+ 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,
+ 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) \
+ & whoosh.query.Term(u'forme_name', u'__empty__')
+
+ # If there's a space in the input, this might be a form
+ if ' ' in name:
+ form, formless_name = name.split(' ', 1)
+ form_query = whoosh.query.Term(u'name', formless_name) \
+ & whoosh.query.Term(u'forme_name', form)
+ query = query | form_query
+
+ ### 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:
+ exact = False
+ results = []
+
+ for suggestion in self.speller.suggest(
+ name, self.INTERMEDIATE_LOOKUP_RESULTS):
+
+ query = whoosh.query.Term('name', suggestion)
+ results.extend(searcher.search(query))
+
+ ### Convert results to db objects
+ objects = self._whoosh_records_to_results(results, exact=exact)
+
+ # Only return up to 10 matches; beyond that, something is wrong. We
+ # strip out duplicate entries above, so it's remotely possible that we
+ # should have more than 10 here and lost a few. The speller returns 25
+ # to give us some padding, and should avoid that problem. Not a big
+ # deal if we lose the 25th-most-likely match anyway.
+ return objects[:self.MAX_LOOKUP_RESULTS]
+
+
+ def random_lookup(self, valid_types=[]):
+ """Returns a random lookup result from one of the provided
+ `valid_types`.
+ """
+
+ tables = []
+ for valid_type in valid_types:
+ table_name = self._parse_table_name(valid_type)
+ if table_name:
+ tables.append(self.indexed_tables[table_name])
+
+ if not tables:
+ # n.b.: It's possible we got a list of valid_types and none of them
+ # were valid, but this function is guaranteed to return
+ # *something*, so it politely selects from the entire index isntead
+ tables = self.indexed_tables.values()
+
+ # Rather than create an array of many hundred items and pick randomly
+ # from it, just pick a number up to the total number of potential
+ # items, then pick randomly from that, and partition the whole range
+ # into chunks. This also avoids the slight problem that the index
+ # contains more rows (for languages) for some items than others.
+ # XXX ought to cache this (in the index?) if possible
+ total = 0
+ partitions = []
+ for table in tables:
+ count = self.session.query(table).count()
+ total += count
+ partitions.append((table, count))
+
+ n = random.randint(1, total)
+ while n > partitions[0][1]:
+ n -= partitions[0][1]
+ partitions.pop(0)
+
+ return self.lookup(unicode(n), valid_types=[ partitions[0][0] ])
+
+ def prefix_lookup(self, prefix):
+ """Returns terms starting with the given exact prefix.
+
+ No special magic is currently done with the name; type prefixes are not
+ recognized.
+ """
+
+ query = whoosh.query.Prefix(u'name', self.normalize_name(prefix))
+
+ searcher = self.index.searcher()
+ searcher.weighting = LanguageWeighting()
+ results = searcher.search(query) # XXX , limit=self.MAX_LOOKUP_RESULTS)
+
+ return self._whoosh_records_to_results(results)
projects / zzz-pokedex.git / blobdiff