+ 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
+
+
+ def _parse_table_name(self, name):
+ """Takes a singular table name, table name, or table object and returns
+ the table name.
+
+ Returns None for a bogus name.
+ """
+ if hasattr(name, '__tablename__'):
+ return getattr(name, '__tablename__')
+ elif name in self.indexed_tables:
+ return name
+ elif name + 's' in self.indexed_tables:
+ return name + 's'
+ else:
+ # Bogus. Be nice and return dummy
+ return None
+
+ def _whoosh_records_to_results(self, records, exact=True):
+ """Converts a list of whoosh's indexed records to LookupResult tuples
+ containing database objects.
+ """
+ # XXX this 'exact' thing is getting kinda leaky. would like a better
+ # way to handle it, since only lookup() cares about fuzzy results
+ seen = {}
+ results = []
+ for record in records:
+ # Skip dupes
+ seen_key = record['table'], record['row_id']
+ if seen_key in seen:
+ continue
+ seen[seen_key] = True
+
+ cls = self.indexed_tables[record['table']]
+ obj = self.session.query(cls).get(record['row_id'])
+
+ results.append(LookupResult(object=obj,
+ indexed_name=record['name'],
+ name=record['display_name'],
+ language=record['language'],
+ iso3166=record['iso3166'],
+ exact=exact))
+
+ return results
+
+
+ def lookup(self, input, valid_types=[], exact_only=False):
+ """Attempts to find some sort of object, given a name.
+
+ Returns a list of named (object, name, language, iso3166, exact)
+ tuples. `object` is a database object, `name` is the name under which
+ the object was found, `language` and `iso3166` are the name and country
+ code of the language in which the name was found, and `exact` is True
+ iff this was an
+ exact match.
+
+ This function currently ONLY does fuzzy matching if there are no exact
+ matches.
+
+ Formes are not returned unless requested; "Shaymin" will return only
+ grass Shaymin.
+
+ Extraneous whitespace is removed with extreme prejudice.
+
+ Recognizes:
+ - Names: "Eevee", "Surf", "Run Away", "Payapa Berry", etc.
+ - Foreign names: "Iibui", "Eivui"
+ - Fuzzy names in whatever language: "Evee", "Ibui"
+ - IDs: "133", "192", "250"
+ Also:
+ - Type restrictions. "type:psychic" will only return the type. This
+ is how to make ID lookup useful. Multiple type specs can be entered
+ with commas, as "move,item:1". If `valid_types` are provided, any
+ type prefix will be ignored.
+ - Alternate formes can be specified merely like "wash rotom".
+
+ `input`
+ Name of the thing to look for.
+
+ `valid_types`
+ A list of table objects or names, e.g., `['pokemon', 'moves']`. If
+ this is provided, only results in one of the given tables will be
+ returned.
+
+ `exact_only`
+ If True, only exact matches are returned. If set to False (the
+ default), and the provided `name` doesn't match anything exactly,
+ spelling correction will be attempted.
+ """
+
+ name = self.normalize_name(input)
+ exact = True
+ form = None
+
+ # Remove any type prefix (pokemon:133) before constructing a query
+ if ':' in name:
+ prefix_chunk, name = name.split(':', 1)
+ name = name.strip()
+
+ if not valid_types:
+ # Only use types from the query string if none were explicitly
+ # provided
+ prefixes = prefix_chunk.split(',')
+ valid_types = [_.strip() for _ in prefixes]
+
+ # Random lookup
+ if name == 'random':
+ return self.random_lookup(valid_types=valid_types)
+
+ # Do different things depending what the query looks like
+ # Note: Term objects do an exact match, so we don't have to worry about
+ # a query parser tripping on weird characters in the input
+ if '*' in name or '?' in name:
+ exact_only = True
+ query = whoosh.query.Wildcard(u'name', name)
+ elif rx_is_number.match(name):
+ # Don't spell-check numbers!
+ exact_only = True
+ query = whoosh.query.Term(u'row_id', name)
+ else:
+ # Not an integer
+ query = whoosh.query.Term(u'name', name)
+
+ ### Filter by type of object
+ type_terms = []
+ for valid_type in valid_types:
+ table_name = self._parse_table_name(valid_type)
+ if table_name:
+ # Quietly ignore bogus valid_types; more likely to DTRT
+ type_terms.append(whoosh.query.Term(u'table', table_name))
+
+ if type_terms:
+ query = query & whoosh.query.Or(type_terms)
+
+
+ ### Actual searching
+ searcher = self.index.searcher()
+ # XXX is this kosher? docs say search() takes a weighting arg, but it
+ # certainly does not
+ searcher.weighting = LanguageWeighting()
+ results = searcher.search(query,
+ limit=self.INTERMEDIATE_LOOKUP_RESULTS)
+
+ # Look for some fuzzy matches if necessary
+ if not exact_only and not results:
projects / zzz-pokedex.git / blobdiff