2009-07-21 07:12:25 +00:00
|
|
|
# encoding: utf8
|
2009-08-17 04:03:49 +00:00
|
|
|
import os, os.path
|
2009-08-25 05:59:39 +00:00
|
|
|
import random
|
2009-07-25 08:28:33 +00:00
|
|
|
import re
|
2009-08-25 04:50:19 +00:00
|
|
|
import shutil
|
2010-03-22 05:10:43 +00:00
|
|
|
import unicodedata
|
2009-07-25 08:28:33 +00:00
|
|
|
|
2009-07-21 07:12:25 +00:00
|
|
|
from sqlalchemy.sql import func
|
2009-07-25 08:28:33 +00:00
|
|
|
import whoosh
|
2009-08-17 04:03:49 +00:00
|
|
|
import whoosh.filedb.filestore
|
|
|
|
import whoosh.filedb.fileindex
|
|
|
|
import whoosh.index
|
2009-07-25 08:28:33 +00:00
|
|
|
from whoosh.qparser import QueryParser
|
2009-08-23 02:44:57 +00:00
|
|
|
import whoosh.scoring
|
2009-07-25 08:28:33 +00:00
|
|
|
import whoosh.spelling
|
2009-07-21 07:12:25 +00:00
|
|
|
|
2011-04-14 10:46:12 +00:00
|
|
|
from pokedex.compatibility import namedtuple
|
2010-04-25 07:14:29 +00:00
|
|
|
|
2009-08-19 06:42:59 +00:00
|
|
|
from pokedex.db import connect
|
2009-07-21 07:12:25 +00:00
|
|
|
import pokedex.db.tables as tables
|
2009-08-21 07:30:01 +00:00
|
|
|
from pokedex.roomaji import romanize
|
2010-05-13 17:33:07 +00:00
|
|
|
from pokedex.defaults import get_default_index_dir
|
2009-07-21 07:12:25 +00:00
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
__all__ = ['PokedexLookup']
|
2009-07-25 08:28:33 +00:00
|
|
|
|
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
rx_is_number = re.compile('^\d+$')
|
2009-08-19 06:42:59 +00:00
|
|
|
|
2010-08-25 03:06:40 +00:00
|
|
|
LookupResult = namedtuple('LookupResult', [
|
|
|
|
'object', 'indexed_name', 'name', 'language', 'iso639', 'iso3166', 'exact',
|
|
|
|
])
|
2009-07-25 08:28:33 +00:00
|
|
|
|
2010-05-13 05:36:12 +00:00
|
|
|
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."
|
|
|
|
)
|
|
|
|
|
2009-08-23 02:44:57 +00:00
|
|
|
class LanguageWeighting(whoosh.scoring.Weighting):
|
|
|
|
"""A scoring class that forces otherwise-equal English results to come
|
|
|
|
before foreign results.
|
|
|
|
"""
|
|
|
|
|
2010-07-01 06:04:39 +00:00
|
|
|
def __init__(self, extra_weights={}, *args, **kwargs):
|
|
|
|
"""`extra_weights` may be a dictionary of weights which will be
|
|
|
|
factored in.
|
|
|
|
|
|
|
|
Intended for use with spelling corrections, which come along with their
|
|
|
|
own weightings.
|
|
|
|
"""
|
|
|
|
self.extra_weights = extra_weights
|
|
|
|
super(LanguageWeighting, self).__init__(*args, **kwargs)
|
|
|
|
|
2009-08-23 02:44:57 +00:00
|
|
|
def score(self, searcher, fieldnum, text, docnum, weight, QTF=1):
|
|
|
|
doc = searcher.stored_fields(docnum)
|
2010-07-01 06:04:39 +00:00
|
|
|
|
|
|
|
# Apply extra weight
|
|
|
|
weight = weight * self.extra_weights.get(text, 1.0)
|
|
|
|
|
2010-09-13 03:51:51 +00:00
|
|
|
language = doc.get('language')
|
|
|
|
if language is None:
|
2009-08-23 02:44:57 +00:00
|
|
|
# English (well, "default"); leave it at 1
|
|
|
|
return weight
|
2010-09-13 03:51:51 +00:00
|
|
|
elif language == u'Roomaji':
|
2010-03-22 06:27:47 +00:00
|
|
|
# Give Roomaji a little boost; it's most likely to be searched
|
2010-07-01 06:04:39 +00:00
|
|
|
return weight * 0.9
|
2009-08-23 02:44:57 +00:00
|
|
|
else:
|
|
|
|
# Everything else can drop down the totem pole
|
2010-07-01 06:04:39 +00:00
|
|
|
return weight * 0.8
|
2009-08-23 02:44:57 +00:00
|
|
|
|
2009-08-23 01:56:02 +00:00
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
class PokedexLookup(object):
|
2010-08-25 03:15:12 +00:00
|
|
|
MAX_FUZZY_RESULTS = 10
|
|
|
|
MAX_EXACT_RESULTS = 43
|
|
|
|
INTERMEDIATE_FACTOR = 2
|
2010-03-22 06:27:47 +00:00
|
|
|
|
2010-07-01 06:04:39 +00:00
|
|
|
# The speller only checks how much the input matches a word; there can be
|
|
|
|
# all manner of extra unmatched junk, and it won't affect the weighting.
|
|
|
|
# To compensate, greatly boost the weighting of matches at the beginning
|
|
|
|
# and end, so nearly-full-word-matches are much better
|
|
|
|
SPELLER_OPTIONS = dict(booststart=10.0, boostend=9.0)
|
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
# Dictionary of table name => table class.
|
|
|
|
# Need the table name so we can get the class from the table name after we
|
|
|
|
# retrieve something from the index
|
|
|
|
indexed_tables = dict(
|
|
|
|
(cls.__tablename__, cls)
|
|
|
|
for cls in (
|
|
|
|
tables.Ability,
|
|
|
|
tables.Item,
|
2010-03-24 05:56:45 +00:00
|
|
|
tables.Location,
|
2010-03-22 06:27:47 +00:00
|
|
|
tables.Move,
|
2010-04-19 05:43:33 +00:00
|
|
|
tables.Nature,
|
2011-04-29 23:10:57 +00:00
|
|
|
tables.PokemonSpecies,
|
2010-11-29 23:51:50 +00:00
|
|
|
tables.PokemonForm,
|
2010-03-22 06:27:47 +00:00
|
|
|
tables.Type,
|
|
|
|
)
|
|
|
|
)
|
2009-08-19 06:42:59 +00:00
|
|
|
|
2009-07-25 08:28:33 +00:00
|
|
|
|
2010-05-13 05:36:12 +00:00
|
|
|
def __init__(self, directory=None, session=None):
|
2010-03-22 06:27:47 +00:00
|
|
|
"""Opens the whoosh index stored in the named directory. If the index
|
|
|
|
doesn't already exist, it will be created.
|
2009-08-23 23:27:13 +00:00
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
`directory`
|
|
|
|
Directory containing the index. Defaults to a location within the
|
|
|
|
`pokedex` egg directory.
|
2009-09-14 03:09:31 +00:00
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
`session`
|
2010-05-13 05:36:12 +00:00
|
|
|
Used for creating the index and retrieving objects. Defaults to an
|
|
|
|
attempt to connect to the default SQLite database installed by
|
|
|
|
`pokedex setup`.
|
2010-03-22 06:27:47 +00:00
|
|
|
"""
|
2009-07-25 08:28:33 +00:00
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
# By the time this returns, self.index, self.speller, and self.session
|
|
|
|
# must be set
|
2009-07-21 07:12:25 +00:00
|
|
|
|
2010-05-13 17:33:07 +00:00
|
|
|
# If a directory was not given, use the default
|
|
|
|
if directory is None:
|
|
|
|
directory = get_default_index_dir()
|
2010-05-13 06:23:05 +00:00
|
|
|
|
2010-05-13 05:36:12 +00:00
|
|
|
self.directory = directory
|
2009-08-25 05:59:39 +00:00
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
if session:
|
|
|
|
self.session = session
|
|
|
|
else:
|
|
|
|
self.session = connect()
|
|
|
|
|
|
|
|
# Attempt to open or create the index
|
2010-05-13 05:36:12 +00:00
|
|
|
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)
|
2010-07-01 06:04:39 +00:00
|
|
|
self.speller = whoosh.spelling.SpellChecker(spell_store,
|
|
|
|
**self.SPELLER_OPTIONS)
|
2010-05-13 05:36:12 +00:00
|
|
|
|
|
|
|
|
|
|
|
def rebuild_index(self):
|
|
|
|
"""Creates the index from scratch."""
|
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
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,
|
2010-08-25 03:06:40 +00:00
|
|
|
iso639=whoosh.fields.ID(stored=True),
|
|
|
|
iso3166=whoosh.fields.ID(stored=True),
|
2010-03-22 06:27:47 +00:00
|
|
|
display_name=whoosh.fields.STORED, # non-lowercased name
|
|
|
|
)
|
|
|
|
|
2010-09-13 03:51:51 +00:00
|
|
|
if os.path.exists(self.directory):
|
|
|
|
# create_in() isn't totally reliable, so just nuke whatever's there
|
|
|
|
# manually. Try to be careful about this...
|
|
|
|
for f in os.listdir(self.directory):
|
|
|
|
if re.match('^_?(MAIN|SPELL)_', f):
|
|
|
|
os.remove(os.path.join(self.directory, f))
|
|
|
|
else:
|
2010-05-13 05:36:12 +00:00
|
|
|
os.mkdir(self.directory)
|
|
|
|
|
|
|
|
self.index = whoosh.index.create_in(self.directory, schema=schema,
|
|
|
|
indexname='MAIN')
|
2010-03-22 06:27:47 +00:00
|
|
|
writer = self.index.writer()
|
|
|
|
|
|
|
|
# Index every name in all our tables of interest
|
2010-07-01 06:04:39 +00:00
|
|
|
speller_entries = set()
|
2010-03-22 06:27:47 +00:00
|
|
|
for cls in self.indexed_tables.values():
|
2011-04-29 23:10:57 +00:00
|
|
|
q = self.session.query(cls).order_by(cls.id)
|
2010-03-22 06:27:47 +00:00
|
|
|
|
|
|
|
for row in q.yield_per(5):
|
2010-03-24 01:09:03 +00:00
|
|
|
row_key = dict(table=unicode(cls.__tablename__),
|
|
|
|
row_id=unicode(row.id))
|
2010-03-22 06:27:47 +00:00
|
|
|
|
2010-08-25 03:06:40 +00:00
|
|
|
def add(name, language, iso639, iso3166):
|
2010-03-22 06:27:47 +00:00
|
|
|
normalized_name = self.normalize_name(name)
|
2010-03-24 01:09:03 +00:00
|
|
|
|
|
|
|
writer.add_document(
|
|
|
|
name=normalized_name, display_name=name,
|
2010-08-25 03:06:40 +00:00
|
|
|
language=language, iso639=iso639, iso3166=iso3166,
|
2010-03-24 01:09:03 +00:00
|
|
|
**row_key
|
|
|
|
)
|
2010-03-23 05:49:21 +00:00
|
|
|
|
2010-07-01 06:04:39 +00:00
|
|
|
speller_entries.add(normalized_name)
|
2010-03-22 06:27:47 +00:00
|
|
|
|
|
|
|
|
2011-04-29 23:10:57 +00:00
|
|
|
if cls == tables.PokemonForm:
|
|
|
|
name_map = 'pokemon_name_map'
|
|
|
|
else:
|
|
|
|
name_map = 'name_map'
|
|
|
|
|
|
|
|
seen = set([None])
|
|
|
|
for language, name in sorted(getattr(row, name_map, {}).items(),
|
|
|
|
# Sort English first for now
|
|
|
|
key=lambda (l, n): (l.identifier != 'en', not l.official)):
|
|
|
|
if not name:
|
2010-11-29 23:51:50 +00:00
|
|
|
continue
|
2011-02-04 04:25:23 +00:00
|
|
|
if name in seen:
|
|
|
|
# Don't add the name again as a different
|
2010-03-22 06:27:47 +00:00
|
|
|
# language; no point and it makes spell results
|
|
|
|
# confusing
|
|
|
|
continue
|
2011-02-04 04:25:23 +00:00
|
|
|
seen.add(name)
|
2010-03-22 06:27:47 +00:00
|
|
|
|
2011-02-04 04:25:23 +00:00
|
|
|
add(name, language.name,
|
|
|
|
language.iso639,
|
|
|
|
language.iso3166)
|
2010-03-22 06:27:47 +00:00
|
|
|
|
|
|
|
# Add Roomaji too
|
2011-02-04 04:25:23 +00:00
|
|
|
if language.identifier == 'ja':
|
|
|
|
roomaji = romanize(name)
|
2010-08-25 03:06:40 +00:00
|
|
|
add(roomaji, u'Roomaji', u'ja', u'jp')
|
2010-03-22 06:27:47 +00:00
|
|
|
|
|
|
|
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
|
2010-07-01 06:04:39 +00:00
|
|
|
self.speller = whoosh.spelling.SpellChecker(self.index.storage, mingram=2,
|
|
|
|
**self.SPELLER_OPTIONS)
|
|
|
|
self.speller.add_words(speller_entries)
|
2010-03-22 06:27:47 +00:00
|
|
|
|
|
|
|
|
|
|
|
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)
|
2009-09-29 03:59:53 +00:00
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
name = name.strip()
|
|
|
|
name = name.lower()
|
2009-09-29 03:59:53 +00:00
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
return name
|
2009-09-29 03:59:53 +00:00
|
|
|
|
|
|
|
|
2010-04-19 07:33:09 +00:00
|
|
|
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(',')
|
2010-09-13 04:08:50 +00:00
|
|
|
user_valid_types = []
|
|
|
|
for prefix in prefixes:
|
|
|
|
prefix = prefix.strip()
|
|
|
|
if prefix:
|
|
|
|
user_valid_types.append(prefix)
|
2011-04-29 23:10:57 +00:00
|
|
|
if prefix == 'pokemon':
|
|
|
|
# When the user says 'pokemon', they really meant both
|
|
|
|
# species & form.
|
|
|
|
user_valid_types.append('pokemon_species')
|
|
|
|
user_valid_types.append('pokemon_form')
|
2010-04-19 07:33:09 +00:00
|
|
|
|
|
|
|
# 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
|
2010-08-25 04:21:29 +00:00
|
|
|
# forbidden, completely ignore what the user requested.
|
|
|
|
# And, just to complicate matters: "type" and language need to be
|
|
|
|
# considered separately.
|
|
|
|
def merge_requirements(func):
|
|
|
|
user = filter(func, user_valid_types)
|
|
|
|
system = filter(func, valid_types)
|
|
|
|
|
|
|
|
if user and system:
|
|
|
|
merged = list(set(user) & set(system))
|
|
|
|
if merged:
|
|
|
|
return merged
|
|
|
|
else:
|
|
|
|
# No overlap; use the system restrictions
|
|
|
|
return system
|
|
|
|
else:
|
|
|
|
# One or the other is blank; use the one that's not
|
|
|
|
return user or system
|
2010-04-19 07:33:09 +00:00
|
|
|
|
2010-08-25 04:21:29 +00:00
|
|
|
# @foo means language must be foo; otherwise it's a table name
|
|
|
|
lang_requirements = merge_requirements(lambda req: req[0] == u'@')
|
|
|
|
type_requirements = merge_requirements(lambda req: req[0] != u'@')
|
|
|
|
all_requirements = lang_requirements + type_requirements
|
2010-04-19 07:33:09 +00:00
|
|
|
|
|
|
|
# Construct the term
|
2010-08-25 03:06:40 +00:00
|
|
|
lang_terms = []
|
2010-08-25 04:21:29 +00:00
|
|
|
for lang in lang_requirements:
|
|
|
|
# Allow for either country or language codes
|
|
|
|
lang_code = lang[1:]
|
|
|
|
lang_terms.append(whoosh.query.Term(u'iso639', lang_code))
|
|
|
|
lang_terms.append(whoosh.query.Term(u'iso3166', lang_code))
|
|
|
|
|
|
|
|
type_terms = []
|
|
|
|
for type in type_requirements:
|
|
|
|
table_name = self._parse_table_name(type)
|
2010-08-25 03:06:40 +00:00
|
|
|
|
2010-08-25 04:21:29 +00:00
|
|
|
# Quietly ignore bogus valid_types; more likely to DTRT
|
|
|
|
if table_name:
|
|
|
|
type_terms.append(whoosh.query.Term(u'table', table_name))
|
2010-08-25 03:06:40 +00:00
|
|
|
|
|
|
|
# Combine both kinds of restriction
|
|
|
|
all_terms = []
|
|
|
|
if type_terms:
|
|
|
|
all_terms.append(whoosh.query.Or(type_terms))
|
|
|
|
if lang_terms:
|
|
|
|
all_terms.append(whoosh.query.Or(lang_terms))
|
|
|
|
|
2010-08-25 04:21:29 +00:00
|
|
|
return name, all_requirements, whoosh.query.And(all_terms)
|
2010-04-19 07:33:09 +00:00
|
|
|
|
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
def _parse_table_name(self, name):
|
|
|
|
"""Takes a singular table name, table name, or table object and returns
|
|
|
|
the table name.
|
2009-09-29 03:59:53 +00:00
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
Returns None for a bogus name.
|
|
|
|
"""
|
2010-04-30 07:22:03 +00:00
|
|
|
# Table object
|
2010-03-22 06:27:47 +00:00
|
|
|
if hasattr(name, '__tablename__'):
|
|
|
|
return getattr(name, '__tablename__')
|
2010-04-30 07:22:03 +00:00
|
|
|
|
|
|
|
# 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
|
2010-03-22 06:27:47 +00:00
|
|
|
|
|
|
|
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,
|
2010-03-24 01:09:03 +00:00
|
|
|
indexed_name=record['name'],
|
2010-03-22 06:27:47 +00:00
|
|
|
name=record['display_name'],
|
2010-09-13 03:51:51 +00:00
|
|
|
language=record.get('language'),
|
2010-08-25 03:06:40 +00:00
|
|
|
iso639=record['iso639'],
|
2010-03-22 06:27:47 +00:00
|
|
|
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.
|
|
|
|
|
2010-08-25 03:06:40 +00:00
|
|
|
Returns a list of named (object, name, language, iso639, iso3166,
|
|
|
|
exact) tuples. `object` is a database object, `name` is the name under
|
|
|
|
which the object was found, `language` and the two isos are the name
|
|
|
|
and country codes of the language in which the name was found, and
|
|
|
|
`exact` is True iff this was an exact match.
|
2010-03-22 06:27:47 +00:00
|
|
|
|
|
|
|
This function currently ONLY does fuzzy matching if there are no exact
|
|
|
|
matches.
|
|
|
|
|
|
|
|
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
|
2010-08-25 03:06:40 +00:00
|
|
|
with commas, as "move,item:1".
|
|
|
|
- Language restrictions. "@fr:charge" will only return Tackle, which
|
|
|
|
is called "Charge" in French. These can be combined with type
|
|
|
|
restrictions, e.g., "@fr,move:charge".
|
2010-03-22 06:27:47 +00:00
|
|
|
|
|
|
|
`input`
|
|
|
|
Name of the thing to look for.
|
|
|
|
|
|
|
|
`valid_types`
|
2010-08-25 03:06:40 +00:00
|
|
|
A list of type or language restrictions, e.g., `['pokemon',
|
|
|
|
'@ja']`. If this is provided, only results in one of the given
|
|
|
|
tables will be returned.
|
2010-03-22 06:27:47 +00:00
|
|
|
|
|
|
|
`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
|
|
|
|
|
2010-04-19 07:33:09 +00:00
|
|
|
# Pop off any type prefix and merge with valid_types
|
|
|
|
name, merged_valid_types, type_term = \
|
|
|
|
self._apply_valid_types(name, valid_types)
|
2010-03-22 06:27:47 +00:00
|
|
|
|
|
|
|
# Random lookup
|
|
|
|
if name == 'random':
|
2010-04-19 07:33:09 +00:00
|
|
|
return self.random_lookup(valid_types=merged_valid_types)
|
2010-03-22 06:27:47 +00:00
|
|
|
|
|
|
|
# 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
|
2010-04-28 17:26:27 +00:00
|
|
|
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
|
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
if '*' in name or '?' in name:
|
|
|
|
exact_only = True
|
|
|
|
query = whoosh.query.Wildcard(u'name', name)
|
2010-04-28 17:26:27 +00:00
|
|
|
elif name_as_number is not None:
|
2010-03-22 06:27:47 +00:00
|
|
|
# Don't spell-check numbers!
|
|
|
|
exact_only = True
|
2010-04-28 17:26:27 +00:00
|
|
|
query = whoosh.query.Term(u'row_id', unicode(name_as_number))
|
2010-03-22 06:27:47 +00:00
|
|
|
else:
|
|
|
|
# Not an integer
|
2010-03-24 02:43:42 +00:00
|
|
|
query = whoosh.query.Term(u'name', name)
|
2010-03-22 06:27:47 +00:00
|
|
|
|
2010-04-19 07:33:09 +00:00
|
|
|
if type_term:
|
|
|
|
query = query & type_term
|
2010-03-22 06:27:47 +00:00
|
|
|
|
|
|
|
|
|
|
|
### Actual searching
|
2010-08-25 03:15:12 +00:00
|
|
|
# Limits; result limits are constants, and intermediate results (before
|
|
|
|
# duplicate items are stripped out) are capped at the result limit
|
|
|
|
# times another constant.
|
|
|
|
# Fuzzy are capped at 10, beyond which something is probably very
|
|
|
|
# wrong. Exact matches -- that is, wildcards and ids -- are far less
|
|
|
|
# constrained.
|
|
|
|
# Also, exact matches are sorted by name, since weight doesn't matter.
|
|
|
|
sort_by = dict()
|
|
|
|
if exact_only:
|
|
|
|
max_results = self.MAX_EXACT_RESULTS
|
|
|
|
sort_by['sortedby'] = (u'table', u'name')
|
|
|
|
else:
|
|
|
|
max_results = self.MAX_FUZZY_RESULTS
|
|
|
|
|
|
|
|
searcher = self.index.searcher(weighting=LanguageWeighting())
|
|
|
|
results = searcher.search(
|
|
|
|
query,
|
|
|
|
limit=int(max_results * self.INTERMEDIATE_FACTOR),
|
|
|
|
**sort_by
|
|
|
|
)
|
2010-03-22 06:27:47 +00:00
|
|
|
|
|
|
|
# Look for some fuzzy matches if necessary
|
|
|
|
if not exact_only and not results:
|
|
|
|
exact = False
|
|
|
|
results = []
|
|
|
|
|
2010-07-01 06:04:39 +00:00
|
|
|
fuzzy_query_parts = []
|
|
|
|
fuzzy_weights = {}
|
|
|
|
min_weight = [None]
|
|
|
|
for suggestion, _, weight in self.speller.suggestions_and_scores(name):
|
|
|
|
# Only allow the top 50% of scores; otherwise there will always
|
|
|
|
# be a lot of trailing junk
|
|
|
|
if min_weight[0] is None:
|
|
|
|
min_weight[0] = weight * 0.5
|
|
|
|
elif weight < min_weight[0]:
|
|
|
|
break
|
|
|
|
|
|
|
|
fuzzy_query_parts.append(whoosh.query.Term('name', suggestion))
|
|
|
|
fuzzy_weights[suggestion] = weight
|
|
|
|
|
|
|
|
if not fuzzy_query_parts:
|
|
|
|
# Nothing at all; don't try querying
|
|
|
|
return []
|
|
|
|
|
|
|
|
fuzzy_query = whoosh.query.Or(fuzzy_query_parts)
|
|
|
|
if type_term:
|
|
|
|
fuzzy_query = fuzzy_query & type_term
|
|
|
|
|
|
|
|
searcher.weighting = LanguageWeighting(extra_weights=fuzzy_weights)
|
|
|
|
results = searcher.search(fuzzy_query)
|
2010-03-22 06:27:47 +00:00
|
|
|
|
|
|
|
### Convert results to db objects
|
|
|
|
objects = self._whoosh_records_to_results(results, exact=exact)
|
|
|
|
|
2010-08-25 03:15:12 +00:00
|
|
|
# Truncate and return
|
|
|
|
return objects[:max_results]
|
2010-03-22 06:27:47 +00:00
|
|
|
|
|
|
|
|
|
|
|
def random_lookup(self, valid_types=[]):
|
|
|
|
"""Returns a random lookup result from one of the provided
|
|
|
|
`valid_types`.
|
|
|
|
"""
|
|
|
|
|
2010-08-25 03:06:40 +00:00
|
|
|
table_names = []
|
2010-03-22 06:27:47 +00:00
|
|
|
for valid_type in valid_types:
|
|
|
|
table_name = self._parse_table_name(valid_type)
|
2010-11-29 23:51:50 +00:00
|
|
|
# Skip anything not recognized. Could be, say, a language code.
|
|
|
|
# XXX The vast majority of Pokémon forms are unnamed and unindexed,
|
|
|
|
# which can produce blank results. So skip them too for now.
|
|
|
|
if table_name and table_name != 'pokemon_forms':
|
2010-08-25 03:06:40 +00:00
|
|
|
table_names.append(table_name)
|
2010-03-22 06:27:47 +00:00
|
|
|
|
2010-08-25 03:06:40 +00:00
|
|
|
if not table_names:
|
2010-03-22 06:27:47 +00:00
|
|
|
# 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
|
2010-08-25 03:06:40 +00:00
|
|
|
# *something*, so it politely selects from the entire index instead
|
|
|
|
table_names = self.indexed_tables.keys()
|
2010-11-29 23:51:50 +00:00
|
|
|
table_names.remove('pokemon_forms')
|
2010-03-22 06:27:47 +00:00
|
|
|
|
2011-04-03 08:56:27 +00:00
|
|
|
# Pick a random table, then pick a random item from it. Small tables
|
|
|
|
# like Type will have an unnatural bias. The alternative is that a
|
|
|
|
# simple search for "random" will do some eight queries, counting the
|
|
|
|
# rows in every single indexed table, and that's awful.
|
|
|
|
# XXX Can we improve on this, reasonably?
|
|
|
|
table_name = random.choice(table_names)
|
|
|
|
count = self.session.query(self.indexed_tables[table_name]).count()
|
|
|
|
id, = self.session.query(self.indexed_tables[table_name].id) \
|
|
|
|
.offset(random.randint(0, count - 1)) \
|
|
|
|
.first()
|
|
|
|
|
|
|
|
return self.lookup(unicode(id), valid_types=[table_name])
|
2010-03-22 06:27:47 +00:00
|
|
|
|
2010-04-19 07:33:09 +00:00
|
|
|
def prefix_lookup(self, prefix, valid_types=[]):
|
2010-03-22 06:27:47 +00:00
|
|
|
"""Returns terms starting with the given exact prefix.
|
|
|
|
|
2010-04-19 07:33:09 +00:00
|
|
|
Type prefixes are recognized, but no other name munging is done.
|
2010-03-22 06:27:47 +00:00
|
|
|
"""
|
|
|
|
|
2010-04-19 07:33:09 +00:00
|
|
|
# Pop off any type prefix and merge with valid_types
|
|
|
|
prefix, merged_valid_types, type_term = \
|
|
|
|
self._apply_valid_types(prefix, valid_types)
|
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
query = whoosh.query.Prefix(u'name', self.normalize_name(prefix))
|
|
|
|
|
2010-04-19 07:33:09 +00:00
|
|
|
if type_term:
|
|
|
|
query = query & type_term
|
|
|
|
|
2010-03-22 06:27:47 +00:00
|
|
|
searcher = self.index.searcher()
|
|
|
|
searcher.weighting = LanguageWeighting()
|
|
|
|
results = searcher.search(query) # XXX , limit=self.MAX_LOOKUP_RESULTS)
|
|
|
|
|
|
|
|
return self._whoosh_records_to_results(results)
|