1 """Internationalization and localization support. 2 3 This module provides internationalization (I18N) and localization (L10N) 4 support for your Python programs by providing an interface to the GNU gettext 5 message catalog library. 6 7 I18N refers to the operation by which a program is made aware of multiple 8 languages. L10N refers to the adaptation of your program, once 9 internationalized, to the local language and cultural habits. 10 11 """ 12 13 # This module represents the integration of work, contributions, feedback, and 14 # suggestions from the following people: 15 # 16 # Martin von Loewis, who wrote the initial implementation of the underlying 17 # C-based libintlmodule (later renamed _gettext), along with a skeletal 18 # gettext.py implementation. 19 # 20 # Peter Funk, who wrote fintl.py, a fairly complete wrapper around intlmodule, 21 # which also included a pure-Python implementation to read .mo files if 22 # intlmodule wasn't available. 23 # 24 # James Henstridge, who also wrote a gettext.py module, which has some 25 # interesting, but currently unsupported experimental features: the notion of 26 # a Catalog class and instances, and the ability to add to a catalog file via 27 # a Python API. 28 # 29 # Barry Warsaw integrated these modules, wrote the .install() API and code, 30 # and conformed all C and Python code to Python's coding standards. 31 # 32 # Francois Pinard and Marc-Andre Lemburg also contributed valuably to this 33 # module. 34 # 35 # J. David Ibanez implemented plural forms. Bruno Haible fixed some bugs. 36 # 37 # TODO: 38 # - Lazy loading of .mo files. Currently the entire catalog is loaded into 39 # memory, but that's probably bad for large translated programs. Instead, 40 # the lexical sort of original strings in GNU .mo files should be exploited 41 # to do binary searches and lazy initializations. Or you might want to use 42 # the undocumented double-hash algorithm for .mo files with hash tables, but 43 # you'll need to study the GNU gettext code to do this. 44 # 45 # - Support Solaris .mo file formats. Unfortunately, we've been unable to 46 # find this format documented anywhere. 47 48 49 import locale 50 import os 51 import re 52 import sys 53 54 55 __all__ = ['NullTranslations', 'GNUTranslations', 'Catalog', 56 'find', 'translation', 'install', 'textdomain', 'bindtextdomain', 57 'bind_textdomain_codeset', 58 'dgettext', 'dngettext', 'gettext', 'lgettext', 'ldgettext', 59 'ldngettext', 'lngettext', 'ngettext', 60 ] 61 62 _default_localedir = os.path.join(sys.base_prefix, 'share', 'locale') 63 64 # Expression parsing for plural form selection. 65 # 66 # The gettext library supports a small subset of C syntax. The only 67 # incompatible difference is that integer literals starting with zero are 68 # decimal. 69 # 70 # https://www.gnu.org/software/gettext/manual/gettext.html#Plural-forms 71 # http://git.savannah.gnu.org/cgit/gettext.git/tree/gettext-runtime/intl/plural.y 72 73 _token_pattern = re.compile(r""" 74 (?P<WHITESPACES>[ \t]+) | # spaces and horizontal tabs 75 (?P<NUMBER>[0-9]+\b) | # decimal integer 76 (?P<NAME>n\b) | # only n is allowed 77 (?P<PARENTHESIS>[()]) | 78 (?P<OPERATOR>[-*/%+?:]|[><!]=?|==|&&|\|\|) | # !, *, /, %, +, -, <, >, 79 # <=, >=, ==, !=, &&, ||, 80 # ? : 81 # unary and bitwise ops 82 # not allowed 83 (?P<INVALID>\w+|.) # invalid token 84 """, re.VERBOSE|re.DOTALL) 85 86 def _tokenize(plural): 87 for mo in re.finditer(_token_pattern, plural): 88 kind = mo.lastgroup 89 if kind == 'WHITESPACES': 90 continue 91 value = mo.group(kind) 92 if kind == 'INVALID': 93 raise ValueError('invalid token in plural form: %s' % value) 94 yield value 95 yield '' 96 97 def _error(value): 98 if value: 99 return ValueError('unexpected token in plural form: %s' % value) 100 else: 101 return ValueError('unexpected end of plural form') 102 103 _binary_ops = ( 104 ('||',), 105 ('&&',), 106 ('==', '!='), 107 ('<', '>', '<=', '>='), 108 ('+', '-'), 109 ('*', '/', '%'), 110 ) 111 _binary_ops = {op: i for i, ops in enumerate(_binary_ops, 1) for op in ops} 112 _c2py_ops = {'||': 'or', '&&': 'and', '/': '//'} 113 114 def _parse(tokens, priority=-1): 115 result = '' 116 nexttok = next(tokens) 117 while nexttok == '!': 118 result += 'not ' 119 nexttok = next(tokens) 120 121 if nexttok == '(': 122 sub, nexttok = _parse(tokens) 123 result = '%s(%s)' % (result, sub) 124 if nexttok != ')': 125 raise ValueError('unbalanced parenthesis in plural form') 126 elif nexttok == 'n': 127 result = '%s%s' % (result, nexttok) 128 else: 129 try: 130 value = int(nexttok, 10) 131 except ValueError: 132 raise _error(nexttok) from None 133 result = '%s%d' % (result, value) 134 nexttok = next(tokens) 135 136 j = 100 137 while nexttok in _binary_ops: 138 i = _binary_ops[nexttok] 139 if i < priority: 140 break 141 # Break chained comparisons 142 if i in (3, 4) and j in (3, 4): # '==', '!=', '<', '>', '<=', '>=' 143 result = '(%s)' % result 144 # Replace some C operators by their Python equivalents 145 op = _c2py_ops.get(nexttok, nexttok) 146 right, nexttok = _parse(tokens, i + 1) 147 result = '%s %s %s' % (result, op, right) 148 j = i 149 if j == priority == 4: # '<', '>', '<=', '>=' 150 result = '(%s)' % result 151 152 if nexttok == '?' and priority <= 0: 153 if_true, nexttok = _parse(tokens, 0) 154 if nexttok != ':': 155 raise _error(nexttok) 156 if_false, nexttok = _parse(tokens) 157 result = '%s if %s else %s' % (if_true, result, if_false) 158 if priority == 0: 159 result = '(%s)' % result 160 161 return result, nexttok 162 163 def _as_int(n): 164 try: 165 i = round(n) 166 except TypeError: 167 raise TypeError('Plural value must be an integer, got %s' % 168 (n.__class__.__name__,)) from None 169 import warnings 170 warnings.warn('Plural value must be an integer, got %s' % 171 (n.__class__.__name__,), 172 DeprecationWarning, 4) 173 return n 174 175 def c2py(plural): 176 """Gets a C expression as used in PO files for plural forms and returns a 177 Python function that implements an equivalent expression. 178 """ 179 180 if len(plural) > 1000: 181 raise ValueError('plural form expression is too long') 182 try: 183 result, nexttok = _parse(_tokenize(plural)) 184 if nexttok: 185 raise _error(nexttok) 186 187 depth = 0 188 for c in result: 189 if c == '(': 190 depth += 1 191 if depth > 20: 192 # Python compiler limit is about 90. 193 # The most complex example has 2. 194 raise ValueError('plural form expression is too complex') 195 elif c == ')': 196 depth -= 1 197 198 ns = {'_as_int': _as_int} 199 exec('''if True: 200 def func(n): 201 if not isinstance(n, int): 202 n = _as_int(n) 203 return int(%s) 204 ''' % result, ns) 205 return ns['func'] 206 except RecursionError: 207 # Recursion error can be raised in _parse() or exec(). 208 raise ValueError('plural form expression is too complex') 209 210 211 def _expand_lang(loc): 212 loc = locale.normalize(loc) 213 COMPONENT_CODESET = 1 << 0 214 COMPONENT_TERRITORY = 1 << 1 215 COMPONENT_MODIFIER = 1 << 2 216 # split up the locale into its base components 217 mask = 0 218 pos = loc.find('@') 219 if pos >= 0: 220 modifier = loc[pos:] 221 loc = loc[:pos] 222 mask |= COMPONENT_MODIFIER 223 else: 224 modifier = '' 225 pos = loc.find('.') 226 if pos >= 0: 227 codeset = loc[pos:] 228 loc = loc[:pos] 229 mask |= COMPONENT_CODESET 230 else: 231 codeset = '' 232 pos = loc.find('_') 233 if pos >= 0: 234 territory = loc[pos:] 235 loc = loc[:pos] 236 mask |= COMPONENT_TERRITORY 237 else: 238 territory = '' 239 language = loc 240 ret = [] 241 for i in range(mask+1): 242 if not (i & ~mask): # if all components for this combo exist ... 243 val = language 244 if i & COMPONENT_TERRITORY: val += territory 245 if i & COMPONENT_CODESET: val += codeset 246 if i & COMPONENT_MODIFIER: val += modifier 247 ret.append(val) 248 ret.reverse() 249 return ret 250 251 252 253 class NullTranslations: 254 def __init__(self, fp=None): 255 self._info = {} 256 self._charset = None 257 self._output_charset = None 258 self._fallback = None 259 if fp is not None: 260 self._parse(fp) 261 262 def _parse(self, fp): 263 pass 264 265 def add_fallback(self, fallback): 266 if self._fallback: 267 self._fallback.add_fallback(fallback) 268 else: 269 self._fallback = fallback 270 271 def gettext(self, message): 272 if self._fallback: 273 return self._fallback.gettext(message) 274 return message 275 276 def lgettext(self, message): 277 if self._fallback: 278 return self._fallback.lgettext(message) 279 if self._output_charset: 280 return message.encode(self._output_charset) 281 return message.encode(locale.getpreferredencoding()) 282 283 def ngettext(self, msgid1, msgid2, n): 284 if self._fallback: 285 return self._fallback.ngettext(msgid1, msgid2, n) 286 if n == 1: 287 return msgid1 288 else: 289 return msgid2 290 291 def lngettext(self, msgid1, msgid2, n): 292 if self._fallback: 293 return self._fallback.lngettext(msgid1, msgid2, n) 294 if n == 1: 295 tmsg = msgid1 296 else: 297 tmsg = msgid2 298 if self._output_charset: 299 return tmsg.encode(self._output_charset) 300 return tmsg.encode(locale.getpreferredencoding()) 301 302 def info(self): 303 return self._info 304 305 def charset(self): 306 return self._charset 307 308 def output_charset(self): 309 return self._output_charset 310 311 def set_output_charset(self, charset): 312 self._output_charset = charset 313 314 def install(self, names=None): 315 import builtins 316 builtins.__dict__['_'] = self.gettext 317 if hasattr(names, "__contains__"): 318 if "gettext" in names: 319 builtins.__dict__['gettext'] = builtins.__dict__['_'] 320 if "ngettext" in names: 321 builtins.__dict__['ngettext'] = self.ngettext 322 if "lgettext" in names: 323 builtins.__dict__['lgettext'] = self.lgettext 324 if "lngettext" in names: 325 builtins.__dict__['lngettext'] = self.lngettext 326 327 328 class GNUTranslations(NullTranslations): 329 # Magic number of .mo files 330 LE_MAGIC = 0x950412de 331 BE_MAGIC = 0xde120495 332 333 # Acceptable .mo versions 334 VERSIONS = (0, 1) 335 336 def _get_versions(self, version): 337 """Returns a tuple of major version, minor version""" 338 return (version >> 16, version & 0xffff) 339 340 def _parse(self, fp): 341 """Override this method to support alternative .mo formats.""" 342 # Delay struct import for speeding up gettext import when .mo files 343 # are not used. 344 from struct import unpack 345 filename = getattr(fp, 'name', '') 346 # Parse the .mo file header, which consists of 5 little endian 32 347 # bit words. 348 self._catalog = catalog = {} 349 self.plural = lambda n: int(n != 1) # germanic plural by default 350 buf = fp.read() 351 buflen = len(buf) 352 # Are we big endian or little endian? 353 magic = unpack('<I', buf[:4])[0] 354 if magic == self.LE_MAGIC: 355 version, msgcount, masteridx, transidx = unpack('<4I', buf[4:20]) 356 ii = '<II' 357 elif magic == self.BE_MAGIC: 358 version, msgcount, masteridx, transidx = unpack('>4I', buf[4:20]) 359 ii = '>II' 360 else: 361 raise OSError(0, 'Bad magic number', filename) 362 363 major_version, minor_version = self._get_versions(version) 364 365 if major_version not in self.VERSIONS: 366 raise OSError(0, 'Bad version number ' + str(major_version), filename) 367 368 # Now put all messages from the .mo file buffer into the catalog 369 # dictionary. 370 for i in range(0, msgcount): 371 mlen, moff = unpack(ii, buf[masteridx:masteridx+8]) 372 mend = moff + mlen 373 tlen, toff = unpack(ii, buf[transidx:transidx+8]) 374 tend = toff + tlen 375 if mend < buflen and tend < buflen: 376 msg = buf[moff:mend] 377 tmsg = buf[toff:tend] 378 else: 379 raise OSError(0, 'File is corrupt', filename) 380 # See if we're looking at GNU .mo conventions for metadata 381 if mlen == 0: 382 # Catalog description 383 lastk = None 384 for b_item in tmsg.split(b'\n'): 385 item = b_item.decode().strip() 386 if not item: 387 continue 388 k = v = None 389 if ':' in item: 390 k, v = item.split(':', 1) 391 k = k.strip().lower() 392 v = v.strip() 393 self._info[k] = v 394 lastk = k 395 elif lastk: 396 self._info[lastk] += '\n' + item 397 if k == 'content-type': 398 self._charset = v.split('charset=')[1] 399 elif k == 'plural-forms': 400 v = v.split(';') 401 plural = v[1].split('plural=')[1] 402 self.plural = c2py(plural) 403 # Note: we unconditionally convert both msgids and msgstrs to 404 # Unicode using the character encoding specified in the charset 405 # parameter of the Content-Type header. The gettext documentation 406 # strongly encourages msgids to be us-ascii, but some applications 407 # require alternative encodings (e.g. Zope's ZCML and ZPT). For 408 # traditional gettext applications, the msgid conversion will 409 # cause no problems since us-ascii should always be a subset of 410 # the charset encoding. We may want to fall back to 8-bit msgids 411 # if the Unicode conversion fails. 412 charset = self._charset or 'ascii' 413 if b'\x00' in msg: 414 # Plural forms 415 msgid1, msgid2 = msg.split(b'\x00') 416 tmsg = tmsg.split(b'\x00') 417 msgid1 = str(msgid1, charset) 418 for i, x in enumerate(tmsg): 419 catalog[(msgid1, i)] = str(x, charset) 420 else: 421 catalog[str(msg, charset)] = str(tmsg, charset) 422 # advance to next entry in the seek tables 423 masteridx += 8 424 transidx += 8 425 426 def lgettext(self, message): 427 missing = object() 428 tmsg = self._catalog.get(message, missing) 429 if tmsg is missing: 430 if self._fallback: 431 return self._fallback.lgettext(message) 432 tmsg = message 433 if self._output_charset: 434 return tmsg.encode(self._output_charset) 435 return tmsg.encode(locale.getpreferredencoding()) 436 437 def lngettext(self, msgid1, msgid2, n): 438 try: 439 tmsg = self._catalog[(msgid1, self.plural(n))] 440 except KeyError: 441 if self._fallback: 442 return self._fallback.lngettext(msgid1, msgid2, n) 443 if n == 1: 444 tmsg = msgid1 445 else: 446 tmsg = msgid2 447 if self._output_charset: 448 return tmsg.encode(self._output_charset) 449 return tmsg.encode(locale.getpreferredencoding()) 450 451 def gettext(self, message): 452 missing = object() 453 tmsg = self._catalog.get(message, missing) 454 if tmsg is missing: 455 if self._fallback: 456 return self._fallback.gettext(message) 457 return message 458 return tmsg 459 460 def ngettext(self, msgid1, msgid2, n): 461 try: 462 tmsg = self._catalog[(msgid1, self.plural(n))] 463 except KeyError: 464 if self._fallback: 465 return self._fallback.ngettext(msgid1, msgid2, n) 466 if n == 1: 467 tmsg = msgid1 468 else: 469 tmsg = msgid2 470 return tmsg 471 472 473 # Locate a .mo file using the gettext strategy 474 def find(domain, localedir=None, languages=None, all=False): 475 # Get some reasonable defaults for arguments that were not supplied 476 if localedir is None: 477 localedir = _default_localedir 478 if languages is None: 479 languages = [] 480 for envar in ('LANGUAGE', 'LC_ALL', 'LC_MESSAGES', 'LANG'): 481 val = os.environ.get(envar) 482 if val: 483 languages = val.split(':') 484 break 485 if 'C' not in languages: 486 languages.append('C') 487 # now normalize and expand the languages 488 nelangs = [] 489 for lang in languages: 490 for nelang in _expand_lang(lang): 491 if nelang not in nelangs: 492 nelangs.append(nelang) 493 # select a language 494 if all: 495 result = [] 496 else: 497 result = None 498 for lang in nelangs: 499 if lang == 'C': 500 break 501 mofile = os.path.join(localedir, lang, 'LC_MESSAGES', '%s.mo' % domain) 502 if os.path.exists(mofile): 503 if all: 504 result.append(mofile) 505 else: 506 return mofile 507 return result 508 509 510 511 # a mapping between absolute .mo file path and Translation object 512 _translations = {} 513 514 def translation(domain, localedir=None, languages=None, 515 class_=None, fallback=False, codeset=None): 516 if class_ is None: 517 class_ = GNUTranslations 518 mofiles = find(domain, localedir, languages, all=True) 519 if not mofiles: 520 if fallback: 521 return NullTranslations() 522 from errno import ENOENT 523 raise FileNotFoundError(ENOENT, 524 'No translation file found for domain', domain) 525 # Avoid opening, reading, and parsing the .mo file after it's been done 526 # once. 527 result = None 528 for mofile in mofiles: 529 key = (class_, os.path.abspath(mofile)) 530 t = _translations.get(key) 531 if t is None: 532 with open(mofile, 'rb') as fp: 533 t = _translations.setdefault(key, class_(fp)) 534 # Copy the translation object to allow setting fallbacks and 535 # output charset. All other instance data is shared with the 536 # cached object. 537 # Delay copy import for speeding up gettext import when .mo files 538 # are not used. 539 import copy 540 t = copy.copy(t) 541 if codeset: 542 t.set_output_charset(codeset) 543 if result is None: 544 result = t 545 else: 546 result.add_fallback(t) 547 return result 548 549 550 def install(domain, localedir=None, codeset=None, names=None): 551 t = translation(domain, localedir, fallback=True, codeset=codeset) 552 t.install(names) 553 554 555 556 # a mapping b/w domains and locale directories 557 _localedirs = {} 558 # a mapping b/w domains and codesets 559 _localecodesets = {} 560 # current global domain, `messages' used for compatibility w/ GNU gettext 561 _current_domain = 'messages' 562 563 564 def textdomain(domain=None): 565 global _current_domain 566 if domain is not None: 567 _current_domain = domain 568 return _current_domain 569 570 571 def bindtextdomain(domain, localedir=None): 572 global _localedirs 573 if localedir is not None: 574 _localedirs[domain] = localedir 575 return _localedirs.get(domain, _default_localedir) 576 577 578 def bind_textdomain_codeset(domain, codeset=None): 579 global _localecodesets 580 if codeset is not None: 581 _localecodesets[domain] = codeset 582 return _localecodesets.get(domain) 583 584 585 def dgettext(domain, message): 586 try: 587 t = translation(domain, _localedirs.get(domain, None), 588 codeset=_localecodesets.get(domain)) 589 except OSError: 590 return message 591 return t.gettext(message) 592 593 def ldgettext(domain, message): 594 codeset = _localecodesets.get(domain) 595 try: 596 t = translation(domain, _localedirs.get(domain, None), codeset=codeset) 597 except OSError: 598 return message.encode(codeset or locale.getpreferredencoding()) 599 return t.lgettext(message) 600 601 def dngettext(domain, msgid1, msgid2, n): 602 try: 603 t = translation(domain, _localedirs.get(domain, None), 604 codeset=_localecodesets.get(domain)) 605 except OSError: 606 if n == 1: 607 return msgid1 608 else: 609 return msgid2 610 return t.ngettext(msgid1, msgid2, n) 611 612 def ldngettext(domain, msgid1, msgid2, n): 613 codeset = _localecodesets.get(domain) 614 try: 615 t = translation(domain, _localedirs.get(domain, None), codeset=codeset) 616 except OSError: 617 if n == 1: 618 tmsg = msgid1 619 else: 620 tmsg = msgid2 621 return tmsg.encode(codeset or locale.getpreferredencoding()) 622 return t.lngettext(msgid1, msgid2, n) 623 624 def gettext(message): 625 return dgettext(_current_domain, message) 626 627 def lgettext(message): 628 return ldgettext(_current_domain, message) 629 630 def ngettext(msgid1, msgid2, n): 631 return dngettext(_current_domain, msgid1, msgid2, n) 632 633 def lngettext(msgid1, msgid2, n): 634 return ldngettext(_current_domain, msgid1, msgid2, n) 635 636 # dcgettext() has been deemed unnecessary and is not implemented. 637 638 # James Henstridge's Catalog constructor from GNOME gettext. Documented usage 639 # was: 640 # 641 # import gettext 642 # cat = gettext.Catalog(PACKAGE, localedir=LOCALEDIR) 643 # _ = cat.gettext 644 # print _('Hello World') 645 646 # The resulting catalog object currently don't support access through a 647 # dictionary API, which was supported (but apparently unused) in GNOME 648 # gettext. 649 650 Catalog = translation 651