+ def __init__(self, directory=None, session=None, recreate=False):
+ """Opens the whoosh index stored in the named directory. If the index
+ doesn't already exist, it will be created.
+
+ `directory`
+ Directory containing the index. Defaults to a location within the
+ `pokedex` egg directory.
+
+ `session`
+ If the index needs to be created, this database session will be
+ used. Defaults to an attempt to connect to the default SQLite
+ database installed by `pokedex setup`.
+
+ `recreate`
+ If set to True, the whoosh index will be created even if it already
+ exists.
+ """
+
+ # By the time this returns, self.index, self.speller, and self.session
+ # must be set
+
+ # Defaults
+ if not directory:
+ directory = pkg_resources.resource_filename('pokedex',
+ 'data/whoosh-index')
+
+ if session:
+ self.session = session
+ else:
+ self.session = connect()
+
+ # Attempt to open or create the index
+ directory_exists = os.path.exists(directory)
+ if directory_exists and not recreate:
+ # Already exists; should be an index! Bam, done.
+ try:
+ self.index = whoosh.index.open_dir(directory, indexname='MAIN')
+ spell_store = whoosh.filedb.filestore.FileStorage(directory)
+ self.speller = whoosh.spelling.SpellChecker(spell_store)
+ return
+ except whoosh.index.EmptyIndexError as e:
+ # Apparently not a real index. Fall out and create it
+ pass
+
+ # Delete and start over if we're going to bail anyway.
+ if directory_exists and recreate:
+ # Be safe and only delete if it looks like a whoosh index, i.e.,
+ # everything starts with _
+ if all(f[0] == '_' for f in os.listdir(directory)):
+ shutil.rmtree(directory)
+ directory_exists = False
+
+ if not directory_exists:
+ os.mkdir(directory)
+
+
+ ### Create index
+ schema = whoosh.fields.Schema(
+ name=whoosh.fields.ID(stored=True),
+ table=whoosh.fields.ID(stored=True),
+ row_id=whoosh.fields.ID(stored=True),
+ language=whoosh.fields.STORED,
+ iso3166=whoosh.fields.STORED,
+ display_name=whoosh.fields.STORED, # non-lowercased name
+ )
+
+ self.index = whoosh.index.create_in(directory, schema=schema,
+ indexname='MAIN')
+ writer = self.index.writer()
+
+ # Index every name in all our tables of interest
+ # speller_entries becomes a list of (word, score) tuples; the score is
+ # 2 for English names, 1.5 for Roomaji, and 1 for everything else. I
+ # think this biases the results in the direction most people expect,
+ # especially when e.g. German names are very similar to English names
+ speller_entries = []
+ for cls in self.indexed_tables.values():
+ q = session.query(cls)
+
+ for row in q.yield_per(5):
+ row_key = dict(table=unicode(cls.__tablename__),
+ row_id=unicode(row.id))
+
+ def add(name, language, iso3166, score):
+ normalized_name = self.normalize_name(name)
+
+ writer.add_document(
+ name=normalized_name, display_name=name,
+ language=language, iso3166=iso3166,
+ **row_key
+ )
+
+ speller_entries.append((normalized_name, score))
+
+
+ # Add the basic English name to the index
+ if cls == tables.Pokemon:
+ # Pokémon need their form name added
+ # XXX kinda kludgy
+ add(row.full_name, None, u'us', 1)
+
+ # If this is a default form, ALSO add the unadorned name,
+ # so 'Deoxys' alone will still do the right thing
+ if row.forme_name and not row.forme_base_pokemon_id:
+ add(row.name, None, u'us', 1)
+ else:
+ add(row.name, None, u'us', 1)
+
+ # Some things also have other languages' names
+ # XXX other language form names..?
+ for foreign_name in getattr(row, 'foreign_names', []):
+ moonspeak = foreign_name.name
+ if row.name == moonspeak:
+ # Don't add the English name again as a different
+ # language; no point and it makes spell results
+ # confusing
+ continue
+
+ add(moonspeak, foreign_name.language.name,
+ foreign_name.language.iso3166,
+ 3)
+
+ # Add Roomaji too
+ if foreign_name.language.name == 'Japanese':
+ roomaji = romanize(foreign_name.name)
+ add(roomaji, u'Roomaji', u'jp', 8)
+
+ writer.commit()
+
+ # Construct and populate a spell-checker index. Quicker to do it all
+ # at once, as every call to add_* does a commit(), and those seem to be
+ # expensive
+ self.speller = whoosh.spelling.SpellChecker(self.index.storage)
+ self.speller.add_scored_words(speller_entries)
+
+
+ def normalize_name(self, name):
+ """Strips irrelevant formatting junk from name input.
+
+ Specifically: everything is lowercased, and accents are removed.
+ """
+ # http://stackoverflow.com/questions/517923/what-is-the-best-way-to-remove-accents-in-a-python-unicode-string
+ # Makes sense to me. Decompose by Unicode rules, then remove combining
+ # characters, then recombine. I'm explicitly doing it this way instead
+ # of testing combining() because Korean characters apparently
+ # decompose! But the results are considered letters, not combining
+ # characters, so testing for Mn works well, and combining them again
+ # makes them look right.
+ nkfd_form = unicodedata.normalize('NFKD', unicode(name))
+ name = u"".join(c for c in nkfd_form
+ if unicodedata.category(c) != 'Mn')
+ name = unicodedata.normalize('NFC', name)
+
+ name = name.strip()
+ name = name.lower()
+
+ return name
+
projects / zzz-pokedex.git / blobdiff