PostgreSQL Source Code  git master
generate_unaccent_rules.py
Go to the documentation of this file.
1 #!/usr/bin/python
2 # -*- coding: utf-8 -*-
3 #
4 # This script builds unaccent.rules on standard output when given the
5 # contents of UnicodeData.txt [1] and Latin-ASCII.xml [2] given as
6 # arguments. Optionally includes ligature expansion and Unicode CLDR
7 # Latin-ASCII transliterator, enabled by default, this can be disabled
8 # with "--no-ligatures-expansion" command line option.
9 #
10 # The approach is to use the Unicode decomposition data to identify
11 # precomposed codepoints that are equivalent to a ligature of several
12 # letters, or a base letter with any number of diacritical marks.
13 #
14 # This approach handles most letters with diacritical marks and some
15 # ligatures. However, several characters (notably a majority of
16 # ligatures) don't have decomposition. To handle all these cases, one can
17 # use a standard Unicode transliterator available in Common Locale Data
18 # Repository (CLDR): Latin-ASCII. This transliterator associates Unicode
19 # characters to ASCII-range equivalent. Unless "--no-ligatures-expansion"
20 # option is enabled, the XML file of this transliterator [2] -- given as a
21 # command line argument -- will be parsed and used.
22 #
23 # Ideally you should use the latest release for each data set. This
24 # script is compatible with at least CLDR release 29.
25 #
26 # [1] https://www.unicode.org/Public/${UNICODE_VERSION}/ucd/UnicodeData.txt
27 # [2] https://raw.githubusercontent.com/unicode-org/cldr/${TAG}/common/transforms/Latin-ASCII.xml
28 
29 import argparse
30 import codecs
31 import re
32 import sys
33 import xml.etree.ElementTree as ET
34 
35 sys.stdout = codecs.getwriter('utf8')(sys.stdout.buffer)
36 
37 # The ranges of Unicode characters that we consider to be "plain letters".
38 # For now we are being conservative by including only Latin and Greek. This
39 # could be extended in future based on feedback from people with relevant
40 # language knowledge.
41 PLAIN_LETTER_RANGES = ((ord('a'), ord('z')), # Latin lower case
42  (ord('A'), ord('Z')), # Latin upper case
43  (0x03b1, 0x03c9), # GREEK SMALL LETTER ALPHA, GREEK SMALL LETTER OMEGA
44  (0x0391, 0x03a9)) # GREEK CAPITAL LETTER ALPHA, GREEK CAPITAL LETTER OMEGA
45 
46 # Combining marks follow a "base" character, and result in a composite
47 # character. Example: "U&'A\0300'"produces "À".There are three types of
48 # combining marks: enclosing (Me), non-spacing combining (Mn), spacing
49 # combining (Mc). We identify the ranges of marks we feel safe removing.
50 # References:
51 # https://en.wikipedia.org/wiki/Combining_character
52 # https://www.unicode.org/charts/PDF/U0300.pdf
53 # https://www.unicode.org/charts/PDF/U20D0.pdf
54 COMBINING_MARK_RANGES = ((0x0300, 0x0362), # Mn: Accents, IPA
55  (0x20dd, 0x20E0), # Me: Symbols
56  (0x20e2, 0x20e4),) # Me: Screen, keycap, triangle
57 
58 
59 def print_record(codepoint, letter):
60  if letter:
61  output = chr(codepoint) + "\t" + letter
62  else:
63  output = chr(codepoint)
64 
65  print(output)
66 
67 
68 class Codepoint:
69  def __init__(self, id, general_category, combining_ids):
70  self.idid = id
71  self.general_categorygeneral_category = general_category
72  self.combining_idscombining_ids = combining_ids
73 
74 
75 def is_mark_to_remove(codepoint):
76  """Return true if this is a combining mark to remove."""
77  if not is_mark(codepoint):
78  return False
79 
80  for begin, end in COMBINING_MARK_RANGES:
81  if codepoint.id >= begin and codepoint.id <= end:
82  return True
83  return False
84 
85 
86 def is_plain_letter(codepoint):
87  """Return true if codepoint represents a "plain letter"."""
88  for begin, end in PLAIN_LETTER_RANGES:
89  if codepoint.id >= begin and codepoint.id <= end:
90  return True
91  return False
92 
93 
94 def is_mark(codepoint):
95  """Returns true for diacritical marks (combining codepoints)."""
96  return codepoint.general_category in ("Mn", "Me", "Mc")
97 
98 
99 def is_letter_with_marks(codepoint, table):
100  """Returns true for letters combined with one or more marks."""
101  # See https://www.unicode.org/reports/tr44/tr44-14.html#General_Category_Values
102 
103  # Letter may have no combining characters, in which case it has
104  # no marks.
105  if len(codepoint.combining_ids) == 1:
106  return False
107 
108  # A letter without diacritical marks has none of them.
109  if any(is_mark(table[i]) for i in codepoint.combining_ids[1:]) is False:
110  return False
111 
112  # Check if the base letter of this letter has marks.
113  codepoint_base = codepoint.combining_ids[0]
114  if is_plain_letter(table[codepoint_base]) is False and \
115  is_letter_with_marks(table[codepoint_base], table) is False:
116  return False
117 
118  return True
119 
120 
121 def is_letter(codepoint, table):
122  """Return true for letter with or without diacritical marks."""
123  return is_plain_letter(codepoint) or is_letter_with_marks(codepoint, table)
124 
125 
126 def get_plain_letter(codepoint, table):
127  """Return the base codepoint without marks. If this codepoint has more
128  than one combining character, do a recursive lookup on the table to
129  find out its plain base letter."""
130  if is_letter_with_marks(codepoint, table):
131  if len(table[codepoint.combining_ids[0]].combining_ids) > 1:
132  return get_plain_letter(table[codepoint.combining_ids[0]], table)
133  elif is_plain_letter(table[codepoint.combining_ids[0]]):
134  return table[codepoint.combining_ids[0]]
135 
136  # Should not come here
137  assert False, 'Codepoint U+%0.2X' % codepoint.id
138  elif is_plain_letter(codepoint):
139  return codepoint
140 
141  # Should not come here
142  assert False, 'Codepoint U+%0.2X' % codepoint.id
143 
144 
145 def is_ligature(codepoint, table):
146  """Return true for letters combined with letters."""
147  return all(is_letter(table[i], table) for i in codepoint.combining_ids)
148 
149 
150 def get_plain_letters(codepoint, table):
151  """Return a list of plain letters from a ligature."""
152  assert(is_ligature(codepoint, table))
153  return [get_plain_letter(table[id], table) for id in codepoint.combining_ids]
154 
155 
157  """Parse the XML file and return a set of tuples (src, trg), where "src"
158  is the original character and "trg" the substitute."""
159  charactersSet = set()
160 
161  # RegEx to parse rules
162  rulePattern = re.compile(r'^(?:(.)|(\\u[0-9a-fA-F]{4})) \u2192 (?:\'(.+)\'|(.+)) ;')
163 
164  # construct tree from XML
165  transliterationTree = ET.parse(latinAsciiFilePath)
166  transliterationTreeRoot = transliterationTree.getroot()
167 
168  # Fetch all the transliteration rules. Since release 29 of Latin-ASCII.xml
169  # all the transliteration rules are located in a single tRule block with
170  # all rules separated into separate lines.
171  blockRules = transliterationTreeRoot.findall("./transforms/transform/tRule")
172  assert(len(blockRules) == 1)
173 
174  # Split the block of rules into one element per line.
175  rules = blockRules[0].text.splitlines()
176 
177  # And finish the processing of each individual rule.
178  for rule in rules:
179  matches = rulePattern.search(rule)
180 
181  # The regular expression capture four groups corresponding
182  # to the characters.
183  #
184  # Group 1: plain "src" char. Empty if group 2 is not.
185  # Group 2: unicode-escaped "src" char (e.g. "\u0110"). Empty if group 1 is not.
186  #
187  # Group 3: plain "trg" char. Empty if group 4 is not.
188  # Group 4: plain "trg" char between quotes. Empty if group 3 is not.
189  if matches is not None:
190  src = matches.group(1) if matches.group(1) is not None else bytes(matches.group(2), 'UTF-8').decode('unicode-escape')
191  trg = matches.group(3) if matches.group(3) is not None else matches.group(4)
192 
193  # "'" and """ are escaped
194  trg = trg.replace("\\'", "'").replace('\\"', '"')
195 
196  # the parser of unaccent only accepts non-whitespace characters
197  # for "src" and "trg" (see unaccent.c)
198  if not src.isspace() and not trg.isspace():
199  charactersSet.add((ord(src), trg))
200 
201  return charactersSet
202 
203 
205  """Returns the special cases which are not handled by other methods"""
206  charactersSet = set()
207 
208  # Cyrillic
209  charactersSet.add((0x0401, "\u0415")) # CYRILLIC CAPITAL LETTER IO
210  charactersSet.add((0x0451, "\u0435")) # CYRILLIC SMALL LETTER IO
211 
212  # Symbols of "Letterlike Symbols" Unicode Block (U+2100 to U+214F)
213  charactersSet.add((0x2103, "\xb0C")) # DEGREE CELSIUS
214  charactersSet.add((0x2109, "\xb0F")) # DEGREE FAHRENHEIT
215 
216  return charactersSet
217 
218 
219 def main(args):
220  # https://www.unicode.org/reports/tr44/tr44-14.html#Character_Decomposition_Mappings
221  decomposition_type_pattern = re.compile(" *<[^>]*> *")
222 
223  table = {}
224  all = []
225 
226  # unordered set for ensure uniqueness
227  charactersSet = set()
228 
229  # read file UnicodeData.txt
230  with codecs.open(
231  args.unicodeDataFilePath, mode='r', encoding='UTF-8',
232  ) as unicodeDataFile:
233  # read everything we need into memory
234  for line in unicodeDataFile:
235  fields = line.split(";")
236  if len(fields) > 5:
237  # https://www.unicode.org/reports/tr44/tr44-14.html#UnicodeData.txt
238  general_category = fields[2]
239  decomposition = fields[5]
240  decomposition = re.sub(decomposition_type_pattern, ' ', decomposition)
241  id = int(fields[0], 16)
242  combining_ids = [int(s, 16) for s in decomposition.split(" ") if s != ""]
243  codepoint = Codepoint(id, general_category, combining_ids)
244  table[id] = codepoint
245  all.append(codepoint)
246 
247  # walk through all the codepoints looking for interesting mappings
248  for codepoint in all:
249  if codepoint.general_category.startswith('L') and \
250  len(codepoint.combining_ids) > 1:
251  if is_letter_with_marks(codepoint, table):
252  charactersSet.add((codepoint.id,
253  chr(get_plain_letter(codepoint, table).id)))
254  elif args.noLigaturesExpansion is False and is_ligature(codepoint, table):
255  charactersSet.add((codepoint.id,
256  "".join(chr(combining_codepoint.id)
257  for combining_codepoint
258  in get_plain_letters(codepoint, table))))
259  elif is_mark_to_remove(codepoint):
260  charactersSet.add((codepoint.id, None))
261 
262  # add CLDR Latin-ASCII characters
263  if not args.noLigaturesExpansion:
264  charactersSet |= parse_cldr_latin_ascii_transliterator(args.latinAsciiFilePath)
265  charactersSet |= special_cases()
266 
267  # sort for more convenient display
268  charactersList = sorted(charactersSet, key=lambda characterPair: characterPair[0])
269 
270  for characterPair in charactersList:
271  print_record(characterPair[0], characterPair[1])
272 
273 
274 if __name__ == "__main__":
275  parser = argparse.ArgumentParser(description='This script builds unaccent.rules on standard output when given the contents of UnicodeData.txt and Latin-ASCII.xml given as arguments.')
276  parser.add_argument("--unicode-data-file", help="Path to formatted text file corresponding to UnicodeData.txt.", type=str, required=True, dest='unicodeDataFilePath')
277  parser.add_argument("--latin-ascii-file", help="Path to XML file from Unicode Common Locale Data Repository (CLDR) corresponding to Latin-ASCII transliterator (Latin-ASCII.xml).", type=str, dest='latinAsciiFilePath')
278  parser.add_argument("--no-ligatures-expansion", help="Do not expand ligatures and do not use Unicode CLDR Latin-ASCII transliterator. By default, this option is not enabled and \"--latin-ascii-file\" argument is required. If this option is enabled, \"--latin-ascii-file\" argument is optional and ignored.", action="store_true", dest='noLigaturesExpansion')
279  args = parser.parse_args()
280 
281  if args.noLigaturesExpansion is False and args.latinAsciiFilePath is None:
282  sys.stderr.write('You must specify the path to Latin-ASCII transliterator file with \"--latin-ascii-file\" option or use \"--no-ligatures-expansion\" option. Use \"-h\" option for help.')
283  sys.exit(1)
284 
285  main(args)
void print(const void *obj)
Definition: print.c:36
def __init__(self, id, general_category, combining_ids)
def get_plain_letters(codepoint, table)
def print_record(codepoint, letter)
def get_plain_letter(codepoint, table)
def is_letter_with_marks(codepoint, table)
def is_ligature(codepoint, table)
def is_letter(codepoint, table)
def parse_cldr_latin_ascii_transliterator(latinAsciiFilePath)
const void size_t len
#define assert(x)
Definition: regcustom.h:55