flask_whooshalchemy.py

'''

    whooshalchemy flask extension
    ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

    Adds whoosh indexing capabilities to SQLAlchemy models for Flask
    applications.

    :copyright: (c) 2012 by Karl Gyllstrom
    :license: BSD (see LICENSE.txt)

'''

from __future__ import with_statement
from __future__ import absolute_import


import flask.ext.sqlalchemy as flask_sqlalchemy

import sqlalchemy

from whoosh.qparser import OrGroup
from whoosh.qparser import AndGroup
from whoosh.qparser import MultifieldParser
from whoosh.analysis import StemmingAnalyzer
import whoosh.index
from whoosh.fields import Schema
#from whoosh.fields import ID, TEXT, KEYWORD, STORED

import heapq
import os


__searchable__ = '__searchable__'


DEFAULT_WHOOSH_INDEX_NAME = 'whoosh_index'


class _QueryProxy(flask_sqlalchemy.BaseQuery):
    # We're replacing the model's ``query`` field with this proxy. The main
    # thing this proxy does is override the __iter__ method so that results are
    # returned in the order of the whoosh score to reflect text-based ranking.

    def __init__(self, entities, session=None):
        super(_QueryProxy, self).__init__(entities, session)

        self._modelclass = self._mapper_zero().class_
        self._primary_key_name = self._modelclass.whoosh_primary_key
        self._whoosh_searcher = self._modelclass.pure_whoosh

        # Stores whoosh results from query. If ``None``, indicates that no
        # whoosh query was performed.
        self._whoosh_rank = None

    def __iter__(self):
        ''' Reorder ORM-db results according to Whoosh relevance score. '''

        super_iter = super(_QueryProxy, self).__iter__()

        if self._whoosh_rank is None:
            # Whoosh search hasn't been run so behave as normal.

            return super_iter

        # Iterate through the values and re-order by whoosh relevance.
        ordered_by_whoosh_rank = []

        for row in super_iter:
            # Push items onto heap, where sort value is the rank provided by
            # Whoosh

            heapq.heappush(ordered_by_whoosh_rank,
                (self._whoosh_rank[unicode(getattr(row,
                    self._primary_key_name))], row))

        def _inner():
            while ordered_by_whoosh_rank:
                yield heapq.heappop(ordered_by_whoosh_rank)[1]

        return _inner()

    def whoosh_search(self, query, limit=None, fields=None, or_=False):
        '''

        Execute text query on database. Results have a text-based
        match to the query, ranked by the scores from the underlying Whoosh
        index.

        By default, the search is executed on all of the indexed fields as an
        OR conjunction. For example, if a model has 'title' and 'content'
        indicated as ``__searchable__``, a query will be checked against both
        fields, returning any instance whose title or content are a content
        match for the query. To specify particular fields to be checked,
        populate the ``fields`` parameter with the desired fields.

        By default, results will only be returned if they contain all of the
        query terms (AND). To switch to an OR grouping, set the ``or_``
        parameter to ``True``.

        '''
            
        if not isinstance(query, unicode):
            query = unicode(query)

        results = self._whoosh_searcher(query, limit, fields, or_)

        if not results:
            # We don't want to proceed with empty results because we get a
            # stderr warning from sqlalchemy when executing 'in_' on empty set.
            # However we cannot just return an empty list because it will not
            # be a query.

            # XXX is this efficient?
            return self.filter('null')

        result_set = set()
        result_ranks = {}

        for rank, result in enumerate(results):
            pk = result[self._primary_key_name]
            result_set.add(pk)
            result_ranks[pk] = rank

        f = self.filter(getattr(self._modelclass,
            self._primary_key_name).in_(result_set))

        f._whoosh_rank = result_ranks

        return f


class _Searcher(object):
    ''' Assigned to a Model class as ``pure_search``, which enables
    text-querying to whoosh hit list. Also used by ``query.whoosh_search``'''

    def __init__(self, primary, indx):
        self.primary_key_name = primary
        self._index = indx
        self.searcher = indx.searcher()
        self._all_fields = list(set(indx.schema._fields.keys()) -
                set([self.primary_key_name]))

    def __call__(self, query, limit=None, fields=None, or_=False):
        if fields is None:
            fields = self._all_fields

        group = OrGroup if or_ else AndGroup
        parser = MultifieldParser(fields, self._index.schema, group=group)
        return self._index.searcher().search(parser.parse(query),
                limit=limit)


def whoosh_index(app, model):
    ''' Create whoosh index for ``model``, if one does not exist. If 
    the index exists it is opened and cached. '''

    # gets the whoosh index for this model, creating one if it does not exist.
    # A dict of model -> whoosh index is added to the ``app`` variable.

    if not hasattr(app, 'whoosh_indexes'):
        app.whoosh_indexes = {}

    return app.whoosh_indexes.get(model.__name__,
                _create_index(app, model))

def _get_analyzer(app, model):
    analyzer = getattr(model, '__analyzer__', None)

    if not analyzer and app.config.get('WHOOSH_ANALYZER'):
        analyzer = app.config['WHOOSH_ANALYZER']

    if not analyzer:
        analyzer = StemmingAnalyzer()

    return analyzer

def _create_index(app, model):
    # a schema is created based on the fields of the model. Currently we only
    # support primary key -> whoosh.ID, and sqlalchemy.(String, Unicode, Text)
    # -> whoosh.TEXT.

    if not app.config.get('WHOOSH_BASE'):
        # XXX todo: is there a better approach to handle the absenSe of a
        # config value for whoosh base? Should we throw an exception? If
        # so, this exception will be thrown in the after_commit function,
        # which is probably not ideal.

        app.config['WHOOSH_BASE'] = DEFAULT_WHOOSH_INDEX_NAME

    # we index per model.
    wi = os.path.join(app.config.get('WHOOSH_BASE'),
            model.__name__)

    analyzer = _get_analyzer(app, model)
    schema, primary_key = _get_whoosh_schema_and_primary_key(model, analyzer)

    if whoosh.index.exists_in(wi):
        indx = whoosh.index.open_dir(wi)
    else:
        if not os.path.exists(wi):
            os.makedirs(wi)
        indx = whoosh.index.create_in(wi, schema)

    app.whoosh_indexes[model.__name__] = indx

    model.pure_whoosh = _Searcher(primary_key, indx)
    model.whoosh_primary_key = primary_key

    # change the query class of this model to our own
    model.query_class = _QueryProxy
    
    return indx


def _get_whoosh_schema_and_primary_key(model, analyzer):
    schema = {}
    primary = None
    searchable = set(model.__searchable__)

    for field in model.__table__.columns:
        if field.primary_key:
            schema[field.name] = whoosh.fields.ID(stored=True, unique=True)
            primary = field.name

        if field.name in searchable and isinstance(field.type,
                (sqlalchemy.types.Text, sqlalchemy.types.String,
                    sqlalchemy.types.Unicode)):

            schema[field.name] = whoosh.fields.TEXT(analyzer=analyzer)

    return Schema(**schema), primary


def _after_flush(app, changes):
    # Any db updates go through here. We check if any of these models have
    # ``__searchable__`` fields, indicating they need to be indexed. With these
    # we update the whoosh index for the model. If no index exists, it will be
    # created here; this could impose a penalty on the initial commit of a
    # model.

    bytype = {}  # sort changes by type so we can use per-model writer
    for change in changes:
        update = change[1] in ('update', 'insert')

        if hasattr(change[0].__class__, __searchable__):
            bytype.setdefault(change[0].__class__.__name__, []).append((update,
                change[0]))

    for model, values in bytype.iteritems():
        index = whoosh_index(app, values[0][1].__class__)
        with index.writer() as writer:
            primary_field = values[0][1].pure_whoosh.primary_key_name
            searchable = values[0][1].__searchable__

            for update, v in values:
                if update:
                    attrs = {}
                    for key in searchable:
                        try:
                            attrs[key] = unicode(getattr(v, key))
                        except AttributeError:
                            raise AttributeError('{0} does not have {1} field {2}'
                                    .format(model, __searchable__, key))

                    attrs[primary_field] = unicode(getattr(v, primary_field))
                    writer.update_document(**attrs)
                else:
                    writer.delete_by_term(primary_field, unicode(getattr(v,
                        primary_field)))


flask_sqlalchemy.models_committed.connect(_after_flush)