+ 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,
+ indexed_name=record['name'],
+ name=record['display_name'],
+ language=record['language'],
+ iso639=record['iso639'],
+ 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, 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.
+
+ 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".
+ - 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".
+ - Alternate formes can be specified merely like "wash rotom".
+
+ `input`
+ Name of the thing to look for.
+
+ `valid_types`
+ 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.
+
+ `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
+
+ # 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=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 name_as_number is not None:
+ # Don't spell-check numbers!
+ exact_only = True
+ query = whoosh.query.Term(u'row_id', unicode(name_as_number))
+ else:
+ # Not an integer
+ query = whoosh.query.Term(u'name', name)
+
+ if type_term:
+ query = query & type_term
+
+
+ ### Actual searching
+ # 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
+ )
+
+ # Look for some fuzzy matches if necessary
+ if not exact_only and not results:
+ exact = False
+ results = []
+
+ 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)
+
+ ### Convert results to db objects
+ objects = self._whoosh_records_to_results(results, exact=exact)
+
+ # Truncate and return
+ return objects[:max_results]
+
+
+ def random_lookup(self, valid_types=[]):
+ """Returns a random lookup result from one of the provided
+ `valid_types`.
+ """
+
+ table_names = []
+ for valid_type in valid_types:
+ table_name = self._parse_table_name(valid_type)
+ # Skip anything not recognized. Could be, say, a language code
+ if table_name:
+ table_names.append(table_name)
+
+ if not table_names:
+ # 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 instead
+ table_names = self.indexed_tables.keys()
+
+ # 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_name in table_names:
+ count = self.session.query(self.indexed_tables[table_name]).count()
+ total += count
+ partitions.append((table_name, 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, valid_types=[]):
+ """Returns terms starting with the given exact prefix.
+
+ 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)
+
+ return self._whoosh_records_to_results(results)
projects / zzz-pokedex.git / blobdiff