Author: runa
Date: 2011-03-17 17:52:17 +0000 (Thu, 17 Mar 2011)
New Revision: 24379
Added:
translation/trunk/tools/polib.py
translation/trunk/tools/validate.py
Log:
a script to validate translation input, thanks chiiph
Added: translation/trunk/tools/polib.py
===================================================================
--- translation/trunk/tools/polib.py (rev 0)
+++ translation/trunk/tools/polib.py 2011-03-17 17:52:17 UTC (rev 24379)
@@ -0,0 +1,1560 @@
+# -*- coding: utf-8 -*-
+#
+# License: MIT (see LICENSE file provided)
+# vim: set expandtab tabstop=4 shiftwidth=4 softtabstop=4:
+
+"""
+**polib** allows you to manipulate, create, modify gettext files (pot, po and
+mo files). You can load existing files, iterate through it's entries, add,
+modify entries, comments or metadata, etc. or create new po files from scratch.
+
+**polib** provides a simple and pythonic API via the :func:`~polib.pofile` and
+:func:`~polib.mofile` convenience functions.
+"""
+
+__author__ = 'David Jean Louis <izimobil(a)gmail.com>'
+__version__ = '0.6.3'
+__all__ = ['pofile', 'POFile', 'POEntry', 'mofile', 'MOFile', 'MOEntry',
+ 'detect_encoding', 'escape', 'unescape', 'detect_encoding',]
+
+import array
+import codecs
+import os
+import re
+import struct
+import sys
+import textwrap
+import types
+
+
+# the default encoding to use when autodetect_encoding is disabled
+default_encoding = 'utf-8'
+
+# _pofile_or_mofile {{{
+
+def _pofile_or_mofile(f, type, **kwargs):
+ """
+ Internal function used by :func:`polib.pofile` and :func:`polib.mofile` to
+ honor the DRY concept.
+ """
+ # get the file encoding
+ if kwargs.get('autodetect_encoding', True):
+ enc = detect_encoding(f, type == 'mofile')
+ else:
+ enc = kwargs.get('encoding', default_encoding)
+
+ # parse the file
+ kls = type == 'pofile' and _POFileParser or _MOFileParser
+ parser = kls(
+ f,
+ encoding=enc,
+ check_for_duplicates=kwargs.get('check_for_duplicates', False)
+ )
+ instance = parser.parse()
+ instance.wrapwidth = kwargs.get('wrapwidth', 78)
+ return instance
+
+# }}}
+# function pofile() {{{
+
+def pofile(pofile, **kwargs):
+ """
+ Convenience function that parses the po or pot file ``pofile`` and returns
+ a :class:`~polib.POFile` instance.
+
+ Arguments:
+
+ ``pofile``
+ string, full or relative path to the po/pot file or its content (data).
+
+ ``wrapwidth``
+ integer, the wrap width, only useful when the ``-w`` option was passed
+ to xgettext (optional, default: ``78``).
+
+ ``autodetect_encoding``
+ boolean, if set to ``False`` the function will not try to detect the
+ po file encoding and will use either the value of the ``encoding``
+ argument or the ``default_encoding`` (optional, default: ``True``).
+
+ ``encoding``
+ string, the encoding to use (e.g. "utf-8"), only relevant if
+ ``autodetect_encoding`` is set to ``False``.
+
+ ``check_for_duplicates``
+ whether to check for duplicate entries when adding entries to the
+ file (optional, default: ``False``).
+ """
+ return _pofile_or_mofile(pofile, 'pofile', **kwargs)
+
+# }}}
+# function mofile() {{{
+
+def mofile(mofile, **kwargs):
+ """
+ Convenience function that parses the mo file ``mofile`` and returns a
+ :class:`~polib.MOFile` instance.
+
+ Arguments:
+
+ ``mofile``
+ string, full or relative path to the mo file or its content (data).
+
+ ``wrapwidth``
+ integer, the wrap width, only useful when the ``-w`` option was passed
+ to xgettext to generate the po file that was used to format the mo file
+ (optional, default: ``78``).
+
+ ``autodetect_encoding``
+ boolean, if set to ``False`` the function will not try to detect the
+ mo file encoding (optional, default: ``True``).
+
+ ``encoding``
+ string, the encoding to use, only relevant if ``autodetect_encoding``
+ is set to ``False``.
+
+ ``check_for_duplicates``
+ whether to check for duplicate entries when adding entries to the
+ file (optional, default: ``False``).
+ """
+ return _pofile_or_mofile(mofile, 'mofile', **kwargs)
+
+# }}}
+# function detect_encoding() {{{
+
+def detect_encoding(file, binary_mode=False):
+ """
+ Try to detect the encoding used by the ``file``. The ``file`` argument can
+ be a PO or MO file path or a string containing the contents of the file.
+ If the encoding cannot be detected, the function will return the value of
+ ``default_encoding``.
+
+ Arguments:
+
+ ``file``
+ string, full or relative path to the po/mo file or its content.
+
+ ``binary_mode``
+ boolean, set this to True if ``file`` is a mo file.
+ """
+ rx = re.compile(r'"?Content-Type:.+? charset=([\w_\-:\.]+)')
+
+ def charset_exists(charset):
+ """Check whether ``charset`` is valid or not."""
+ try:
+ codecs.lookup(charset)
+ except LookupError:
+ return False
+ return True
+
+ if not os.path.exists(file):
+ match = rx.search(file)
+ if match:
+ enc = match.group(1).strip()
+ if charset_exists(enc):
+ return enc
+ else:
+ if binary_mode:
+ mode = 'rb'
+ else:
+ mode = 'r'
+ f = open(file, mode)
+ for l in f.readlines():
+ match = rx.search(l)
+ if match:
+ f.close()
+ enc = match.group(1).strip()
+ if charset_exists(enc):
+ return enc
+ f.close()
+ return default_encoding
+
+# }}}
+# function escape() {{{
+
+def escape(st):
+ """
+ Escapes the characters ``\\\\``, ``\\t``, ``\\n``, ``\\r`` and ``"`` in
+ the given string ``st`` and returns it.
+ """
+ return st.replace('\\', r'\\')\
+ .replace('\t', r'\t')\
+ .replace('\r', r'\r')\
+ .replace('\n', r'\n')\
+ .replace('\"', r'\"')
+
+# }}}
+# function unescape() {{{
+
+def unescape(st):
+ """
+ Unescapes the characters ``\\\\``, ``\\t``, ``\\n``, ``\\r`` and ``"`` in
+ the given string ``st`` and returns it.
+ """
+ def unescape_repl(m):
+ m = m.group(1)
+ if m == 'n':
+ return '\n'
+ if m == 't':
+ return '\t'
+ if m == 'r':
+ return '\r'
+ if m == '\\':
+ return '\\'
+ return m # handles escaped double quote
+ return re.sub(r'\\(\\|n|t|r|")', unescape_repl, st)
+
+# }}}
+# class _BaseFile {{{
+
+class _BaseFile(list):
+ """
+ Common base class for the :class:`~polib.POFile` and :class:`~polib.MOFile`
+ classes. This class should **not** be instanciated directly.
+ """
+
+ def __init__(self, *args, **kwargs):
+ """
+ Constructor, accepts the following keyword arguments:
+
+ ``pofile``
+ string, the path to the po or mo file, or its content as a string.
+
+ ``wrapwidth``
+ integer, the wrap width, only useful when the ``-w`` option was
+ passed to xgettext (optional, default: ``78``).
+
+ ``encoding``
+ string, the encoding to use, defaults to ``default_encoding``
+ global variable (optional).
+
+ ``check_for_duplicates``
+ whether to check for duplicate entries when adding entries to the
+ file, (optional, default: ``False``).
+ """
+ list.__init__(self)
+ # the opened file handle
+ pofile = kwargs.get('pofile', None)
+ if pofile and os.path.exists(pofile):
+ self.fpath = pofile
+ else:
+ self.fpath = kwargs.get('fpath')
+ # the width at which lines should be wrapped
+ self.wrapwidth = kwargs.get('wrapwidth', 78)
+ # the file encoding
+ self.encoding = kwargs.get('encoding', default_encoding)
+ # whether to check for duplicate entries or not
+ self.check_for_duplicates = kwargs.get('check_for_duplicates', False)
+ # header
+ self.header = ''
+ # both po and mo files have metadata
+ self.metadata = {}
+ self.metadata_is_fuzzy = 0
+
+ def __unicode__(self):
+ """
+ Returns the unicode representation of the file.
+ """
+ ret = []
+ entries = [self.metadata_as_entry()] + \
+ [e for e in self if not e.obsolete]
+ for entry in entries:
+ ret.append(entry.__unicode__(self.wrapwidth))
+ for entry in self.obsolete_entries():
+ ret.append(entry.__unicode__(self.wrapwidth))
+ ret = '\n'.join(ret)
+
+ if type(ret) != types.UnicodeType:
+ return unicode(ret, self.encoding)
+ return ret
+
+ def __str__(self):
+ """
+ Returns the string representation of the file.
+ """
+ return unicode(self).encode(self.encoding)
+
+ def __contains__(self, entry):
+ """
+ Overriden ``list`` method to implement the membership test (in and
+ not in).
+ The method considers that an entry is in the file if it finds an entry
+ that has the same msgid (the test is **case sensitive**).
+
+ Argument:
+
+ ``entry``
+ an instance of :class:`~polib._BaseEntry`.
+ """
+ return self.find(entry.msgid, by='msgid') is not None
+
+ def __eq__(self, other):
+ return unicode(self) == unicode(other)
+
+ def append(self, entry):
+ """
+ Overriden method to check for duplicates entries, if a user tries to
+ add an entry that is already in the file, the method will raise a
+ ``ValueError`` exception.
+
+ Argument:
+
+ ``entry``
+ an instance of :class:`~polib._BaseEntry`.
+ """
+ if self.check_for_duplicates and entry in self:
+ raise ValueError('Entry "%s" already exists' % entry.msgid)
+ super(_BaseFile, self).append(entry)
+
+ def insert(self, index, entry):
+ """
+ Overriden method to check for duplicates entries, if a user tries to
+ add an entry that is already in the file, the method will raise a
+ ``ValueError`` exception.
+
+ Arguments:
+
+ ``index``
+ index at which the entry should be inserted.
+
+ ``entry``
+ an instance of :class:`~polib._BaseEntry`.
+ """
+ if self.check_for_duplicates and entry in self:
+ raise ValueError('Entry "%s" already exists' % entry.msgid)
+ super(_BaseFile, self).insert(index, entry)
+
+ def metadata_as_entry(self):
+ """
+ Returns the file metadata as a :class:`~polib.POFile` instance.
+ """
+ e = POEntry(msgid='')
+ mdata = self.ordered_metadata()
+ if mdata:
+ strs = []
+ for name, value in mdata:
+ # Strip whitespace off each line in a multi-line entry
+ strs.append('%s: %s' % (name, value))
+ e.msgstr = '\n'.join(strs) + '\n'
+ if self.metadata_is_fuzzy:
+ e.flags.append('fuzzy')
+ return e
+
+ def save(self, fpath=None, repr_method='__str__'):
+ """
+ Saves the po file to ``fpath``.
+ If it is an existing file and no ``fpath`` is provided, then the
+ existing file is rewritten with the modified data.
+
+ Keyword arguments:
+
+ ``fpath``
+ string, full or relative path to the file.
+
+ ``repr_method``
+ string, the method to use for output.
+ """
+ if self.fpath is None and fpath is None:
+ raise IOError('You must provide a file path to save() method')
+ contents = getattr(self, repr_method)()
+ if fpath is None:
+ fpath = self.fpath
+ if repr_method == 'to_binary':
+ fhandle = open(fpath, 'wb')
+ else:
+ fhandle = codecs.open(fpath, 'w', self.encoding)
+ if type(contents) != types.UnicodeType:
+ contents = contents.decode(self.encoding)
+ fhandle.write(contents)
+ fhandle.close()
+ # set the file path if not set
+ if self.fpath is None and fpath:
+ self.fpath = fpath
+
+ def find(self, st, by='msgid', include_obsolete_entries=False,
+ msgctxt=False):
+ """
+ Find the entry which msgid (or property identified by the ``by``
+ argument) matches the string ``st``.
+
+ Keyword arguments:
+
+ ``st``
+ string, the string to search for.
+
+ ``by``
+ string, the property to use for comparison (default: ``msgid``).
+
+ ``include_obsolete_entries``
+ boolean, whether to also search in entries that are obsolete.
+
+ ``msgctxt``
+ string, allows to specify a specific message context for the
+ search.
+ """
+ if include_obsolete_entries:
+ entries = self[:]
+ else:
+ entries = [e for e in self if not e.obsolete]
+ for e in entries:
+ if getattr(e, by) == st:
+ if msgctxt and e.msgctxt != msgctxt:
+ continue
+ return e
+ return None
+
+ def ordered_metadata(self):
+ """
+ Convenience method that returns an ordered version of the metadata
+ dictionnary. The return value is list of tuples (metadata name,
+ metadata_value).
+ """
+ # copy the dict first
+ metadata = self.metadata.copy()
+ data_order = [
+ 'Project-Id-Version',
+ 'Report-Msgid-Bugs-To',
+ 'POT-Creation-Date',
+ 'PO-Revision-Date',
+ 'Last-Translator',
+ 'Language-Team',
+ 'MIME-Version',
+ 'Content-Type',
+ 'Content-Transfer-Encoding'
+ ]
+ ordered_data = []
+ for data in data_order:
+ try:
+ value = metadata.pop(data)
+ ordered_data.append((data, value))
+ except KeyError:
+ pass
+ # the rest of the metadata will be alphabetically ordered since there
+ # are no specs for this AFAIK
+ keys = metadata.keys()
+ keys.sort()
+ for data in keys:
+ value = metadata[data]
+ ordered_data.append((data, value))
+ return ordered_data
+
+ def to_binary(self):
+ """
+ Return the binary representation of the file.
+ """
+ offsets = []
+ entries = self.translated_entries()
+ # the keys are sorted in the .mo file
+ def cmp(_self, other):
+ # msgfmt compares entries with msgctxt if it exists
+ self_msgid = _self.msgctxt and _self.msgctxt or _self.msgid
+ other_msgid = other.msgctxt and other.msgctxt or other.msgid
+ if self_msgid > other_msgid:
+ return 1
+ elif self_msgid < other_msgid:
+ return -1
+ else:
+ return 0
+ # add metadata entry
+ entries.sort(cmp)
+ mentry = self.metadata_as_entry()
+ #mentry.msgstr = mentry.msgstr.replace('\\n', '').lstrip()
+ entries = [mentry] + entries
+ entries_len = len(entries)
+ ids, strs = '', ''
+ for e in entries:
+ # For each string, we need size and file offset. Each string is
+ # NUL terminated; the NUL does not count into the size.
+ msgid = ''
+ if e.msgctxt:
+ # Contexts are stored by storing the concatenation of the
+ # context, a <EOT> byte, and the original string
+ msgid = self._encode(e.msgctxt + '\4')
+ if e.msgid_plural:
+ indexes = e.msgstr_plural.keys()
+ indexes.sort()
+ msgstr = []
+ for index in indexes:
+ msgstr.append(e.msgstr_plural[index])
+ msgid += self._encode(e.msgid + '\0' + e.msgid_plural)
+ msgstr = self._encode('\0'.join(msgstr))
+ else:
+ msgid += self._encode(e.msgid)
+ msgstr = self._encode(e.msgstr)
+ offsets.append((len(ids), len(msgid), len(strs), len(msgstr)))
+ ids += msgid + '\0'
+ strs += msgstr + '\0'
+
+ # The header is 7 32-bit unsigned integers.
+ keystart = 7*4+16*entries_len
+ # and the values start after the keys
+ valuestart = keystart + len(ids)
+ koffsets = []
+ voffsets = []
+ # The string table first has the list of keys, then the list of values.
+ # Each entry has first the size of the string, then the file offset.
+ for o1, l1, o2, l2 in offsets:
+ koffsets += [l1, o1+keystart]
+ voffsets += [l2, o2+valuestart]
+ offsets = koffsets + voffsets
+ # check endianness for magic number
+ if struct.pack('@h', 1) == struct.pack('<h', 1):
+ magic_number = MOFile.LITTLE_ENDIAN
+ else:
+ magic_number = MOFile.BIG_ENDIAN
+
+ output = struct.pack(
+ "Iiiiiii",
+ magic_number, # Magic number
+ 0, # Version
+ entries_len, # # of entries
+ 7*4, # start of key index
+ 7*4+entries_len*8, # start of value index
+ 0, keystart # size and offset of hash table
+ # Important: we don't use hash tables
+ )
+ output += array.array("i", offsets).tostring()
+ output += ids
+ output += strs
+ return output
+
+ def _encode(self, mixed):
+ """
+ Encodes the given ``mixed`` argument with the file encoding if and
+ only if it's an unicode string and returns the encoded string.
+ """
+ if type(mixed) == types.UnicodeType:
+ return mixed.encode(self.encoding)
+ return mixed
+
+# }}}
+# class POFile {{{
+
+class POFile(_BaseFile):
+ """
+ Po (or Pot) file reader/writer.
+ This class inherits the :class:`~polib._BaseFile` class and, by extension,
+ the python ``list`` type.
+ """
+
+ def __unicode__(self):
+ """
+ Returns the unicode representation of the po file.
+ """
+ ret, headers = '', self.header.split('\n')
+ for header in headers:
+ if header[:1] in [',', ':']:
+ ret += '#%s\n' % header
+ else:
+ ret += '# %s\n' % header
+
+ if type(ret) != types.UnicodeType:
+ ret = unicode(ret, self.encoding)
+
+ return ret + _BaseFile.__unicode__(self)
+
+ def save_as_mofile(self, fpath):
+ """
+ Saves the binary representation of the file to given ``fpath``.
+
+ Keyword argument:
+
+ ``fpath``
+ string, full or relative path to the mo file.
+ """
+ _BaseFile.save(self, fpath, 'to_binary')
+
+ def percent_translated(self):
+ """
+ Convenience method that returns the percentage of translated
+ messages.
+ """
+ total = len([e for e in self if not e.obsolete])
+ if total == 0:
+ return 100
+ translated = len(self.translated_entries())
+ return int((100.00 / float(total)) * translated)
+
+ def translated_entries(self):
+ """
+ Convenience method that returns the list of translated entries.
+ """
+ return [e for e in self if e.translated()]
+
+ def untranslated_entries(self):
+ """
+ Convenience method that returns the list of untranslated entries.
+ """
+ return [e for e in self if not e.translated() and not e.obsolete \
+ and not 'fuzzy' in e.flags]
+
+ def fuzzy_entries(self):
+ """
+ Convenience method that returns the list of fuzzy entries.
+ """
+ return [e for e in self if 'fuzzy' in e.flags]
+
+ def obsolete_entries(self):
+ """
+ Convenience method that returns the list of obsolete entries.
+ """
+ return [e for e in self if e.obsolete]
+
+ def merge(self, refpot):
+ """
+ Convenience method that merges the current pofile with the pot file
+ provided. It behaves exactly as the gettext msgmerge utility:
+
+ * comments of this file will be preserved, but extracted comments and
+ occurrences will be discarded;
+ * any translations or comments in the file will be discarded, however,
+ dot comments and file positions will be preserved;
+ * the fuzzy flags are preserved.
+
+ Keyword argument:
+
+ ``refpot``
+ object POFile, the reference catalog.
+ """
+ for entry in refpot:
+ e = self.find(entry.msgid, include_obsolete_entries=True)
+ if e is None:
+ e = POEntry()
+ self.append(e)
+ e.merge(entry)
+ # ok, now we must "obsolete" entries that are not in the refpot anymore
+ for entry in self:
+ if refpot.find(entry.msgid) is None:
+ entry.obsolete = True
+
+# }}}
+# class MOFile {{{
+
+class MOFile(_BaseFile):
+ """
+ Mo file reader/writer.
+ This class inherits the :class:`~polib._BaseFile` class and, by
+ extension, the python ``list`` type.
+ """
+ BIG_ENDIAN = 0xde120495
+ LITTLE_ENDIAN = 0x950412de
+
+ def __init__(self, *args, **kwargs):
+ """
+ Constructor, accepts all keywords arguments accepted by
+ :class:`~polib._BaseFile` class.
+ """
+ _BaseFile.__init__(self, *args, **kwargs)
+ self.magic_number = None
+ self.version = 0
+
+ def save_as_pofile(self, fpath):
+ """
+ Saves the mofile as a pofile to ``fpath``.
+
+ Keyword argument:
+
+ ``fpath``
+ string, full or relative path to the file.
+ """
+ _BaseFile.save(self, fpath)
+
+ def save(self, fpath=None):
+ """
+ Saves the mofile to ``fpath``.
+
+ Keyword argument:
+
+ ``fpath``
+ string, full or relative path to the file.
+ """
+ _BaseFile.save(self, fpath, 'to_binary')
+
+ def percent_translated(self):
+ """
+ Convenience method to keep the same interface with POFile instances.
+ """
+ return 100
+
+ def translated_entries(self):
+ """
+ Convenience method to keep the same interface with POFile instances.
+ """
+ return self
+
+ def untranslated_entries(self):
+ """
+ Convenience method to keep the same interface with POFile instances.
+ """
+ return []
+
+ def fuzzy_entries(self):
+ """
+ Convenience method to keep the same interface with POFile instances.
+ """
+ return []
+
+ def obsolete_entries(self):
+ """
+ Convenience method to keep the same interface with POFile instances.
+ """
+ return []
+
+# }}}
+# class _BaseEntry {{{
+
+class _BaseEntry(object):
+ """
+ Base class for :class:`~polib.POEntry` and :class:`~polib.MOEntry` classes.
+ This class should **not** be instanciated directly.
+ """
+
+ def __init__(self, *args, **kwargs):
+ """
+ TODO: document keyword arguments.
+ """
+ self.msgid = kwargs.get('msgid', '')
+ self.msgstr = kwargs.get('msgstr', '')
+ self.msgid_plural = kwargs.get('msgid_plural', '')
+ self.msgstr_plural = kwargs.get('msgstr_plural', {})
+ self.obsolete = kwargs.get('obsolete', False)
+ self.encoding = kwargs.get('encoding', default_encoding)
+ self.msgctxt = kwargs.get('msgctxt', None)
+
+ def __repr__(self):
+ """
+ Returns the python representation of the entry.
+ """
+ return '<%s instance at %x>' % (self.__class__.__name__, id(self))
+
+ def __unicode__(self, wrapwidth=78):
+ """
+ Returns the unicode representation of the entry.
+ """
+ if self.obsolete:
+ delflag = '#~ '
+ else:
+ delflag = ''
+ ret = []
+ # write the msgctxt if any
+ if self.msgctxt is not None:
+ ret += self._str_field("msgctxt", delflag, "", self.msgctxt, wrapwidth)
+ # write the msgid
+ ret += self._str_field("msgid", delflag, "", self.msgid, wrapwidth)
+ # write the msgid_plural if any
+ if self.msgid_plural:
+ ret += self._str_field("msgid_plural", delflag, "", self.msgid_plural, wrapwidth)
+ if self.msgstr_plural:
+ # write the msgstr_plural if any
+ msgstrs = self.msgstr_plural
+ keys = list(msgstrs)
+ keys.sort()
+ for index in keys:
+ msgstr = msgstrs[index]
+ plural_index = '[%s]' % index
+ ret += self._str_field("msgstr", delflag, plural_index, msgstr, wrapwidth)
+ else:
+ # otherwise write the msgstr
+ ret += self._str_field("msgstr", delflag, "", self.msgstr, wrapwidth)
+ ret.append('')
+ ret = '\n'.join(ret)
+
+ if type(ret) != types.UnicodeType:
+ return unicode(ret, self.encoding)
+ return ret
+
+ def __str__(self):
+ """
+ Returns the string representation of the entry.
+ """
+ return unicode(self).encode(self.encoding)
+
+ def __eq__(self, other):
+ return unicode(self) == unicode(other)
+
+ def _str_field(self, fieldname, delflag, plural_index, field, wrapwidth=78):
+ lines = field.splitlines(True)
+ if len(lines) > 1:
+ lines = ['']+lines # start with initial empty line
+ else:
+ escaped_field = escape(field)
+ specialchars_count = 0
+ for c in ['\\', '\n', '\r', '\t', '"']:
+ specialchars_count += field.count(c)
+ # comparison must take into account fieldname length + one space
+ # + 2 quotes (eg. msgid "<string>")
+ real_wrapwidth = wrapwidth - (len(fieldname)+3) + specialchars_count
+ if wrapwidth > 0 and len(field) > real_wrapwidth:
+ # Wrap the line but take field name into account
+ lines = [''] + [unescape(item) for item in wrap(
+ escaped_field,
+ wrapwidth - 2, # 2 for quotes ""
+ drop_whitespace=False,
+ break_long_words=False
+ )]
+ else:
+ lines = [field] # needed for the empty string case
+ #lines = [field] # needed for the empty string case
+ if fieldname.startswith('previous_'):
+ # quick and dirty trick to get the real field name
+ fieldname = fieldname[9:]
+
+ ret = ['%s%s%s "%s"' % (delflag, fieldname, plural_index,
+ escape(lines.pop(0)))]
+ for mstr in lines:
+ ret.append('%s"%s"' % (delflag, escape(mstr)))
+ return ret
+
+# }}}
+# class POEntry {{{
+
+class POEntry(_BaseEntry):
+ """
+ Represents a po file entry.
+ """
+
+ def __init__(self, *args, **kwargs):
+ """
+ TODO: document keyword arguments.
+ """
+ _BaseEntry.__init__(self, *args, **kwargs)
+ self.comment = kwargs.get('comment', '')
+ self.tcomment = kwargs.get('tcomment', '')
+ self.occurrences = kwargs.get('occurrences', [])
+ self.flags = kwargs.get('flags', [])
+ self.previous_msgctxt = kwargs.get('previous_msgctxt', None)
+ self.previous_msgid = kwargs.get('previous_msgid', None)
+ self.previous_msgid_plural = kwargs.get('previous_msgid_plural', None)
+
+ def __unicode__(self, wrapwidth=78):
+ """
+ Returns the unicode representation of the entry.
+ """
+ if self.obsolete:
+ return _BaseEntry.__unicode__(self, wrapwidth)
+ ret = []
+ # comment first, if any (with text wrapping as xgettext does)
+ if self.comment != '':
+ for comment in self.comment.split('\n'):
+ if wrapwidth > 0 and len(comment) > wrapwidth-3:
+ ret += wrap(comment, wrapwidth,
+ initial_indent='#. ',
+ subsequent_indent='#. ',
+ break_long_words=False)
+ else:
+ ret.append('#. %s' % comment)
+ # translator comment, if any (with text wrapping as xgettext does)
+ if self.tcomment != '':
+ for tcomment in self.tcomment.split('\n'):
+ if wrapwidth > 0 and len(tcomment) > wrapwidth-2:
+ ret += wrap(tcomment, wrapwidth,
+ initial_indent='# ',
+ subsequent_indent='# ',
+ break_long_words=False)
+ else:
+ ret.append('# %s' % tcomment)
+ # occurrences (with text wrapping as xgettext does)
+ if self.occurrences:
+ filelist = []
+ for fpath, lineno in self.occurrences:
+ if lineno:
+ filelist.append('%s:%s' % (fpath, lineno))
+ else:
+ filelist.append(fpath)
+ filestr = ' '.join(filelist)
+ if wrapwidth > 0 and len(filestr)+3 > wrapwidth:
+ # XXX textwrap split words that contain hyphen, this is not
+ # what we want for filenames, so the dirty hack is to
+ # temporally replace hyphens with a char that a file cannot
+ # contain, like "*"
+ lines = wrap(filestr.replace('-', '*'),
+ wrapwidth,
+ initial_indent='#: ',
+ subsequent_indent='#: ',
+ break_long_words=False)
+ # end of the replace hack
+ for line in lines:
+ ret.append(line.replace('*', '-'))
+ else:
+ ret.append('#: '+filestr)
+ # flags
+ if self.flags:
+ flags = []
+ for flag in self.flags:
+ flags.append(flag)
+ ret.append('#, %s' % ', '.join(flags))
+
+ # previous context and previous msgid/msgid_plural
+ if self.previous_msgctxt:
+ ret += self._str_field("previous_msgctxt", "#| ", "",
+ self.previous_msgctxt, wrapwidth)
+ if self.previous_msgid:
+ ret += self._str_field("previous_msgid", "#| ", "",
+ self.previous_msgid, wrapwidth)
+ if self.previous_msgid_plural:
+ ret += self._str_field("previous_msgid_plural", "#| ", "",
+ self.previous_msgid_plural, wrapwidth)
+
+ ret.append(_BaseEntry.__unicode__(self, wrapwidth))
+ ret = '\n'.join(ret)
+
+ if type(ret) != types.UnicodeType:
+ return unicode(ret, self.encoding)
+ return ret
+
+ def __cmp__(self, other):
+ """
+ Called by comparison operations if rich comparison is not defined.
+ """
+ def compare_occurrences(a, b):
+ """
+ Compare an entry occurrence with another one.
+ """
+ if a[0] != b[0]:
+ return a[0] < b[0]
+ if a[1] != b[1]:
+ return a[1] < b[1]
+ return 0
+
+ # First: Obsolete test
+ if self.obsolete != other.obsolete:
+ if self.obsolete:
+ return -1
+ else:
+ return 1
+ # Work on a copy to protect original
+ occ1 = self.occurrences[:]
+ occ2 = other.occurrences[:]
+ # Sorting using compare method
+ occ1.sort(compare_occurrences)
+ occ2.sort(compare_occurrences)
+ # Comparing sorted occurrences
+ pos = 0
+ for entry1 in occ1:
+ try:
+ entry2 = occ2[pos]
+ except IndexError:
+ return 1
+ pos = pos + 1
+ if entry1[0] != entry2[0]:
+ if entry1[0] > entry2[0]:
+ return 1
+ else:
+ return -1
+ if entry1[1] != entry2[1]:
+ if entry1[1] > entry2[1]:
+ return 1
+ else:
+ return -1
+ # Finally: Compare message ID
+ if self.msgid > other.msgid: return 1
+ else: return -1
+
+ def translated(self):
+ """
+ Returns ``True`` if the entry has been translated or ``False``
+ otherwise.
+ """
+ if self.obsolete or 'fuzzy' in self.flags:
+ return False
+ if self.msgstr != '':
+ return True
+ if self.msgstr_plural:
+ for pos in self.msgstr_plural:
+ if self.msgstr_plural[pos] == '':
+ return False
+ return True
+ return False
+
+ def merge(self, other):
+ """
+ Merge the current entry with the given pot entry.
+ """
+ self.msgid = other.msgid
+ self.msgctxt = other.msgctxt
+ self.occurrences = other.occurrences
+ self.comment = other.comment
+ fuzzy = 'fuzzy' in self.flags
+ self.flags = other.flags[:] # clone flags
+ if fuzzy:
+ self.flags.append('fuzzy')
+ self.msgid_plural = other.msgid_plural
+ self.obsolete = other.obsolete
+ self.previous_msgctxt = other.previous_msgctxt
+ self.previous_msgid = other.previous_msgid
+ self.previous_msgid_plural = other.previous_msgid_plural
+ if other.msgstr_plural:
+ for pos in other.msgstr_plural:
+ try:
+ # keep existing translation at pos if any
+ self.msgstr_plural[pos]
+ except KeyError:
+ self.msgstr_plural[pos] = ''
+
+# }}}
+# class MOEntry {{{
+
+class MOEntry(_BaseEntry):
+ """
+ Represents a mo file entry.
+ """
+ pass
+
+# }}}
+# class _POFileParser {{{
+
+class _POFileParser(object):
+ """
+ A finite state machine to parse efficiently and correctly po
+ file format.
+ """
+
+ def __init__(self, pofile, *args, **kwargs):
+ """
+ Constructor.
+
+ Keyword arguments:
+
+ ``pofile``
+ string, path to the po file or its content
+
+ ``encoding``
+ string, the encoding to use, defaults to ``default_encoding``
+ global variable (optional).
+
+ ``check_for_duplicates``
+ whether to check for duplicate entries when adding entries to the
+ file (optional, default: ``False``).
+ """
+ enc = kwargs.get('encoding', default_encoding)
+ if os.path.exists(pofile):
+ try:
+ self.fhandle = codecs.open(pofile, 'rU', enc)
+ except LookupError:
+ enc = default_encoding
+ self.fhandle = codecs.open(pofile, 'rU', enc)
+ else:
+ self.fhandle = pofile.splitlines()
+
+ self.instance = POFile(
+ pofile=pofile,
+ encoding=enc,
+ check_for_duplicates=kwargs.get('check_for_duplicates', False)
+ )
+ self.transitions = {}
+ self.current_entry = POEntry()
+ self.current_state = 'ST'
+ self.current_token = None
+ self.current_line = 0
+ # two memo flags used in handlers
+ self.msgstr_index = 0
+ self.entry_obsolete = 0
+ # Configure the state machine, by adding transitions.
+ # Signification of symbols:
+ # * ST: Beginning of the file (start)
+ # * HE: Header
+ # * TC: a translation comment
+ # * GC: a generated comment
+ # * OC: a file/line occurence
+ # * FL: a flags line
+ # * CT: a message context
+ # * PC: a previous msgctxt
+ # * PM: a previous msgid
+ # * PP: a previous msgid_plural
+ # * MI: a msgid
+ # * MP: a msgid plural
+ # * MS: a msgstr
+ # * MX: a msgstr plural
+ # * MC: a msgid or msgstr continuation line
+ all = ['ST', 'HE', 'GC', 'OC', 'FL', 'CT', 'PC', 'PM', 'PP', 'TC',
+ 'MS', 'MP', 'MX', 'MI']
+
+ self.add('TC', ['ST', 'HE'], 'HE')
+ self.add('TC', ['GC', 'OC', 'FL', 'TC', 'PC', 'PM', 'PP', 'MS',
+ 'MP', 'MX', 'MI'], 'TC')
+ self.add('GC', all, 'GC')
+ self.add('OC', all, 'OC')
+ self.add('FL', all, 'FL')
+ self.add('PC', all, 'PC')
+ self.add('PM', all, 'PM')
+ self.add('PP', all, 'PP')
+ self.add('CT', ['ST', 'HE', 'GC', 'OC', 'FL', 'TC', 'PC', 'PM',
+ 'PP', 'MS', 'MX'], 'CT')
+ self.add('MI', ['ST', 'HE', 'GC', 'OC', 'FL', 'CT', 'TC', 'PC',
+ 'PM', 'PP', 'MS', 'MX'], 'MI')
+ self.add('MP', ['TC', 'GC', 'PC', 'PM', 'PP', 'MI'], 'MP')
+ self.add('MS', ['MI', 'MP', 'TC'], 'MS')
+ self.add('MX', ['MI', 'MX', 'MP', 'TC'], 'MX')
+ self.add('MC', ['CT', 'MI', 'MP', 'MS', 'MX', 'PM', 'PP', 'PC'], 'MC')
+
+ def parse(self):
+ """
+ Run the state machine, parse the file line by line and call process()
+ with the current matched symbol.
+ """
+ i, lastlen = 1, 0
+ for line in self.fhandle:
+ line = line.strip()
+ if line == '':
+ i = i+1
+ continue
+ if line[:3] == '#~ ':
+ line = line[3:]
+ self.entry_obsolete = 1
+ else:
+ self.entry_obsolete = 0
+ self.current_token = line
+ self.current_line = i
+ if line[:2] == '#:':
+ # we are on a occurrences line
+ self.process('OC', i)
+ elif line[:9] == 'msgctxt "':
+ # we are on a msgctxt
+ self.process('CT', i)
+ elif line[:7] == 'msgid "':
+ # we are on a msgid
+ self.process('MI', i)
+ elif line[:8] == 'msgstr "':
+ # we are on a msgstr
+ self.process('MS', i)
+ elif line[:1] == '"' or line[:4] == '#| "':
+ # we are on a continuation line or some metadata
+ self.process('MC', i)
+ elif line[:14] == 'msgid_plural "':
+ # we are on a msgid plural
+ self.process('MP', i)
+ elif line[:7] == 'msgstr[':
+ # we are on a msgstr plural
+ self.process('MX', i)
+ elif line[:3] == '#, ':
+ # we are on a flags line
+ self.process('FL', i)
+ elif line[:2] == '# ' or line == '#':
+ if line == '#': line = line + ' '
+ # we are on a translator comment line
+ self.process('TC', i)
+ elif line[:2] == '#.':
+ # we are on a generated comment line
+ self.process('GC', i)
+ elif line[:15] == '#| msgid_plural':
+ # we are on a previous msgid_plural
+ self.process('PP', i)
+ elif line[:8] == '#| msgid':
+ self.process('PM', i)
+ # we are on a previous msgid
+ elif line[:10] == '#| msgctxt':
+ # we are on a previous msgctxt
+ self.process('PC', i)
+ i = i+1
+
+ if self.current_entry:
+ # since entries are added when another entry is found, we must add
+ # the last entry here (only if there are lines)
+ self.instance.append(self.current_entry)
+ # before returning the instance, check if there's metadata and if
+ # so extract it in a dict
+ firstentry = self.instance[0]
+ if firstentry.msgid == '': # metadata found
+ # remove the entry
+ firstentry = self.instance.pop(0)
+ self.instance.metadata_is_fuzzy = firstentry.flags
+ key = None
+ for msg in firstentry.msgstr.splitlines():
+ try:
+ key, val = msg.split(':', 1)
+ self.instance.metadata[key] = val.strip()
+ except:
+ if key is not None:
+ self.instance.metadata[key] += '\n'+ msg.strip()
+ # close opened file
+ if isinstance(self.fhandle, file):
+ self.fhandle.close()
+ return self.instance
+
+ def add(self, symbol, states, next_state):
+ """
+ Add a transition to the state machine.
+
+ Keywords arguments:
+
+ ``symbol``
+ string, the matched token (two chars symbol).
+
+ ``states``
+ list, a list of states (two chars symbols).
+
+ ``next_state``
+ the next state the fsm will have after the action.
+ """
+ for state in states:
+ action = getattr(self, 'handle_%s' % next_state.lower())
+ self.transitions[(symbol, state)] = (action, next_state)
+
+ def process(self, symbol, linenum):
+ """
+ Process the transition corresponding to the current state and the
+ symbol provided.
+
+ Keywords arguments:
+
+ ``symbol``
+ string, the matched token (two chars symbol).
+
+ ``linenum``
+ integer, the current line number of the parsed file.
+ """
+ try:
+ (action, state) = self.transitions[(symbol, self.current_state)]
+ if action():
+ self.current_state = state
+ except Exception, exc:
+ raise IOError('Syntax error in po file (line %s)' % linenum)
+
+ # state handlers
+
+ def handle_he(self):
+ """Handle a header comment."""
+ if self.instance.header != '':
+ self.instance.header += '\n'
+ self.instance.header += self.current_token[2:]
+ return 1
+
+ def handle_tc(self):
+ """Handle a translator comment."""
+ if self.current_state in ['MC', 'MS', 'MX']:
+ self.instance.append(self.current_entry)
+ self.current_entry = POEntry()
+ if self.current_entry.tcomment != '':
+ self.current_entry.tcomment += '\n'
+ self.current_entry.tcomment += self.current_token[2:]
+ return True
+
+ def handle_gc(self):
+ """Handle a generated comment."""
+ if self.current_state in ['MC', 'MS', 'MX']:
+ self.instance.append(self.current_entry)
+ self.current_entry = POEntry()
+ if self.current_entry.comment != '':
+ self.current_entry.comment += '\n'
+ self.current_entry.comment += self.current_token[3:]
+ return True
+
+ def handle_oc(self):
+ """Handle a file:num occurence."""
+ if self.current_state in ['MC', 'MS', 'MX']:
+ self.instance.append(self.current_entry)
+ self.current_entry = POEntry()
+ occurrences = self.current_token[3:].split()
+ for occurrence in occurrences:
+ if occurrence != '':
+ try:
+ fil, line = occurrence.split(':')
+ if not line.isdigit():
+ fil = fil + line
+ line = ''
+ self.current_entry.occurrences.append((fil, line))
+ except:
+ self.current_entry.occurrences.append((occurrence, ''))
+ return True
+
+ def handle_fl(self):
+ """Handle a flags line."""
+ if self.current_state in ['MC', 'MS', 'MX']:
+ self.instance.append(self.current_entry)
+ self.current_entry = POEntry()
+ self.current_entry.flags += self.current_token[3:].split(', ')
+ return True
+
+ def handle_pp(self):
+ """Handle a previous msgid_plural line."""
+ if self.current_state in ['MC', 'MS', 'MX']:
+ self.instance.append(self.current_entry)
+ self.current_entry = POEntry()
+ self.current_entry.previous_msgid_plural = \
+ unescape(self.current_token[17:-1])
+ return True
+
+ def handle_pm(self):
+ """Handle a previous msgid line."""
+ if self.current_state in ['MC', 'MS', 'MX']:
+ self.instance.append(self.current_entry)
+ self.current_entry = POEntry()
+ self.current_entry.previous_msgid = \
+ unescape(self.current_token[10:-1])
+ return True
+
+ def handle_pc(self):
+ """Handle a previous msgctxt line."""
+ if self.current_state in ['MC', 'MS', 'MX']:
+ self.instance.append(self.current_entry)
+ self.current_entry = POEntry()
+ self.current_entry.previous_msgctxt = \
+ unescape(self.current_token[12:-1])
+ return True
+
+ def handle_ct(self):
+ """Handle a msgctxt."""
+ if self.current_state in ['MC', 'MS', 'MX']:
+ self.instance.append(self.current_entry)
+ self.current_entry = POEntry()
+ self.current_entry.msgctxt = unescape(self.current_token[9:-1])
+ return True
+
+ def handle_mi(self):
+ """Handle a msgid."""
+ if self.current_state in ['MC', 'MS', 'MX']:
+ self.instance.append(self.current_entry)
+ self.current_entry = POEntry()
+ self.current_entry.obsolete = self.entry_obsolete
+ self.current_entry.msgid = unescape(self.current_token[7:-1])
+ return True
+
+ def handle_mp(self):
+ """Handle a msgid plural."""
+ self.current_entry.msgid_plural = unescape(self.current_token[14:-1])
+ return True
+
+ def handle_ms(self):
+ """Handle a msgstr."""
+ self.current_entry.msgstr = unescape(self.current_token[8:-1])
+ self.current_entry.linenum = str(self.current_line)
+ return True
+
+ def handle_mx(self):
+ """Handle a msgstr plural."""
+ index, value = self.current_token[7], self.current_token[11:-1]
+ self.current_entry.msgstr_plural[index] = unescape(value)
+ self.msgstr_index = index
+ return True
+
+ def handle_mc(self):
+ """Handle a msgid or msgstr continuation line."""
+ token = unescape(self.current_token[1:-1])
+ if self.current_state == 'CT':
+ typ = 'msgctxt'
+ self.current_entry.msgctxt += token
+ elif self.current_state == 'MI':
+ typ = 'msgid'
+ self.current_entry.msgid += token
+ elif self.current_state == 'MP':
+ typ = 'msgid_plural'
+ self.current_entry.msgid_plural += token
+ elif self.current_state == 'MS':
+ typ = 'msgstr'
+ self.current_entry.msgstr += token
+ elif self.current_state == 'MX':
+ typ = 'msgstr[%s]' % self.msgstr_index
+ self.current_entry.msgstr_plural[self.msgstr_index] += token
+ elif self.current_state == 'PP':
+ typ = 'previous_msgid_plural'
+ token = token[3:]
+ self.current_entry.previous_msgid_plural += token
+ elif self.current_state == 'PM':
+ typ = 'previous_msgid'
+ token = token[3:]
+ self.current_entry.previous_msgid += token
+ elif self.current_state == 'PC':
+ typ = 'previous_msgctxt'
+ token = token[3:]
+ self.current_entry.previous_msgctxt += token
+ # don't change the current state
+ return False
+
+# }}}
+# class _MOFileParser {{{
+
+class _MOFileParser(object):
+ """
+ A class to parse binary mo files.
+ """
+
+ def __init__(self, mofile, *args, **kwargs):
+ """
+ Constructor.
+
+ Keyword arguments:
+
+ ``mofile``
+ string, path to the mo file or its content
+
+ ``encoding``
+ string, the encoding to use, defaults to ``default_encoding``
+ global variable (optional).
+
+ ``check_for_duplicates``
+ whether to check for duplicate entries when adding entries to the
+ file (optional, default: ``False``).
+ """
+ self.fhandle = open(mofile, 'rb')
+ self.instance = MOFile(
+ fpath=mofile,
+ encoding=kwargs.get('encoding', default_encoding),
+ check_for_duplicates=kwargs.get('check_for_duplicates', False)
+ )
+
+ def parse(self):
+ """
+ Build the instance with the file handle provided in the
+ constructor.
+ """
+ # parse magic number
+ magic_number = self._readbinary('<I', 4)
+ if magic_number == MOFile.LITTLE_ENDIAN:
+ ii = '<II'
+ elif magic_number == MOFile.BIG_ENDIAN:
+ ii = '>II'
+ else:
+ raise IOError('Invalid mo file, magic number is incorrect !')
+ self.instance.magic_number = magic_number
+ # parse the version number and the number of strings
+ self.instance.version, numofstrings = self._readbinary(ii, 8)
+ # original strings and translation strings hash table offset
+ msgids_hash_offset, msgstrs_hash_offset = self._readbinary(ii, 8)
+ # move to msgid hash table and read length and offset of msgids
+ self.fhandle.seek(msgids_hash_offset)
+ msgids_index = []
+ for i in range(numofstrings):
+ msgids_index.append(self._readbinary(ii, 8))
+ # move to msgstr hash table and read length and offset of msgstrs
+ self.fhandle.seek(msgstrs_hash_offset)
+ msgstrs_index = []
+ for i in range(numofstrings):
+ msgstrs_index.append(self._readbinary(ii, 8))
+ # build entries
+ for i in range(numofstrings):
+ self.fhandle.seek(msgids_index[i][1])
+ msgid = self.fhandle.read(msgids_index[i][0])
+ self.fhandle.seek(msgstrs_index[i][1])
+ msgstr = self.fhandle.read(msgstrs_index[i][0])
+ if i == 0: # metadata
+ raw_metadata, metadata = msgstr.split('\n'), {}
+ for line in raw_metadata:
+ tokens = line.split(':', 1)
+ if tokens[0] != '':
+ try:
+ metadata[tokens[0]] = tokens[1].strip()
+ except IndexError:
+ metadata[tokens[0]] = ''
+ self.instance.metadata = metadata
+ continue
+ # test if we have a plural entry
+ msgid_tokens = msgid.split('\0')
+ if len(msgid_tokens) > 1:
+ entry = MOEntry(
+ msgid=msgid_tokens[0],
+ msgid_plural=msgid_tokens[1],
+ msgstr_plural=dict((k,v) for k,v in \
+ enumerate(msgstr.split('\0')))
+ )
+ else:
+ entry = MOEntry(msgid=msgid, msgstr=msgstr)
+ self.instance.append(entry)
+ # close opened file
+ self.fhandle.close()
+ return self.instance
+
+ def _readbinary(self, fmt, numbytes):
+ """
+ Private method that unpack n bytes of data using format <fmt>.
+ It returns a tuple or a mixed value if the tuple length is 1.
+ """
+ bytes = self.fhandle.read(numbytes)
+ tup = struct.unpack(fmt, bytes)
+ if len(tup) == 1:
+ return tup[0]
+ return tup
+
+# }}}
+# class TextWrapper {{{
+
+class TextWrapper(textwrap.TextWrapper):
+ """
+ Subclass of textwrap.TextWrapper that backport the
+ drop_whitespace option.
+ """
+ def __init__(self, *args, **kwargs):
+ drop_whitespace = kwargs.pop('drop_whitespace', True)
+ textwrap.TextWrapper.__init__(self, *args, **kwargs)
+ self.drop_whitespace = drop_whitespace
+
+ def _wrap_chunks(self, chunks):
+ """_wrap_chunks(chunks : [string]) -> [string]
+
+ Wrap a sequence of text chunks and return a list of lines of
+ length 'self.width' or less. (If 'break_long_words' is false,
+ some lines may be longer than this.) Chunks correspond roughly
+ to words and the whitespace between them: each chunk is
+ indivisible (modulo 'break_long_words'), but a line break can
+ come between any two chunks. Chunks should not have internal
+ whitespace; ie. a chunk is either all whitespace or a "word".
+ Whitespace chunks will be removed from the beginning and end of
+ lines, but apart from that whitespace is preserved.
+ """
+ lines = []
+ if self.width <= 0:
+ raise ValueError("invalid width %r (must be > 0)" % self.width)
+
+ # Arrange in reverse order so items can be efficiently popped
+ # from a stack of chucks.
+ chunks.reverse()
+
+ while chunks:
+
+ # Start the list of chunks that will make up the current line.
+ # cur_len is just the length of all the chunks in cur_line.
+ cur_line = []
+ cur_len = 0
+
+ # Figure out which static string will prefix this line.
+ if lines:
+ indent = self.subsequent_indent
+ else:
+ indent = self.initial_indent
+
+ # Maximum width for this line.
+ width = self.width - len(indent)
+
+ # First chunk on line is whitespace -- drop it, unless this
+ # is the very beginning of the text (ie. no lines started yet).
+ if self.drop_whitespace and chunks[-1].strip() == '' and lines:
+ del chunks[-1]
+
+ while chunks:
+ l = len(chunks[-1])
+
+ # Can at least squeeze this chunk onto the current line.
+ if cur_len + l <= width:
+ cur_line.append(chunks.pop())
+ cur_len += l
+
+ # Nope, this line is full.
+ else:
+ break
+
+ # The current line is full, and the next chunk is too big to
+ # fit on *any* line (not just this one).
+ if chunks and len(chunks[-1]) > width:
+ self._handle_long_word(chunks, cur_line, cur_len, width)
+
+ # If the last chunk on this line is all whitespace, drop it.
+ if self.drop_whitespace and cur_line and cur_line[-1].strip() == '':
+ del cur_line[-1]
+
+ # Convert current line back to a string and store it in list
+ # of all lines (return value).
+ if cur_line:
+ lines.append(indent + ''.join(cur_line))
+
+ return lines
+
+# }}}
+# function wrap() {{{
+
+def wrap(text, width=70, **kwargs):
+ """
+ Wrap a single paragraph of text, returning a list of wrapped lines.
+ """
+ if sys.version_info < (2, 6):
+ return TextWrapper(width=width, **kwargs).wrap(text)
+ return textwrap.wrap(text, width=width, **kwargs)
+
+#}}}
Added: translation/trunk/tools/validate.py
===================================================================
--- translation/trunk/tools/validate.py (rev 0)
+++ translation/trunk/tools/validate.py 2011-03-17 17:52:17 UTC (rev 24379)
@@ -0,0 +1,94 @@
+#! /bin/env python
+
+from __future__ import unicode_literals, print_function
+
+import polib
+import sys
+import getopt
+
+symbols = [
+ '(', ')', '<', '>', '{',
+ '}', '[', ']', '\"', ';'
+ ]
+
+counts = {}
+
+def reset():
+ for s in symbols:
+ counts[s] = 0
+
+def parse_orig_string(string):
+ for ch in string:
+ if ch in symbols:
+ counts[ch] += 1
+
+def parse_trans_string(string):
+ for ch in string:
+ if ch in symbols:
+ counts[ch] -= 1
+
+def get_strings(file):
+ entries = []
+ po = polib.pofile(file, autodetect_encoding=True)
+
+ for entry in po:
+ entries.append((entry.msgid, entry.msgstr, entry.linenum))
+
+ return entries
+
+def warn():
+ for s in symbols:
+ if counts[s] < 0:
+ return True
+ return False
+
+def usage():
+ print("Usage: %s -i/--input=<file.po> -l/--logfile=<logfile>" % sys.argv[0])
+
+def log(string, file_to_check, log_file, linenum = -1):
+ f = open(log_file, "a+")
+ if linenum == -1:
+ f.write(("%s: %s\n" % (file_to_check, string)).encode("utf-8"))
+ else:
+ f.write(("%s (%s): %s\n" % (file_to_check, linenum, string)).encode("utf-8"))
+ f.close()
+
+def check(file_to_check, log_file):
+ errors = 0
+
+ strings = get_strings(file_to_check)
+ for (orig, trans, linenum) in strings:
+ reset()
+ parse_orig_string(orig)
+ parse_trans_string(trans)
+ if warn():
+ errors += 1
+ log(trans, file_to_check, log_file, linenum)
+
+ if errors != 0:
+ log("Total count of warnings %d\n" % errors, file_to_check, log_file)
+
+if __name__ == '__main__':
+ try:
+ opts, args = getopt.getopt(sys.argv[1:], "i:hl:", ["input=", "help", "logfile="])
+ except getopt.GetoptError, err:
+ print(str(err))
+ sys.exit(2)
+
+ file_to_check = None
+ log_file = None
+
+ for opt, arg in opts:
+ if opt in ("-i", "--input"):
+ file_to_check = arg
+ elif opt in ("-h", "--help"):
+ usage()
+ sys.exit()
+ elif opt in ("-l", "--logfile"):
+ log_file = arg
+
+ if file_to_check is None or log_file is None:
+ print("ERROR: You need to specify both the input and the logfile")
+ sys.exit(2)
+
+ check(file_to_check, log_file)