X-Git-Url: http://git.veekun.com/zzz-pokedex.git/blobdiff_plain/4fe48baffb5156027c6d186308998412554a6906..b7268b5c97e6c0f5c2b92e5befb53d1a251e42b3:/pokedex/lookup.py

diff --git a/pokedex/lookup.py b/pokedex/lookup.py
index dcc3e7a..55e1b5a 100644
--- a/pokedex/lookup.py
+++ b/pokedex/lookup.py
@@ -1,11 +1,17 @@
 # encoding: utf8
+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
 
 # Dictionary of table name => table class.
@@ -13,33 +19,85 @@ import pokedex.db.tables as tables
 # retrieve something from the index
 indexed_tables = {}
 for cls in [
+        tables.Ability,
+        tables.Item,
+        tables.Move,
         tables.Pokemon,
+        tables.Type,
     ]:
     indexed_tables[cls.__tablename__] = cls
 
-index_bits = {}
-def get_index(session):
-    """Returns (index, speller).
-
-    Creates an index if one does not exist.
+# Dictionary of extra keys to file types of objects under, e.g. Pokémon can
+# also be looked up purely by number
+extra_keys = {
+    tables.Move: [
+        lambda row: u"move %d" % row.id,
+    ],
+    tables.Pokemon: [
+        lambda row: unicode(row.id),
+    ],
+}
+
+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.
+
+    `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='pokedex')
+            speller = whoosh.index.open_dir(directory, indexname='spelling')
+            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),
-        spelling_name=whoosh.fields.ID(stored=True),
         table=whoosh.fields.STORED,
         row_id=whoosh.fields.STORED,
-        language_id=whoosh.fields.STORED,
+        language=whoosh.fields.STORED,
+
+        # Whoosh 0.2 explodes when using a file-stored schema with no TEXT
+        # columns.  Appease it
+        dummy=whoosh.fields.TEXT,
     )
 
-    index = whoosh.index.Index(store, schema=schema, create=True)
+    index = whoosh.index.create_in(directory, schema=schema,
+                                              indexname='pokedex')
     writer = index.writer()
 
     # Index every name in all our tables of interest
+    speller_entries = []
     for cls in indexed_tables.values():
         q = session.query(cls)
 
@@ -48,31 +106,61 @@ def get_index(session):
             q = q.filter_by(forme_base_pokemon_id=None)
 
         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()
-            spelling_name = re.sub('[^a-z]', '', name)
-            writer.add_document(name=name,
-                                spelling_name=spelling_name,
-                                table=cls.__tablename__,
-                                row_id=row.id)
+            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)
 
     writer.commit()
 
-    ### Construct a spell-checker index
-    speller = whoosh.spelling.SpellChecker(index.storage)
+    # XXX GIHWEGREHKG
+    old__schema = whoosh.spelling.SpellChecker._schema
+    def new__schema(self):
+        schema = old__schema(self)
+        schema.add('dummy', whoosh.fields.TEXT)
+        return schema
+    whoosh.spelling.SpellChecker._schema = new__schema
+
+    # 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, indexname='spelling')
+    # 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 = speller.index(create=True).writer()
+    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
 
-    # Can't use speller.add_field because it tries to intuit a frequency, and
-    # names are in an ID field, which seems to be immune to frequency.
-    # Not hard to add everything ourselves, though
-    reader = index.doc_reader()
-    speller.add_words([ _['spelling_name'] for _ in reader ])
-    reader.close()
+    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):
+def lookup(name, session=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
@@ -84,39 +172,43 @@ def lookup(session, name, exact_only=False):
 
     Currently 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.
+
+    `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.
     """
 
+    if not session:
+        session = connect()
+
+    index, speller = open_index()
+
     exact = True
 
-    # Alas!  We have to make three attempts to find anything with this index.
-    # First: Try an exact match for a name in the index.
-    # Second: Try an exact match for a stripped-down name in the index.
-    # Third: Get spelling suggestions.
-    # The spelling module apparently only indexes *words* -- that is, [a-z]+.
-    # So we have a separate field that contains the same name, stripped down to
-    # just [a-z]+.
-    # Unfortunately, exact matches aren't returned as spelling suggestions, so
-    # we also have to do a regular index match against this separate field.
-    # Otherwise, 'nidoran' will never match anything
-    index, speller = get_index(session)
-
-    # Look for exact name
-    parser = QueryParser('name', schema=index.schema)
-    results = index.find(name.lower(), parser=parser)
+    # 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.lower())
+    results = searcher.search(query)
 
     if not exact_only:
-        # Look for a match with a reduced a-z name
-        if not results:
-            parser = QueryParser('spelling_name', schema=index.schema)
-            results = index.find(name.lower(), parser=parser)
-
         # Look for some fuzzy matches
         if not results:
-            results = []
             exact = False
+            results = []
 
             for suggestion in speller.suggest(name, 3):
-                results.extend( index.find(suggestion, parser=parser) )
+                query = whoosh.query.Term('name', suggestion)
+                results.extend(searcher.search(query))
 
     # Convert results to db objects
     objects = []