 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.
+    (4) Take the first alternative in any '|' division, unless other
+    alternatives would involve different parameters.
+    (5) Ignore comments. Error on all other non-capturing (?...) forms (e.g.
+    look-ahead and look-behind matches).
+
+    Returns a list of tuples, each tuple containing (a) a pattern, (b) the
+    number of parameters, (c) the names of the parameters. Any unnamed
+    parameters are called '_0', '_1', etc.
+    """
+    # 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.
+    groups = []             # (start, end)
+    quantifiers = []        # start pos
+    ranges = []             # (start, end)
+    eols = []               # pos
+    disjunctions = []       # pos
+    unclosed_groups = []
+    unclosed_ranges = []
+    escaped = False
+    quantify = False
+    in_range = False
+    for pos, c in enumerate(pattern):
+        if in_range and c != ']' or (c == ']' and
+                unclosed_ranges[-1] == pos - 1):
+            continue
+        elif c == '[':
+            unclosed_ranges.append(pos)
+        elif c == ']':
+            ranges.append((unclosed_ranges.pop(), pos + 1))
+            in_range = False
+        elif c == '.':
+            # Treat this as a one-character long range:
+            ranges.append((pos, pos + 1))
+        elif escaped or c == '\\':
+            escaped = not escaped
+        elif c == '(':
+            unclosed_groups.append(pos)
+        elif c == ')':
+            groups.append((unclosed_groups.pop(), pos + 1))
+        elif quantify and c == '?':
+            quantify = False
+        elif c in '?*+{':
+            quantifiers.append(pos)
+            quantify = True
+        elif c == '$':
+            eols.append(pos)
+        elif c == '|':
+            disjunctions.append(pos)
+
+    # Now classify each of the parenthetical groups to work out which ones take
+    # parameters. Only the outer-most of a set of nested capturing groups is
+    # important.
+    groups.sort()
+    params = []
+    comments = []
+    last_end = 0
+    for start, end in groups:
+        if start < last_end:
+            # Skip over inner nested capturing groups.
+            continue
+        m = GROUP_CLASS.match(pattern, start)
+        if m.group('positional'):
+            params.append((start, end, '_%d' % len(params), start + 1))
+        elif m.group('named'):
+            params.append((start, end, m.group('named'), m.start('contents')))
+        elif m.group('repeat'):
+            params.append((start, end, m.group('repeat'), start + 1))
+        elif m.group('illegal'):
+            raise ValueError('The pattern construct %r is not valid here.'
+                    % pattern[start:end])
+        elif m.group('comment'):
+            comments.extend([start, end])
+        else:
+            # This is a non-capturing set, so nesting prohibitions don't apply
+            # to any inner groups.
+            continue
+        last_end = end
+
+    # XXX: Got to here!
+    results = []
+    end = groups[0][0]
+    # The first bit, before the first group starts.
+    if end == 0:
+        # FIXME: don't want to handle this case just yet.
+        raise Exception
+
+    quant_end = bisect(quantifiers, end)
+    range_end = bisect(ranges, end)
+    dis_end = bisect(disjunctions, end)