# encoding: utf8
+from collections import namedtuple
+import os, os.path
+import pkg_resources
import re
from sqlalchemy.sql import func
import whoosh
+import whoosh.filedb.filestore
+import whoosh.filedb.fileindex
+import whoosh.index
from whoosh.qparser import QueryParser
import whoosh.spelling
+from pokedex.db import connect
import pokedex.db.tables as tables
+from pokedex.roomaji import romanize
# Dictionary of table name => table class.
# Need the table name so we can get the class from the table name after we
],
}
-index_bits = {}
-def get_index(session):
- """Returns (index, speller).
+def open_index(directory=None, session=None, recreate=False):
+ """Opens the whoosh index stored in the named directory and returns (index,
+ speller). If the index doesn't already exist, it will be created.
- Creates an index if one does not exist.
+ `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.
"""
- if index_bits:
- return index_bits['index'], index_bits['speller']
+ # Defaults
+ if not directory:
+ directory = pkg_resources.resource_filename('pokedex',
+ 'data/whoosh_index')
+
+ if not session:
+ 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!
+ try:
+ index = whoosh.index.open_dir(directory, indexname='MAIN')
+ spell_store = whoosh.filedb.filestore.FileStorage(directory)
+ speller = whoosh.spelling.SpellChecker(spell_store)
+ return index, speller
+ except whoosh.index.EmptyIndexError as e:
+ # Apparently not a real index. Fall out of the if and create it
+ pass
+
+ if not directory_exists:
+ os.mkdir(directory)
- store = whoosh.store.RamStorage()
+
+ # Create index
schema = whoosh.fields.Schema(
name=whoosh.fields.ID(stored=True),
table=whoosh.fields.STORED,
row_id=whoosh.fields.STORED,
- language_id=whoosh.fields.STORED,
+ language=whoosh.fields.STORED,
)
- # Construct a straight lookup index
- index = whoosh.index.Index(store, schema=schema, create=True)
+ index = whoosh.index.create_in(directory, schema=schema, indexname='MAIN')
writer = index.writer()
# Index every name in all our tables of interest
for row in q.yield_per(5):
row_key = dict(table=cls.__tablename__, row_id=row.id)
- # Spelling index only indexes strings of letters, alas, so we
- # reduce every name to this to make the index work. However, exact
- # matches are not returned, so e.g. 'nidoran' would neither match
- # exactly nor fuzzy-match. Solution: add the spelling-munged name
- # as a regular index row too.
name = row.name.lower()
writer.add_document(name=name, **row_key)
-
speller_entries.append(name)
for extra_key_func in extra_keys.get(cls, []):
extra_key = extra_key_func(row)
writer.add_document(name=extra_key, **row_key)
+ # Pokemon also get other languages
+ if cls == tables.Pokemon:
+ for foreign_name in row.foreign_names:
+ name = foreign_name.name.lower()
+ writer.add_document(name=name,
+ language=foreign_name.language.name,
+ **row_key)
+ speller_entries.append(name)
+
+ if foreign_name.language.name == 'Japanese':
+ # Add Roomaji too
+ roomaji = romanize(foreign_name.name).lower()
+ writer.add_document(name=roomaji,
+ language='Roomaji',
+ **row_key)
+ speller_entries.append(roomaji)
+
+
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
speller = whoosh.spelling.SpellChecker(index.storage)
- # WARNING: HERE BE DRAGONS
- # whoosh.spelling refuses to index things that don't look like words.
- # Unfortunately, this doesn't work so well for Pokémon (Mr. Mime,
- # Porygon-Z, etc.), and attempts to work around it lead to further
- # complications.
- # The below is copied from SpellChecker.add_scored_words without the check
- # for isalpha(). XXX get whoosh patched to make this unnecessary!
- writer = whoosh.writing.IndexWriter(speller.index())
- for word in speller_entries:
- fields = {"word": word, "score": 1}
- for size in xrange(speller.mingram, speller.maxgram + 1):
- nga = whoosh.analysis.NgramAnalyzer(size)
- gramlist = [t.text for t in nga(word)]
- if len(gramlist) > 0:
- fields["start%s" % size] = gramlist[0]
- fields["end%s" % size] = gramlist[-1]
- fields["gram%s" % size] = " ".join(gramlist)
- writer.add_document(**fields)
- writer.commit()
- # end copy-pasta
+ speller.add_words(speller_entries)
+
+ return index, speller
- index_bits['index'] = index
- index_bits['speller'] = speller
- index_bits['store'] = store
- return index_bits['index'], index_bits['speller']
-def lookup(session, name, exact_only=False):
+LookupResult = namedtuple('LookupResult', ['object', 'language', 'exact'])
+def lookup(name, session=None, indices=None, exact_only=False):
"""Attempts to find some sort of object, given a database session and name.
- Returns (objects, exact) where `objects` is a list of database objects, and
- `exact` is True iff the given name matched the returned objects exactly.
+ Returns a list of named (object, language, exact) tuples. `object` is a
+ database object, `language` is the name of the language in which the name
+ was found, and `exact` is True iff this was an exact match.
- This function ONLY does fuzzy matching if there are no exact matches.
+ This function currently ONLY does fuzzy matching if there are no exact
+ matches.
Formes are not returned; "Shaymin" will return only grass Shaymin.
- Currently recognizes:
+ Recognizes:
- Pokémon names: "Eevee"
+
+ `name`
+ Name of the thing to look for.
+
+ `session`
+ A database session to use for retrieving objects. As with get_index,
+ if this is not provided, a connection to the default database will be
+ attempted.
+
+ `indices`
+ Tuple of index, speller as returned from `open_index()`. Defaults to
+ a call to `open_index()`.
+
+ `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.
"""
- exact = True
+ if not session:
+ session = connect()
+
+ if indices:
+ index, speller = indices
+ else:
+ index, speller = open_index()
- index, speller = get_index(session)
+ name = unicode(name)
+
+ exact = True
# Look for exact name. A Term object does an exact match, so we don't have
# to worry about a query parser tripping on weird characters in the input
searcher = index.searcher()
- query = whoosh.query.Term('name', name)
+ query = whoosh.query.Term('name', name.lower())
results = searcher.search(query)
- if not exact_only:
- # Look for some fuzzy matches
- if not results:
- exact = False
- results = []
+ # Look for some fuzzy matches if necessary
+ if not exact_only and not results:
+ exact = False
+ results = []
- for suggestion in speller.suggest(name, 3):
- query = whoosh.query.Term('name', suggestion)
- results.extend(searcher.search(query))
+ for suggestion in speller.suggest(name, 10):
+ query = whoosh.query.Term('name', suggestion)
+ results.extend(searcher.search(query))
- # Convert results to db objects
+ ### Convert results to db objects
objects = []
seen = {}
for result in results:
cls = indexed_tables[result['table']]
obj = session.query(cls).get(result['row_id'])
- objects.append(obj)
+ objects.append(LookupResult(obj, result['language'], exact))
- return objects, exact
+ return objects
projects / zzz-pokedex.git / blobdiff