X-Git-Url: http://git.veekun.com/zzz-pokedex.git/blobdiff_plain/4603adc7c77a5c04b9f4a1ffca75c94442e51752..dc06f46c000cd0466cb448e548b03b5234e2fdfa:/pokedex/lookup.py diff --git a/pokedex/lookup.py b/pokedex/lookup.py index 7b10974..b00f5a7 100644 --- a/pokedex/lookup.py +++ b/pokedex/lookup.py @@ -1,7 +1,6 @@ # encoding: utf8 from collections import namedtuple import os, os.path -import pkg_resources import random import re import shutil @@ -19,6 +18,7 @@ import whoosh.spelling from pokedex.db import connect import pokedex.db.tables as tables from pokedex.roomaji import romanize +from pokedex.defaults import get_default_index_dir __all__ = ['PokedexLookup'] @@ -26,7 +26,25 @@ __all__ = ['PokedexLookup'] rx_is_number = re.compile('^\d+$') LookupResult = namedtuple('LookupResult', - ['object', 'name', 'language', 'iso3166', 'exact']) + ['object', 'indexed_name', 'name', 'language', 'iso3166', 'exact']) + +class UninitializedIndex(object): + class UninitializedIndexError(Exception): + pass + + def __nonzero__(self): + """Dummy object should identify itself as False.""" + return False + + def __bool__(self): + """Python 3000 version of the above. Future-proofing rules!""" + return False + + def __getattr__(self, *args, **kwargs): + raise self.UninitializedIndexError( + "The lookup index does not exist. Please use `pokedex setup` " + "or lookup.rebuild_index() to create it." + ) class LanguageWeighting(whoosh.scoring.Weighting): """A scoring class that forces otherwise-equal English results to come @@ -58,14 +76,16 @@ class PokedexLookup(object): for cls in ( tables.Ability, tables.Item, + tables.Location, tables.Move, + tables.Nature, tables.Pokemon, tables.Type, ) ) - def __init__(self, directory=None, session=None, recreate=False): + def __init__(self, directory=None, session=None): """Opens the whoosh index stored in the named directory. If the index doesn't already exist, it will be created. @@ -74,22 +94,19 @@ class PokedexLookup(object): `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. + Used for creating the index and retrieving objects. Defaults to an + attempt to connect to the default SQLite database installed by + `pokedex setup`. """ # 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 a directory was not given, use the default + if directory is None: + directory = get_default_index_dir() + + self.directory = directory if session: self.session = session @@ -97,31 +114,33 @@ class PokedexLookup(object): 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 + if not os.path.exists(directory) or not os.listdir(directory): + # Directory doesn't exist OR is empty; caller needs to use + # rebuild_index before doing anything. Provide a dummy object that + # complains when used + self.index = UninitializedIndex() + self.speller = UninitializedIndex() + return + + # Otherwise, already exists; should be an index! Bam, done. + # Note that this will explode if the directory exists but doesn't + # contain an index; that's a feature + try: + self.index = whoosh.index.open_dir(directory, indexname='MAIN') + except whoosh.index.EmptyIndexError: + raise IOError( + "The index directory already contains files. " + "Please use a dedicated directory for the lookup index." + ) + + # Create speller, and done + spell_store = whoosh.filedb.filestore.FileStorage(directory) + self.speller = whoosh.spelling.SpellChecker(spell_store) + + + def rebuild_index(self): + """Creates the index from scratch.""" + schema = whoosh.fields.Schema( name=whoosh.fields.ID(stored=True), table=whoosh.fields.ID(stored=True), @@ -129,11 +148,13 @@ class PokedexLookup(object): language=whoosh.fields.STORED, iso3166=whoosh.fields.STORED, display_name=whoosh.fields.STORED, # non-lowercased name - forme_name=whoosh.fields.ID, ) - self.index = whoosh.index.create_in(directory, schema=schema, - indexname='MAIN') + if not os.path.exists(self.directory): + os.mkdir(self.directory) + + self.index = whoosh.index.create_in(self.directory, schema=schema, + indexname='MAIN') writer = self.index.writer() # Index every name in all our tables of interest @@ -143,47 +164,42 @@ class PokedexLookup(object): # 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) + q = self.session.query(cls) for row in q.yield_per(5): - # Need to give forme_name a dummy value because I can't - # search for explicitly empty fields. Boo. - row_keys = [ - dict(table=unicode(cls.__tablename__), - row_id=unicode(row.id), - forme_name=u'__empty__') - ] - - # If this is a form, mark it as such - # XXX foreign form names...? - if getattr(row, 'forme_name', None): - # ...but if it's also the *default* form, index the name - # bare too - if not getattr(row, 'forme_base_pokemon_id', None): - new_key = row_keys[0].copy() - row_keys.append(new_key) - - row_keys[0]['forme_name'] = row.forme_name + row_key = dict(table=unicode(cls.__tablename__), + row_id=unicode(row.id)) def add(name, language, iso3166, score): normalized_name = self.normalize_name(name) - for row_key in row_keys: - writer.add_document( - name=normalized_name, display_name=name, - language=language, iso3166=iso3166, - **row_key - ) + + writer.add_document( + name=normalized_name, display_name=name, + language=language, iso3166=iso3166, + **row_key + ) speller_entries.append((normalized_name, score)) - name = row.name - add(name, None, u'us', 1) + # 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) - # Pokemon also get other languages + # 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 name == moonspeak: + if row.name == moonspeak: # Don't add the English name again as a different # language; no point and it makes spell results # confusing @@ -230,21 +246,81 @@ class PokedexLookup(object): return name + def _apply_valid_types(self, name, valid_types): + """Combines the enforced `valid_types` with any from the search string + itself and updates the query. + + For example, a name of 'a,b:foo' and valid_types of b,c will search for + only `b`s named "foo". + + Returns `(name, merged_valid_types, term)`, where `name` has had any type + prefix stripped, `merged_valid_types` combines the original + `valid_types` with the type prefix, and `term` is a query term for + limited to just the allowed types. If there are no type restrictions + at all, `term` will be None. + """ + + # Remove any type prefix (pokemon:133) first + user_valid_types = [] + if ':' in name: + prefix_chunk, name = name.split(':', 1) + name = name.strip() + + prefixes = prefix_chunk.split(',') + user_valid_types = [_.strip() for _ in prefixes] + + # Merge the valid types together. Only types that appear in BOTH lists + # may be used. + # As a special case, if the user asked for types that are explicitly + # forbidden, completely ignore what the user requested + combined_valid_types = [] + if user_valid_types and valid_types: + combined_valid_types = list( + set(user_valid_types) & set(combined_valid_types) + ) + + if not combined_valid_types: + # No overlap! Just use the enforced ones + combined_valid_types = valid_types + else: + # One list or the other was blank, so just use the one that isn't + combined_valid_types = valid_types + user_valid_types + + if not combined_valid_types: + # No restrictions + return name, [], None + + # Construct the term + type_terms = [] + final_valid_types = [] + for valid_type in combined_valid_types: + table_name = self._parse_table_name(valid_type) + + # Quietly ignore bogus valid_types; more likely to DTRT + if table_name: + final_valid_types.append(valid_type) + type_terms.append(whoosh.query.Term(u'table', table_name)) + + return name, final_valid_types, whoosh.query.Or(type_terms) + + 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. """ + # Table object 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 + + # Table name + for table in self.indexed_tables.values(): + if name in (table.__tablename__, table.__singlename__): + return table.__tablename__ + + # 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 @@ -265,6 +341,7 @@ class PokedexLookup(object): 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'], @@ -321,53 +398,37 @@ class PokedexLookup(object): 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] + # Pop off any type prefix and merge with valid_types + name, merged_valid_types, type_term = \ + self._apply_valid_types(name, valid_types) # Random lookup if name == 'random': - return self.random_lookup(valid_types=valid_types) + return self.random_lookup(valid_types=merged_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 + try: + # Let Python try to convert to a number, so 0xff works + name_as_number = int(name, base=0) + except ValueError: + # Oh well + name_as_number = None + if '*' in name or '?' in name: exact_only = True query = whoosh.query.Wildcard(u'name', name) - elif rx_is_number.match(name): + elif name_as_number is not None: # Don't spell-check numbers! exact_only = True - query = whoosh.query.Term(u'row_id', name) + query = whoosh.query.Term(u'row_id', unicode(name_as_number)) else: # Not an integer - query = whoosh.query.Term(u'name', name) \ - & whoosh.query.Term(u'forme_name', u'__empty__') + query = whoosh.query.Term(u'name', name) - # 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) + if type_term: + query = query & type_term ### Actual searching @@ -437,15 +498,21 @@ class PokedexLookup(object): return self.lookup(unicode(n), valid_types=[ partitions[0][0] ]) - def prefix_lookup(self, prefix): + def prefix_lookup(self, prefix, valid_types=[]): """Returns terms starting with the given exact prefix. - No special magic is currently done with the name; type prefixes are not - recognized. + Type prefixes are recognized, but no other name munging is done. """ + # Pop off any type prefix and merge with valid_types + prefix, merged_valid_types, type_term = \ + self._apply_valid_types(prefix, valid_types) + query = whoosh.query.Prefix(u'name', self.normalize_name(prefix)) + if type_term: + query = query & type_term + searcher = self.index.searcher() searcher.weighting = LanguageWeighting() results = searcher.search(query) # XXX , limit=self.MAX_LOOKUP_RESULTS)