import json
import re
import sys
import warnings

from peewee import *
from peewee import ColumnBase
from peewee import EnclosedNodeList
from peewee import Entity
from peewee import Expression
from peewee import Insert
from peewee import Node
from peewee import NodeList
from peewee import OP
from peewee import VirtualField
from peewee import merge_dict
from peewee import sqlite3
from playhouse.sqlite_udf import JSON
from playhouse.sqlite_udf import RANK
from playhouse.sqlite_udf import register_udf_groups


if sys.version_info[0] == 3:
    basestring = str


FTS3_MATCHINFO = 'pcx'
FTS4_MATCHINFO = 'pcnalx'
if sqlite3 is not None:
    FTS_VERSION = 4 if sqlite3.sqlite_version_info[:3] >= (3, 7, 4) else 3
else:
    FTS_VERSION = 3

FTS5_MIN_SQLITE_VERSION = (3, 9, 0)


class RowIDField(AutoField):
    auto_increment = True
    column_name = name = required_name = 'rowid'

    def bind(self, model, name, *args):
        if name != self.required_name:
            raise ValueError('%s must be named "%s".' %
                             (type(self), self.required_name))
        super(RowIDField, self).bind(model, name, *args)


class DocIDField(RowIDField):
    column_name = name = required_name = 'docid'


class AutoIncrementField(AutoField):
    def ddl(self, ctx):
        node_list = super(AutoIncrementField, self).ddl(ctx)
        return NodeList((node_list, SQL('AUTOINCREMENT')))


class TDecimalField(DecimalField):
    field_type = 'TEXT'
    def get_modifiers(self): pass


class ISODateTimeField(DateTimeField):
    formats = [
        '%Y-%m-%dT%H:%M:%S.%f%z',
        '%Y-%m-%dT%H:%M:%S%z',
        '%Y-%m-%dT%H:%M:%S.%f',
        '%Y-%m-%dT%H:%M:%S',
        '%Y-%m-%d',
    ]

    def db_value(self, value):
        if value:
            return value.isoformat()


class JSONPath(ColumnBase):
    def __init__(self, field, path=None):
        super(JSONPath, self).__init__()
        self._field = field
        self._path = path or ()

    def _converter(self, value):
        return self._field.python_value(value)

    @property
    def path(self):
        return Value('$%s' % ''.join(self._path))

    def __getitem__(self, idx):
        if isinstance(idx, int) or idx == '#':
            item = '[%s]' % idx
        else:
            item = '.%s' % idx
        return type(self)(self._field, self._path + (item,))

    def append(self, value, as_json=None):
        if as_json or isinstance(value, (list, dict)):
            value = fn.json(self._field._json_dumps(value))
        return fn.json_set(self._field, self['#'].path, value)

    def _json_operation(self, func, value, as_json=None):
        if as_json or isinstance(value, (list, dict)):
            value = fn.json(self._field._json_dumps(value))
        return func(self._field, self.path, value)

    def insert(self, value, as_json=None):
        return self._json_operation(fn.json_insert, value, as_json)

    def set(self, value, as_json=None):
        return self._json_operation(fn.json_set, value, as_json)

    def replace(self, value, as_json=None):
        return self._json_operation(fn.json_replace, value, as_json)

    def update(self, value):
        return self.set(fn.json_patch(self, self._field._json_dumps(value)))

    def remove(self):
        return fn.json_remove(self._field, self.path)

    def json_type(self):
        return fn.json_type(self._field, self.path)

    def length(self):
        return fn.json_array_length(self._field, self.path)

    def children(self):
        return fn.json_each(self._field, self.path)

    def tree(self):
        return fn.json_tree(self._field, self.path)

    def __sql__(self, ctx):
        return ctx.sql(fn.json_extract(self._field, self.path)
                       if self._path else self._field)

class JSONBPath(JSONPath):
    def append(self, value, as_json=None):
        if as_json or isinstance(value, (list, dict)):
            value = fn.jsonb(self._field._json_dumps(value))
        return fn.jsonb_set(self._field, self['#'].path, value)

    def _json_operation(self, func, value, as_json=None):
        if as_json or isinstance(value, (list, dict)):
            value = fn.jsonb(self._field._json_dumps(value))
        return func(self._field, self.path, value)

    def insert(self, value, as_json=None):
        return self._json_operation(fn.jsonb_insert, value, as_json)

    def set(self, value, as_json=None):
        return self._json_operation(fn.jsonb_set, value, as_json)

    def replace(self, value, as_json=None):
        return self._json_operation(fn.jsonb_replace, value, as_json)

    def update(self, value):
        return self.set(fn.jsonb_patch(self, self._field._json_dumps(value)))

    def remove(self):
        return fn.jsonb_remove(self._field, self.path)

    def __sql__(self, ctx):
        return ctx.sql(fn.jsonb_extract(self._field, self.path)
                       if self._path else self._field)


class JSONField(TextField):
    field_type = 'JSON'
    unpack = False
    Path = JSONPath

    def __init__(self, json_dumps=None, json_loads=None, **kwargs):
        self._json_dumps = json_dumps or json.dumps
        self._json_loads = json_loads or json.loads
        super(JSONField, self).__init__(**kwargs)

    def python_value(self, value):
        if value is not None:
            try:
                return self._json_loads(value)
            except (TypeError, ValueError):
                return value

    def db_value(self, value):
        if value is not None:
            if not isinstance(value, Node):
                value = fn.json(self._json_dumps(value))
            return value

    def _e(op):
        def inner(self, rhs):
            if isinstance(rhs, (list, dict)):
                rhs = AsIs(rhs, self.db_value)
            return Expression(self, op, rhs)
        return inner
    __eq__ = _e(OP.EQ)
    __ne__ = _e(OP.NE)
    __gt__ = _e(OP.GT)
    __ge__ = _e(OP.GTE)
    __lt__ = _e(OP.LT)
    __le__ = _e(OP.LTE)
    __hash__ = Field.__hash__

    def __getitem__(self, item):
        return self.Path(self)[item]

    def extract(self, *paths):
        paths = [Value(p, converter=False) for p in paths]
        return fn.json_extract(self, *paths)
    def extract_json(self, path):
        return Expression(self, '->', Value(path, converter=False))
    def extract_text(self, path):
        return Expression(self, '->>', Value(path, converter=False))

    def append(self, value, as_json=None):
        return self.Path(self).append(value, as_json)

    def insert(self, value, as_json=None):
        return self.Path(self).insert(value, as_json)

    def set(self, value, as_json=None):
        return self.Path(self).set(value, as_json)

    def replace(self, value, as_json=None):
        return self.Path(self).replace(value, as_json)

    def update(self, data):
        return self.Path(self).update(data)

    def remove(self, *paths):
        if not paths:
            return self.Path(self).remove()
        return fn.json_remove(self, *paths)

    def json_type(self):
        return fn.json_type(self)

    def length(self, path=None):
        args = (self, path) if path else (self,)
        return fn.json_array_length(*args)

    def children(self):
        """
        Schema of `json_each` and `json_tree`:

        key,
        value,
        type TEXT (object, array, string, etc),
        atom (value for primitive/scalar types, NULL for array and object)
        id INTEGER (unique identifier for element)
        parent INTEGER (unique identifier of parent element or NULL)
        fullkey TEXT (full path describing element)
        path TEXT (path to the container of the current element)
        json JSON hidden (1st input parameter to function)
        root TEXT hidden (2nd input parameter, path at which to start)
        """
        return fn.json_each(self)

    def tree(self):
        return fn.json_tree(self)


class JSONBField(JSONField):
    field_type = 'JSONB'
    Path = JSONBPath

    def db_value(self, value):
        if value is not None:
            if not isinstance(value, Node):
                value = fn.jsonb(self._json_dumps(value))
            return value

    def json(self):
        return fn.json(self)

    def extract(self, *paths):
        paths = [Value(p, converter=False) for p in paths]
        return fn.jsonb_extract(self, *paths)

    def remove(self, *paths):
        if not paths:
            return self.Path(self).remove()
        return fn.jsonb_remove(self, *paths)


class SearchField(Field):
    def __init__(self, unindexed=False, column_name=None, **k):
        if k:
            raise ValueError('SearchField does not accept these keyword '
                             'arguments: %s.' % sorted(k))
        super(SearchField, self).__init__(unindexed=unindexed,
                                          column_name=column_name, null=True)

    def match(self, term):
        return match(self, term)

    @property
    def fts_column_index(self):
        if not hasattr(self, '_fts_column_index'):
            search_fields = [f.name for f in self.model._meta.sorted_fields
                             if isinstance(f, SearchField)]
            self._fts_column_index = search_fields.index(self.name)
        return self._fts_column_index

    def highlight(self, left, right):
        column_idx = self.fts_column_index
        return fn.highlight(self.model._meta.entity, column_idx, left, right)

    def snippet(self, left, right, over_length='...', max_tokens=16):
        if not (0 < max_tokens < 65):
            raise ValueError('max_tokens must be between 1 and 64 (inclusive)')
        column_idx = self.fts_column_index
        return fn.snippet(self.model._meta.entity, column_idx, left, right,
                          over_length, max_tokens)


class VirtualTableSchemaManager(SchemaManager):
    def _create_virtual_table(self, safe=True, **options):
        options = self.model.clean_options(
            merge_dict(self.model._meta.options, options))

        # Structure:
        # CREATE VIRTUAL TABLE <model>
        # USING <extension_module>
        # ([prefix_arguments, ...] fields, ... [arguments, ...], [options...])
        ctx = self._create_context()
        ctx.literal('CREATE VIRTUAL TABLE ')
        if safe:
            ctx.literal('IF NOT EXISTS ')
        (ctx
         .sql(self.model)
         .literal(' USING '))

        ext_module = self.model._meta.extension_module
        if isinstance(ext_module, Node):
            return ctx.sql(ext_module)

        ctx.sql(SQL(ext_module)).literal(' ')
        arguments = []
        meta = self.model._meta

        if meta.prefix_arguments:
            arguments.extend([SQL(a) for a in meta.prefix_arguments])

        # Constraints, data-types, foreign and primary keys are all omitted.
        for field in meta.sorted_fields:
            if isinstance(field, (RowIDField)) or field._hidden:
                continue
            field_def = [Entity(field.column_name)]
            if field.unindexed:
                field_def.append(SQL('UNINDEXED'))
            arguments.append(NodeList(field_def))

        if meta.arguments:
            arguments.extend([SQL(a) for a in meta.arguments])

        if options:
            arguments.extend(self._create_table_option_sql(options))
        return ctx.sql(EnclosedNodeList(arguments))

    def _create_table(self, safe=True, **options):
        if issubclass(self.model, VirtualModel):
            return self._create_virtual_table(safe, **options)

        return super(VirtualTableSchemaManager, self)._create_table(
            safe, **options)


class VirtualModel(Model):
    class Meta:
        arguments = None
        extension_module = None
        prefix_arguments = None
        primary_key = False
        schema_manager_class = VirtualTableSchemaManager

    @classmethod
    def clean_options(cls, options):
        return options


class BaseFTSModel(VirtualModel):
    @classmethod
    def clean_options(cls, options):
        content = options.get('content')
        prefix = options.get('prefix')
        tokenize = options.get('tokenize')
        content_rowid = options.get('content_rowid')

        if isinstance(content, basestring) and content == '':
            # Special-case content-less full-text search tables.
            options['content'] = "''"
        elif isinstance(content, Field):
            # Special-case to ensure fields are fully-qualified.
            options['content'] = Entity(content.model._meta.table_name,
                                        content.column_name)

        if content_rowid is not None:
            options['content_rowid'] = content_rowid

        if prefix:
            if isinstance(prefix, (list, tuple)):
                prefix = ','.join([str(i) for i in prefix])
            options['prefix'] = "'%s'" % prefix.strip("' ")

        if tokenize and cls._meta.extension_module.lower() == 'fts5':
            # Tokenizers need to be in quoted string for FTS5, but not for FTS3
            # or FTS4.
            options['tokenize'] = '"%s"' % tokenize

        return options


class FTSModel(BaseFTSModel):
    """
    VirtualModel class for creating tables that use either the FTS3 or FTS4
    search extensions. Peewee automatically determines which version of the
    FTS extension is supported and will use FTS4 if possible.
    """
    # FTS3/4 uses "docid" in the same way a normal table uses "rowid".
    docid = DocIDField()

    class Meta:
        extension_module = 'FTS%s' % FTS_VERSION

    @classmethod
    def _fts_cmd(cls, cmd):
        tbl = cls._meta.table_name
        res = cls._meta.database.execute_sql(
            "INSERT INTO %s(%s) VALUES('%s');" % (tbl, tbl, cmd))
        return res.fetchone()

    @classmethod
    def optimize(cls):
        return cls._fts_cmd('optimize')

    @classmethod
    def rebuild(cls):
        return cls._fts_cmd('rebuild')

    @classmethod
    def integrity_check(cls):
        return cls._fts_cmd('integrity-check')

    @classmethod
    def merge(cls, blocks=200, segments=8):
        return cls._fts_cmd('merge=%s,%s' % (blocks, segments))

    @classmethod
    def automerge(cls, state=True):
        return cls._fts_cmd('automerge=%s' % (state and '1' or '0'))

    @classmethod
    def match(cls, term):
        """
        Generate a `MATCH` expression appropriate for searching this table.
        """
        return match(cls._meta.entity, term)

    @classmethod
    def rank(cls, *weights):
        matchinfo = fn.matchinfo(cls._meta.entity, FTS3_MATCHINFO)
        return fn.fts_rank(matchinfo, *weights)

    @classmethod
    def bm25(cls, *weights):
        match_info = fn.matchinfo(cls._meta.entity, FTS4_MATCHINFO)
        return fn.fts_bm25(match_info, *weights)

    @classmethod
    def bm25f(cls, *weights):
        match_info = fn.matchinfo(cls._meta.entity, FTS4_MATCHINFO)
        return fn.fts_bm25f(match_info, *weights)

    @classmethod
    def lucene(cls, *weights):
        match_info = fn.matchinfo(cls._meta.entity, FTS4_MATCHINFO)
        return fn.fts_lucene(match_info, *weights)

    @classmethod
    def _search(cls, term, weights, with_score, score_alias, score_fn,
                explicit_ordering):
        if not weights:
            rank = score_fn()
        elif isinstance(weights, dict):
            weight_args = []
            for field in cls._meta.sorted_fields:
                # Attempt to get the specified weight of the field by looking
                # it up using it's field instance followed by name.
                field_weight = weights.get(field, weights.get(field.name, 1.0))
                weight_args.append(field_weight)
            rank = score_fn(*weight_args)
        else:
            rank = score_fn(*weights)

        selection = ()
        order_by = rank
        if with_score:
            selection = (cls, rank.alias(score_alias))
        if with_score and not explicit_ordering:
            order_by = SQL(score_alias)

        return (cls
                .select(*selection)
                .where(cls.match(term))
                .order_by(order_by))

    @classmethod
    def search(cls, term, weights=None, with_score=False, score_alias='score',
               explicit_ordering=False):
        """Full-text search using selected `term`."""
        return cls._search(
            term,
            weights,
            with_score,
            score_alias,
            cls.rank,
            explicit_ordering)

    @classmethod
    def search_bm25(cls, term, weights=None, with_score=False,
                    score_alias='score', explicit_ordering=False):
        """Full-text search for selected `term` using BM25 algorithm."""
        return cls._search(
            term,
            weights,
            with_score,
            score_alias,
            cls.bm25,
            explicit_ordering)

    @classmethod
    def search_bm25f(cls, term, weights=None, with_score=False,
                     score_alias='score', explicit_ordering=False):
        """Full-text search for selected `term` using BM25 algorithm."""
        return cls._search(
            term,
            weights,
            with_score,
            score_alias,
            cls.bm25f,
            explicit_ordering)

    @classmethod
    def search_lucene(cls, term, weights=None, with_score=False,
                      score_alias='score', explicit_ordering=False):
        """Full-text search for selected `term` using BM25 algorithm."""
        return cls._search(
            term,
            weights,
            with_score,
            score_alias,
            cls.lucene,
            explicit_ordering)


_alphabet = 'abcdefghijklmnopqrstuvwxyz'
_alphanum = (set('\t ,"(){}*:_+0123456789') |
             set(_alphabet) |
             set(_alphabet.upper()) |
             set((chr(26),)))
_invalid_ascii = set(chr(p) for p in range(128) if chr(p) not in _alphanum)
del _alphabet
del _alphanum
_quote_re = re.compile(r'[^\s"]+|"[^"\\]*(?:\\.[^"\\]*)*"')


class FTS5Model(BaseFTSModel):
    """
    Requires SQLite >= 3.9.0.

    Table options:

    content: table name of external content, or empty string for "contentless"
    content_rowid: column name of external content primary key
    prefix: integer(s). Ex: '2' or '2 3 4'
    tokenize: porter, unicode61, ascii. Ex: 'porter unicode61'

    The unicode tokenizer supports the following parameters:

    * remove_diacritics (1 or 0, default is 1)
    * tokenchars (string of characters, e.g. '-_'
    * separators (string of characters)

    Parameters are passed as alternating parameter name and value, so:

    {'tokenize': "unicode61 remove_diacritics 0 tokenchars '-_'"}

    Content-less tables:

    If you don't need the full-text content in it's original form, you can
    specify a content-less table. Searches and auxiliary functions will work
    as usual, but the only values returned when SELECT-ing can be rowid. Also
    content-less tables do not support UPDATE or DELETE.

    External content tables:

    You can set up triggers to sync these, e.g.

    -- Create a table. And an external content fts5 table to index it.
    CREATE TABLE tbl(a INTEGER PRIMARY KEY, b);
    CREATE VIRTUAL TABLE ft USING fts5(b, content='tbl', content_rowid='a');

    -- Triggers to keep the FTS index up to date.
    CREATE TRIGGER tbl_ai AFTER INSERT ON tbl BEGIN
      INSERT INTO ft(rowid, b) VALUES (new.a, new.b);
    END;
    CREATE TRIGGER tbl_ad AFTER DELETE ON tbl BEGIN
      INSERT INTO ft(fts_idx, rowid, b) VALUES('delete', old.a, old.b);
    END;
    CREATE TRIGGER tbl_au AFTER UPDATE ON tbl BEGIN
      INSERT INTO ft(fts_idx, rowid, b) VALUES('delete', old.a, old.b);
      INSERT INTO ft(rowid, b) VALUES (new.a, new.b);
    END;

    Built-in auxiliary functions:

    * bm25(tbl[, weight_0, ... weight_n])
    * highlight(tbl, col_idx, prefix, suffix)
    * snippet(tbl, col_idx, prefix, suffix, ?, max_tokens)
    """
    # FTS5 does not support declared primary keys, but we can use the
    # implicit rowid.
    rowid = RowIDField()

    class Meta:
        extension_module = 'fts5'

    _error_messages = {
        'field_type': ('Besides the implicit `rowid` column, all columns must '
                       'be instances of SearchField'),
        'index': 'Secondary indexes are not supported for FTS5 models',
        'pk': 'FTS5 models must use the default `rowid` primary key',
    }

    @classmethod
    def validate_model(cls):
        # Perform FTS5-specific validation and options post-processing.
        if cls._meta.primary_key.name != 'rowid':
            raise ImproperlyConfigured(cls._error_messages['pk'])
        for field in cls._meta.fields.values():
            if not isinstance(field, (SearchField, RowIDField)):
                raise ImproperlyConfigured(cls._error_messages['field_type'])
        if cls._meta.indexes:
            raise ImproperlyConfigured(cls._error_messages['index'])

    @classmethod
    def fts5_installed(cls):
        if sqlite3.sqlite_version_info[:3] < FTS5_MIN_SQLITE_VERSION:
            return False

        # Test in-memory DB to determine if the FTS5 extension is installed.
        tmp_db = sqlite3.connect(':memory:')
        try:
            tmp_db.execute('CREATE VIRTUAL TABLE fts5test USING fts5 (data);')
        except:
            try:
                tmp_db.enable_load_extension(True)
                tmp_db.load_extension('fts5')
            except:
                return False
            else:
                cls._meta.database.load_extension('fts5')
        finally:
            tmp_db.close()

        return True

    @staticmethod
    def validate_query(query):
        """
        Simple helper function to indicate whether a search query is a
        valid FTS5 query. Note: this simply looks at the characters being
        used, and is not guaranteed to catch all problematic queries.
        """
        tokens = _quote_re.findall(query)
        for token in tokens:
            if token.startswith('"') and token.endswith('"'):
                continue
            if set(token) & _invalid_ascii:
                return False
        return True

    @staticmethod
    def clean_query(query, replace=chr(26)):
        """
        Clean a query of invalid tokens.
        """
        accum = []
        any_invalid = False
        tokens = _quote_re.findall(query)
        for token in tokens:
            if token.startswith('"') and token.endswith('"'):
                accum.append(token)
                continue
            token_set = set(token)
            invalid_for_token = token_set & _invalid_ascii
            if invalid_for_token:
                any_invalid = True
                for c in invalid_for_token:
                    token = token.replace(c, replace)
            accum.append(token)

        if any_invalid:
            return ' '.join(accum)
        return query

    @classmethod
    def match(cls, term):
        """
        Generate a `MATCH` expression appropriate for searching this table.
        """
        return match(cls._meta.entity, term)

    @classmethod
    def rank(cls, *args):
        return cls.bm25(*args) if args else SQL('rank')

    @classmethod
    def bm25(cls, *weights):
        return fn.bm25(cls._meta.entity, *weights)

    @classmethod
    def search(cls, term, weights=None, with_score=False, score_alias='score',
               explicit_ordering=False):
        """Full-text search using selected `term`."""
        return cls.search_bm25(
            FTS5Model.clean_query(term),
            weights,
            with_score,
            score_alias,
            explicit_ordering)

    @classmethod
    def search_bm25(cls, term, weights=None, with_score=False,
                    score_alias='score', explicit_ordering=False):
        """Full-text search using selected `term`."""
        if not weights:
            rank = SQL('rank')
        elif isinstance(weights, dict):
            weight_args = []
            for field in cls._meta.sorted_fields:
                if isinstance(field, SearchField) and not field.unindexed:
                    weight_args.append(
                        weights.get(field, weights.get(field.name, 1.0)))
            rank = fn.bm25(cls._meta.entity, *weight_args)
        else:
            rank = fn.bm25(cls._meta.entity, *weights)

        selection = ()
        order_by = rank
        if with_score:
            selection = (cls, rank.alias(score_alias))
        if with_score and not explicit_ordering:
            order_by = SQL(score_alias)

        return (cls
                .select(*selection)
                .where(cls.match(FTS5Model.clean_query(term)))
                .order_by(order_by))

    @classmethod
    def _fts_cmd_sql(cls, cmd, **extra_params):
        tbl = cls._meta.entity
        columns = [tbl]
        values = [cmd]
        for key, value in extra_params.items():
            columns.append(Entity(key))
            values.append(value)

        return NodeList((
            SQL('INSERT INTO'),
            cls._meta.entity,
            EnclosedNodeList(columns),
            SQL('VALUES'),
            EnclosedNodeList(values)))

    @classmethod
    def _fts_cmd(cls, cmd, **extra_params):
        query = cls._fts_cmd_sql(cmd, **extra_params)
        return cls._meta.database.execute(query)

    @classmethod
    def automerge(cls, level):
        if not (0 <= level <= 16):
            raise ValueError('level must be between 0 and 16')
        return cls._fts_cmd('automerge', rank=level)

    @classmethod
    def merge(cls, npages):
        return cls._fts_cmd('merge', rank=npages)

    @classmethod
    def optimize(cls):
        return cls._fts_cmd('optimize')

    @classmethod
    def rebuild(cls):
        return cls._fts_cmd('rebuild')

    @classmethod
    def set_pgsz(cls, pgsz):
        return cls._fts_cmd('pgsz', rank=pgsz)

    @classmethod
    def set_rank(cls, rank_expression):
        return cls._fts_cmd('rank', rank=rank_expression)

    @classmethod
    def delete_all(cls):
        return cls._fts_cmd('delete-all')

    @classmethod
    def integrity_check(cls, rank=0):
        return cls._fts_cmd('integrity-check', rank=rank)

    @classmethod
    def VocabModel(cls, table_type='row', table=None):
        if table_type not in ('row', 'col', 'instance'):
            raise ValueError('table_type must be either "row", "col" or '
                             '"instance".')

        attr = '_vocab_model_%s' % table_type

        if not hasattr(cls, attr):
            class Meta:
                database = cls._meta.database
                table_name = table or cls._meta.table_name + '_v'
                extension_module = fn.fts5vocab(
                    cls._meta.entity,
                    SQL(table_type))

            attrs = {
                'term': VirtualField(TextField),
                'doc': IntegerField(),
                'cnt': IntegerField(),
                'rowid': RowIDField(),
                'Meta': Meta,
            }
            if table_type == 'col':
                attrs['col'] = VirtualField(TextField)
            elif table_type == 'instance':
                attrs['offset'] = VirtualField(IntegerField)

            class_name = '%sVocab' % cls.__name__
            setattr(cls, attr, type(class_name, (VirtualModel,), attrs))

        return getattr(cls, attr)


OP.MATCH = 'MATCH'

def match(lhs, rhs):
    return Expression(lhs, OP.MATCH, rhs)