"""Module to handle the operations within the aggregate pipeline."""

import bisect
import collections
import copy
import datetime
import decimal
import functools
import itertools
import math
import numbers
from packaging import version
import random
import re
import sys
import warnings

import pytz
from sentinels import NOTHING

import mongomock
from mongomock import command_cursor
from mongomock import filtering
from mongomock import helpers
from mongomock import OperationFailure

try:
    from bson.errors import InvalidDocument
    from bson import Regex, decimal128
    decimal_support = True
    _RE_TYPES = (helpers.RE_TYPE, Regex)
except ImportError:
    InvalidDocument = OperationFailure
    decimal_support = False
    _RE_TYPES = (helpers.RE_TYPE,)

_random = random.Random()


group_operators = [
    '$addToSet',
    '$avg',
    '$first',
    '$last',
    '$max',
    '$mergeObjects',
    '$min',
    '$push',
    '$stdDevPop',
    '$stdDevSamp',
    '$sum',
]
unary_arithmetic_operators = {
    '$abs',
    '$ceil',
    '$exp',
    '$floor',
    '$ln',
    '$log10',
    '$sqrt',
    '$trunc',
}
binary_arithmetic_operators = {
    '$divide',
    '$log',
    '$mod',
    '$pow',
    '$subtract',
}
arithmetic_operators = unary_arithmetic_operators | binary_arithmetic_operators | {
    '$add',
    '$multiply',
}
project_operators = [
    '$max',
    '$min',
    '$avg',
    '$sum',
    '$stdDevPop',
    '$stdDevSamp',
    '$arrayElemAt',
    '$first',
    '$last',
]
control_flow_operators = [
    '$switch',
]
projection_operators = [
    '$let',
    '$literal',
]
date_operators = [
    '$dateFromString',
    '$dateToString',
    '$dateFromParts',
    '$dayOfMonth',
    '$dayOfWeek',
    '$dayOfYear',
    '$hour',
    '$isoDayOfWeek',
    '$isoWeek',
    '$isoWeekYear',
    '$millisecond',
    '$minute',
    '$month',
    '$second',
    '$week',
    '$year',
]
conditional_operators = ['$cond', '$ifNull']
array_operators = [
    '$concatArrays',
    '$filter',
    '$indexOfArray',
    '$map',
    '$range',
    '$reduce',
    '$reverseArray',
    '$size',
    '$slice',
    '$zip',
]
object_operators = [
    '$mergeObjects',
]
text_search_operators = ['$meta']
string_operators = [
    '$concat',
    '$indexOfBytes',
    '$indexOfCP',
    '$regexMatch',
    '$split',
    '$strcasecmp',
    '$strLenBytes',
    '$strLenCP',
    '$substr',
    '$substrBytes',
    '$substrCP',
    '$toLower',
    '$toUpper',
    '$trim',
]
comparison_operators = [
    '$cmp',
    '$eq',
    '$ne',
] + list(filtering.SORTING_OPERATOR_MAP.keys())
boolean_operators = ['$and', '$or', '$not']
set_operators = [
    '$in',
    '$setEquals',
    '$setIntersection',
    '$setDifference',
    '$setUnion',
    '$setIsSubset',
    '$anyElementTrue',
    '$allElementsTrue',
]

type_convertion_operators = [
    '$convert',
    '$toString',
    '$toInt',
    '$toDecimal',
    '$toLong',
    '$arrayToObject',
    '$objectToArray',
]
type_operators = [
    '$isNumber',
    '$isArray',
]


def _avg_operation(values):
    values_list = list(v for v in values if isinstance(v, numbers.Number))
    if not values_list:
        return None
    return sum(values_list) / float(len(list(values_list)))


def _group_operation(values, operator):
    values_list = list(v for v in values if v is not None)
    if not values_list:
        return None
    return operator(values_list)


def _sum_operation(values):
    values_list = list()
    if decimal_support:
        for v in values:
            if isinstance(v, numbers.Number):
                values_list.append(v)
            elif isinstance(v, decimal128.Decimal128):
                values_list.append(v.to_decimal())
    else:
        values_list = list(v for v in values if isinstance(v, numbers.Number))
    sum_value = sum(values_list)
    return decimal128.Decimal128(sum_value) if isinstance(sum_value, decimal.Decimal) else sum_value


def _merge_objects_operation(values):
    merged_doc = dict()
    for v in values:
        if isinstance(v, dict):
            merged_doc.update(v)
    return merged_doc


_GROUPING_OPERATOR_MAP = {
    '$sum': _sum_operation,
    '$avg': _avg_operation,
    '$mergeObjects': _merge_objects_operation,
    '$min': lambda values: _group_operation(values, min),
    '$max': lambda values: _group_operation(values, max),
    '$first': lambda values: values[0] if values else None,
    '$last': lambda values: values[-1] if values else None,
}


class _Parser:
    """Helper to parse expressions within the aggregate pipeline."""

    def __init__(self, doc_dict, user_vars=None, ignore_missing_keys=False):
        self._doc_dict = doc_dict
        self._ignore_missing_keys = ignore_missing_keys
        self._user_vars = user_vars or {}

    def parse(self, expression):
        """Parse a MongoDB expression."""
        if not isinstance(expression, dict):
            # May raise a KeyError despite the ignore missing key.
            return self._parse_basic_expression(expression)

        if len(expression) > 1 and any(key.startswith('$') for key in expression):
            raise OperationFailure(
                'an expression specification must contain exactly one field, '
                'the name of the expression. Found %d fields in %s'
                % (len(expression), expression))

        value_dict = {}
        for k, v in expression.items():
            if k in arithmetic_operators:
                return self._handle_arithmetic_operator(k, v)
            if k in project_operators:
                return self._handle_project_operator(k, v)
            if k in projection_operators:
                return self._handle_projection_operator(k, v)
            if k in comparison_operators:
                return self._handle_comparison_operator(k, v)
            if k in date_operators:
                return self._handle_date_operator(k, v)
            if k in array_operators:
                return self._handle_array_operator(k, v)
            if k in conditional_operators:
                return self._handle_conditional_operator(k, v)
            if k in control_flow_operators:
                return self._handle_control_flow_operator(k, v)
            if k in set_operators:
                return self._handle_set_operator(k, v)
            if k in string_operators:
                return self._handle_string_operator(k, v)
            if k in type_convertion_operators:
                return self._handle_type_convertion_operator(k, v)
            if k in type_operators:
                return self._handle_type_operator(k, v)
            if k in boolean_operators:
                return self._handle_boolean_operator(k, v)
            if k in text_search_operators + projection_operators + object_operators:
                raise NotImplementedError(
                    "'%s' is a valid operation but it is not supported by Mongomock yet." % k)
            if k.startswith('$'):
                raise OperationFailure("Unrecognized expression '%s'" % k)
            try:
                value = self.parse(v)
            except KeyError:
                if self._ignore_missing_keys:
                    continue
                raise
            value_dict[k] = value

        return value_dict

    def parse_many(self, values):
        for value in values:
            try:
                yield self.parse(value)
            except KeyError:
                if self._ignore_missing_keys:
                    yield None
                else:
                    raise

    def _parse_to_bool(self, expression):
        """Parse a MongoDB expression and then convert it to bool"""
        # handles converting `undefined` (in form of KeyError) to False
        try:
            return helpers.mongodb_to_bool(self.parse(expression))
        except KeyError:
            return False

    def _parse_or_nothing(self, expression):
        try:
            return self.parse(expression)
        except KeyError:
            return NOTHING

    def _parse_basic_expression(self, expression):
        if isinstance(expression, str) and expression.startswith('$'):
            if expression.startswith('$$'):
                return helpers.get_value_by_dot(dict({
                    'ROOT': self._doc_dict,
                    'CURRENT': self._doc_dict,
                }, **self._user_vars), expression[2:], can_generate_array=True)
            return helpers.get_value_by_dot(self._doc_dict, expression[1:], can_generate_array=True)
        return expression

    def _handle_boolean_operator(self, operator, values):
        if operator == '$and':
            return all([self._parse_to_bool(value) for value in values])
        if operator == '$or':
            return any(self._parse_to_bool(value) for value in values)
        if operator == '$not':
            return not self._parse_to_bool(values)
        # This should never happen: it is only a safe fallback if something went wrong.
        raise NotImplementedError(  # pragma: no cover
            "Although '%s' is a valid boolean operator for the "
            'aggregation pipeline, it is currently not implemented'
            ' in Mongomock.' % operator
        )

    def _handle_arithmetic_operator(self, operator, values):
        if operator in unary_arithmetic_operators:
            try:
                number = self.parse(values)
            except KeyError:
                return None
            if number is None:
                return None
            if not isinstance(number, numbers.Number):
                raise OperationFailure(
                    "Parameter to %s must evaluate to a number, got '%s'" %
                    (operator, type(number)))

            if operator == '$abs':
                return abs(number)
            if operator == '$ceil':
                return math.ceil(number)
            if operator == '$exp':
                return math.exp(number)
            if operator == '$floor':
                return math.floor(number)
            if operator == '$ln':
                return math.log(number)
            if operator == '$log10':
                return math.log10(number)
            if operator == '$sqrt':
                return math.sqrt(number)
            if operator == '$trunc':
                return math.trunc(number)

        if operator in binary_arithmetic_operators:
            if not isinstance(values, (tuple, list)):
                raise OperationFailure(
                    "Parameter to %s must evaluate to a list, got '%s'" %
                    (operator, type(values)))

            if len(values) != 2:
                raise OperationFailure('%s must have only 2 parameters' % operator)
            number_0, number_1 = self.parse_many(values)
            if number_0 is None or number_1 is None:
                return None

            if operator == '$divide':
                return number_0 / number_1
            if operator == '$log':
                return math.log(number_0, number_1)
            if operator == '$mod':
                return math.fmod(number_0, number_1)
            if operator == '$pow':
                return math.pow(number_0, number_1)
            if operator == '$subtract':
                if isinstance(number_0, datetime.datetime) and \
                        isinstance(number_1, (int, float)):
                    number_1 = datetime.timedelta(milliseconds=number_1)
                res = number_0 - number_1
                if isinstance(res, datetime.timedelta):
                    return round(res.total_seconds() * 1000)
                return res

        assert isinstance(values, (tuple, list)), \
            "Parameter to {} must evaluate to a list, got '{}'".format(operator, type(values))

        parsed_values = list(self.parse_many(values))
        assert parsed_values, '%s must have at least one parameter' % operator
        for value in parsed_values:
            if value is None:
                return None
            assert isinstance(value, numbers.Number), '%s only uses numbers' % operator
        if operator == '$add':
            return sum(parsed_values)
        if operator == '$multiply':
            return functools.reduce(lambda x, y: x * y, parsed_values)

        # This should never happen: it is only a safe fallback if something went wrong.
        raise NotImplementedError(  # pragma: no cover
            "Although '%s' is a valid aritmetic operator for the aggregation "
            'pipeline, it is currently not implemented  in Mongomock.' % operator)

    def _handle_project_operator(self, operator, values):
        if operator in _GROUPING_OPERATOR_MAP:
            values = self.parse(values) if isinstance(values, str) else self.parse_many(values)
            return _GROUPING_OPERATOR_MAP[operator](values)
        if operator == '$arrayElemAt':
            key, value = values
            array = self.parse(key)
            index = self.parse(value)
            try:
                return array[index]
            except IndexError as error:
                raise KeyError('Array have length less than index value') from error

        raise NotImplementedError("Although '%s' is a valid project operator for the "
                                  'aggregation pipeline, it is currently not implemented '
                                  'in Mongomock.' % operator)

    def _handle_projection_operator(self, operator, value):
        if operator == '$literal':
            return value
        if operator == '$let':
            if not isinstance(value, dict):
                raise InvalidDocument('$let only supports an object as its argument')
            for field in ('vars', 'in'):
                if field not in value:
                    raise OperationFailure(f"Missing '{field}' parameter to $let")
            if not isinstance(value['vars'], dict):
                raise OperationFailure('invalid parameter: expected an object (vars)')
            user_vars = {
                var_key: self.parse(var_value)
                for var_key, var_value in value['vars'].items()
            }
            return _Parser(
                self._doc_dict,
                dict(self._user_vars, **user_vars),
                ignore_missing_keys=self._ignore_missing_keys,
            ).parse(value['in'])
        raise NotImplementedError("Although '%s' is a valid project operator for the "
                                  'aggregation pipeline, it is currently not implemented '
                                  'in Mongomock.' % operator)

    def _handle_comparison_operator(self, operator, values):
        assert len(values) == 2, 'Comparison requires two expressions'
        a = self.parse(values[0])
        b = self.parse(values[1])
        if operator == '$eq':
            return a == b
        if operator == '$ne':
            return a != b
        if operator in filtering.SORTING_OPERATOR_MAP:
            return filtering.bson_compare(filtering.SORTING_OPERATOR_MAP[operator], a, b)
        raise NotImplementedError(
            "Although '%s' is a valid comparison operator for the "
            'aggregation pipeline, it is currently not implemented '
            ' in Mongomock.' % operator)

    def _handle_string_operator(self, operator, values):
        if operator == '$toLower':
            parsed = self.parse(values)
            return str(parsed).lower() if parsed is not None else ''
        if operator == '$toUpper':
            parsed = self.parse(values)
            return str(parsed).upper() if parsed is not None else ''
        if operator == '$concat':
            parsed_list = list(self.parse_many(values))
            return None if None in parsed_list else ''.join([str(x) for x in parsed_list])
        if operator == '$split':
            if len(values) != 2:
                raise OperationFailure('split must have 2 items')
            try:
                string = self.parse(values[0])
                delimiter = self.parse(values[1])
            except KeyError:
                return None

            if string is None or delimiter is None:
                return None
            if not isinstance(string, str):
                raise TypeError('split first argument must evaluate to string')
            if not isinstance(delimiter, str):
                raise TypeError('split second argument must evaluate to string')
            return string.split(delimiter)
        if operator == '$substr':
            if len(values) != 3:
                raise OperationFailure('substr must have 3 items')
            string = str(self.parse(values[0]))
            first = self.parse(values[1])
            length = self.parse(values[2])
            if string is None:
                return ''
            if first < 0:
                warnings.warn('Negative starting point given to $substr is accepted only until '
                              'MongoDB 3.7. This behavior will change in the future.')
                return ''
            if length < 0:
                warnings.warn('Negative length given to $substr is accepted only until '
                              'MongoDB 3.7. This behavior will change in the future.')
            second = len(string) if length < 0 else first + length
            return string[first:second]
        if operator == '$strcasecmp':
            if len(values) != 2:
                raise OperationFailure('strcasecmp must have 2 items')
            a, b = str(self.parse(values[0])), str(self.parse(values[1]))
            return 0 if a == b else -1 if a < b else 1
        if operator == '$regexMatch':
            if not isinstance(values, dict):
                raise OperationFailure(
                    '$regexMatch expects an object of named arguments but found: %s' % type(values))
            for field in ('input', 'regex'):
                if field not in values:
                    raise OperationFailure("$regexMatch requires '%s' parameter" % field)
            unknown_args = set(values) - {'input', 'regex', 'options'}
            if unknown_args:
                raise OperationFailure(
                    '$regexMatch found an unknown argument: %s' % list(unknown_args)[0])

            try:
                input_value = self.parse(values['input'])
            except KeyError:
                return False
            if not isinstance(input_value, str):
                raise OperationFailure("$regexMatch needs 'input' to be of type string")

            try:
                regex_val = self.parse(values['regex'])
            except KeyError:
                return False
            options = None
            for option in values.get('options', ''):
                if option not in 'imxs':
                    raise OperationFailure(
                        '$regexMatch invalid flag in regex options: %s' % option)
                re_option = getattr(re, option.upper())
                if options is None:
                    options = re_option
                else:
                    options |= re_option
            if isinstance(regex_val, str):
                if options is None:
                    regex = re.compile(regex_val)
                else:
                    regex = re.compile(regex_val, options)
            elif 'options' in values and regex_val.flags:
                raise OperationFailure(
                    "$regexMatch: regex option(s) specified in both 'regex' and 'option' fields")
            elif isinstance(regex_val, helpers.RE_TYPE):
                if options and not regex_val.flags:
                    regex = re.compile(regex_val.pattern, options)
                elif regex_val.flags & ~(re.I | re.M | re.X | re.S):
                    raise OperationFailure(
                        '$regexMatch invalid flag in regex options: %s' % regex_val.flags)
                else:
                    regex = regex_val
            elif isinstance(regex_val, _RE_TYPES):
                # bson.Regex
                if regex_val.flags & ~(re.I | re.M | re.X | re.S):
                    raise OperationFailure(
                        '$regexMatch invalid flag in regex options: %s' % regex_val.flags)
                regex = re.compile(regex_val.pattern, regex_val.flags or options)
            else:
                raise OperationFailure("$regexMatch needs 'regex' to be of type string or regex")

            return bool(regex.search(input_value))

        # This should never happen: it is only a safe fallback if something went wrong.
        raise NotImplementedError(  # pragma: no cover
            "Although '%s' is a valid string operator for the aggregation "
            'pipeline, it is currently not implemented  in Mongomock.' % operator)

    def _handle_date_operator(self, operator, values):
        if isinstance(values, dict) and values.keys() == {'date', 'timezone'}:
            value = self.parse(values['date'])
            target_tz = pytz.timezone(values['timezone'])
            out_value = value.replace(tzinfo=pytz.utc).astimezone(target_tz)
        else:
            out_value = self.parse(values)

        if operator == '$dayOfYear':
            return out_value.timetuple().tm_yday
        if operator == '$dayOfMonth':
            return out_value.day
        if operator == '$dayOfWeek':
            return (out_value.isoweekday() % 7) + 1
        if operator == '$year':
            return out_value.year
        if operator == '$month':
            return out_value.month
        if operator == '$week':
            return int(out_value.strftime('%U'))
        if operator == '$hour':
            return out_value.hour
        if operator == '$minute':
            return out_value.minute
        if operator == '$second':
            return out_value.second
        if operator == '$millisecond':
            return int(out_value.microsecond / 1000)
        if operator == '$dateToString':
            if not isinstance(values, dict):
                raise OperationFailure(
                    '$dateToString operator must correspond a dict'
                    'that has "format" and "date" field.'
                )
            if not isinstance(values, dict) or not {'format', 'date'} <= set(values):
                raise OperationFailure(
                    '$dateToString operator must correspond a dict'
                    'that has "format" and "date" field.'
                )
            if '%L' in out_value['format']:
                raise NotImplementedError(
                    'Although %L is a valid date format for the '
                    '$dateToString operator, it is currently not implemented '
                    ' in Mongomock.'
                )
            if 'onNull' in values:
                raise NotImplementedError(
                    'Although onNull is a valid field for the '
                    '$dateToString operator, it is currently not implemented '
                    ' in Mongomock.'
                )
            if 'timezone' in values.keys():
                raise NotImplementedError(
                    'Although timezone is a valid field for the '
                    '$dateToString operator, it is currently not implemented '
                    ' in Mongomock.'
                )
            return out_value['date'].strftime(out_value['format'])
        if operator == '$dateFromParts':
            if not isinstance(out_value, dict):
                raise OperationFailure(
                    f'{operator} operator must correspond a dict '
                    'that has "year" or "isoWeekYear" field.'
                )
            if len(set(out_value) & {'year', 'isoWeekYear'}) != 1:
                raise OperationFailure(
                    f'{operator} operator must correspond a dict '
                    'that has "year" or "isoWeekYear" field.'
                )
            for field in ('isoWeekYear', 'isoWeek', 'isoDayOfWeek', 'timezone'):
                if field in out_value:
                    raise NotImplementedError(
                        f'Although {field} is a valid field for the '
                        f'{operator} operator, it is currently not implemented '
                        'in Mongomock.'
                    )

            year = out_value['year']
            month = out_value.get('month', 1) or 1
            day = out_value.get('day', 1) or 1
            hour = out_value.get('hour', 0) or 0
            minute = out_value.get('minute', 0) or 0
            second = out_value.get('second', 0) or 0
            millisecond = out_value.get('millisecond', 0) or 0

            return datetime.datetime(
                year=year,
                month=month,
                day=day,
                hour=hour,
                minute=minute,
                second=second,
                microsecond=millisecond
            )

        raise NotImplementedError(
            "Although '%s' is a valid date operator for the "
            'aggregation pipeline, it is currently not implemented '
            ' in Mongomock.' % operator)

    def _handle_array_operator(self, operator, value):
        if operator == '$concatArrays':
            if not isinstance(value, (list, tuple)):
                value = [value]

            parsed_list = list(self.parse_many(value))
            for parsed_item in parsed_list:
                if parsed_item is not None and not isinstance(parsed_item, (list, tuple)):
                    raise OperationFailure(
                        f'$concatArrays only supports arrays, not {type(parsed_item)}'
                    )

            return None if None in parsed_list else list(itertools.chain.from_iterable(parsed_list))

        if operator == '$map':
            if not isinstance(value, dict):
                raise OperationFailure('$map only supports an object as its argument')

            # NOTE: while the two validations below could be achieved with
            # one-liner set operations (e.g. set(value) - {'input', 'as',
            # 'in'}), we prefer the iteration-based approaches in order to
            # mimic MongoDB's behavior regarding the order of evaluation. For
            # example, MongoDB complains about 'input' parameter missing before
            # 'in'.
            for k in ('input', 'in'):
                if k not in value:
                    raise OperationFailure("Missing '%s' parameter to $map" % k)

            for k in value:
                if k not in {'input', 'as', 'in'}:
                    raise OperationFailure('Unrecognized parameter to $map: %s' % k)

            input_array = self._parse_or_nothing(value['input'])

            if input_array is None or input_array is NOTHING:
                return None

            if not isinstance(input_array, (list, tuple)):
                raise OperationFailure('input to $map must be an array not %s' % type(input_array))

            fieldname = value.get('as', 'this')
            in_expr = value['in']
            return [
                _Parser(
                    self._doc_dict,
                    dict(self._user_vars, **{fieldname: item}),
                    ignore_missing_keys=self._ignore_missing_keys,
                ).parse(in_expr)
                for item in input_array
            ]

        if operator == '$size':
            if isinstance(value, list):
                if len(value) != 1:
                    raise OperationFailure('Expression $size takes exactly 1 arguments. '
                                           '%d were passed in.' % len(value))
                value = value[0]
            array_value = self._parse_or_nothing(value)
            if not isinstance(array_value, (list, tuple)):
                raise OperationFailure(
                    'The argument to $size must be an array, but was of type: %s' %
                    ('missing' if array_value is NOTHING else type(array_value)))
            return len(array_value)

        if operator == '$filter':
            if not isinstance(value, dict):
                raise OperationFailure('$filter only supports an object as its argument')
            extra_params = set(value) - {'input', 'cond', 'as'}
            if extra_params:
                raise OperationFailure('Unrecognized parameter to $filter: %s' % extra_params.pop())
            missing_params = {'input', 'cond'} - set(value)
            if missing_params:
                raise OperationFailure("Missing '%s' parameter to $filter" % missing_params.pop())

            input_array = self.parse(value['input'])
            fieldname = value.get('as', 'this')
            cond = value['cond']
            return [
                item for item in input_array
                if _Parser(
                    self._doc_dict,
                    dict(self._user_vars, **{fieldname: item}),
                    ignore_missing_keys=self._ignore_missing_keys,
                ).parse(cond)
            ]
        if operator == '$slice':
            if not isinstance(value, list):
                raise OperationFailure('$slice only supports a list as its argument')
            if len(value) < 2 or len(value) > 3:
                raise OperationFailure('Expression $slice takes at least 2 arguments, and at most '
                                       '3, but {} were passed in'.format(len(value)))
            array_value = self.parse(value[0])
            if not isinstance(array_value, list):
                raise OperationFailure(
                    'First argument to $slice must be an array, but is of type: {}'
                    .format(type(array_value)))
            for num, v in zip(('Second', 'Third'), value[1:]):
                if not isinstance(v, int):
                    raise OperationFailure(
                        '{} argument to $slice must be numeric, but is of type: {}'
                        .format(num, type(v)))
            if len(value) > 2 and value[2] <= 0:
                raise OperationFailure('Third argument to $slice must be '
                                       'positive: {}'.format(value[2]))

            start = value[1]
            if start < 0:
                if len(value) > 2:
                    stop = len(array_value) + start + value[2]
                else:
                    stop = None
            elif len(value) > 2:
                stop = start + value[2]
            else:
                stop = start
                start = 0
            return array_value[start:stop]

        raise NotImplementedError(
            "Although '%s' is a valid array operator for the "
            'aggregation pipeline, it is currently not implemented '
            'in Mongomock.' % operator)

    def _handle_type_convertion_operator(self, operator, values):
        if operator == '$toString':
            try:
                parsed = self.parse(values)
            except KeyError:
                return None
            if isinstance(parsed, bool):
                return str(parsed).lower()
            if isinstance(parsed, datetime.datetime):
                return parsed.isoformat()[:-3] + 'Z'
            return str(parsed)

        if operator == '$toInt':
            try:
                parsed = self.parse(values)
            except KeyError:
                return None
            if decimal_support:
                if isinstance(parsed, decimal128.Decimal128):
                    return int(parsed.to_decimal())
                return int(parsed)
            raise NotImplementedError(
                'You need to import the pymongo library to support decimal128 type.'
            )

        if operator == '$toLong':
            try:
                parsed = self.parse(values)
            except KeyError:
                return None
            if decimal_support:
                if isinstance(parsed, decimal128.Decimal128):
                    return int(parsed.to_decimal())
                return int(parsed)
            raise NotImplementedError(
                'You need to import the pymongo library to support decimal128 type.'
            )

        # Document: https://docs.mongodb.com/manual/reference/operator/aggregation/toDecimal/
        if operator == '$toDecimal':
            if not decimal_support:
                raise NotImplementedError(
                    'You need to import the pymongo library to support decimal128 type.'
                )
            try:
                parsed = self.parse(values)
            except KeyError:
                return None
            if isinstance(parsed, bool):
                parsed = '1' if parsed is True else '0'
                decimal_value = decimal128.Decimal128(parsed)
            elif isinstance(parsed, int):
                decimal_value = decimal128.Decimal128(str(parsed))
            elif isinstance(parsed, float):
                exp = decimal.Decimal('.00000000000000')
                decimal_value = decimal.Decimal(str(parsed)).quantize(exp)
                decimal_value = decimal128.Decimal128(decimal_value)
            elif isinstance(parsed, decimal128.Decimal128):
                decimal_value = parsed
            elif isinstance(parsed, str):
                try:
                    decimal_value = decimal128.Decimal128(parsed)
                except decimal.InvalidOperation as err:
                    raise OperationFailure(
                        "Failed to parse number '%s' in $convert with no onError value:"
                        'Failed to parse string to decimal' % parsed) from err
            elif isinstance(parsed, datetime.datetime):
                epoch = datetime.datetime.utcfromtimestamp(0)
                string_micro_seconds = str((parsed - epoch).total_seconds() * 1000).split('.', 1)[0]
                decimal_value = decimal128.Decimal128(string_micro_seconds)
            else:
                raise TypeError("'%s' type is not supported" % type(parsed))
            return decimal_value

        # Document: https://docs.mongodb.com/manual/reference/operator/aggregation/arrayToObject/
        if operator == '$arrayToObject':
            try:
                parsed = self.parse(values)
            except KeyError:
                return None

            if parsed is None:
                return None

            if not isinstance(parsed, (list, tuple)):
                raise OperationFailure(
                    f'$arrayToObject requires an array input, found: {type(parsed)}'
                )

            if all(isinstance(x, dict) and set(x.keys()) == {'k', 'v'} for x in parsed):
                return {d['k']: d['v'] for d in parsed}

            if all(isinstance(x, (list, tuple)) and len(x) == 2 for x in parsed):
                return dict(parsed)

            raise OperationFailure(
                'arrays used with $arrayToObject must contain documents '
                'with k and v fields or two-element arrays'
            )

        # Document: https://docs.mongodb.com/manual/reference/operator/aggregation/objectToArray/
        if operator == '$objectToArray':
            try:
                parsed = self.parse(values)
            except KeyError:
                return None

            if parsed is None:
                return None

            if not isinstance(parsed, (dict, collections.OrderedDict)):
                raise OperationFailure(
                    f'$objectToArray requires an object input, found: {type(parsed)}'
                )

            if len(parsed) > 1 and sys.version_info < (3, 6):
                raise NotImplementedError(
                    "Although '%s' is a valid type conversion, it is not implemented for Python 2 "
                    'and Python 3.5 in Mongomock yet.' % operator)

            return [{'k': k, 'v': v} for k, v in parsed.items()]

        raise NotImplementedError(
            "Although '%s' is a valid type conversion operator for the "
            'aggregation pipeline, it is currently not implemented '
            'in Mongomock.' % operator)

    def _handle_type_operator(self, operator, values):
        # Document: https://docs.mongodb.com/manual/reference/operator/aggregation/isNumber/
        if operator == '$isNumber':
            try:
                parsed = self.parse(values)
            except KeyError:
                return False
            return False if isinstance(parsed, bool) else isinstance(parsed, numbers.Number)

        # Document: https://docs.mongodb.com/manual/reference/operator/aggregation/isArray/
        if operator == '$isArray':
            try:
                parsed = self.parse(values)
            except KeyError:
                return False
            return isinstance(parsed, (tuple, list))

        raise NotImplementedError(  # pragma: no cover
            "Although '%s' is a valid type operator for the aggregation pipeline, it is currently "
            'not implemented in Mongomock.' % operator)

    def _handle_conditional_operator(self, operator, values):
        if operator == '$ifNull':
            fields = values[:-1]
            if len(fields) > 1 and version.parse(mongomock.SERVER_VERSION) <= version.parse('4.4'):
                raise OperationFailure(
                    '$ifNull supports only one input expression '
                    ' in MongoDB v4.4 and lower'
                )
            fallback = values[-1]
            for field in fields:
                try:
                    out_value = self.parse(field)
                    if out_value is not None:
                        return out_value
                except KeyError:
                    pass
            return self.parse(fallback)
        if operator == '$cond':
            if isinstance(values, list):
                condition, true_case, false_case = values
            elif isinstance(values, dict):
                condition = values['if']
                true_case = values['then']
                false_case = values['else']
            condition_value = self._parse_to_bool(condition)
            expression = true_case if condition_value else false_case
            return self.parse(expression)
        # This should never happen: it is only a safe fallback if something went wrong.
        raise NotImplementedError(  # pragma: no cover
            "Although '%s' is a valid conditional operator for the "
            'aggregation pipeline, it is currently not implemented '
            ' in Mongomock.' % operator)

    def _handle_control_flow_operator(self, operator, values):
        if operator == '$switch':
            if not isinstance(values, dict):
                raise OperationFailure(
                    '$switch requires an object as an argument, '
                    'found: %s' % type(values)
                )

            branches = values.get('branches', [])
            if not isinstance(branches, (list, tuple)):
                raise OperationFailure(
                    "$switch expected an array for 'branches', "
                    'found: %s' % type(branches)
                )
            if not branches:
                raise OperationFailure(
                    '$switch requires at least one branch.'
                )

            for branch in branches:
                if not isinstance(branch, dict):
                    raise OperationFailure(
                        '$switch expected each branch to be an object, '
                        'found: %s' % type(branch)
                    )
                if 'case' not in branch:
                    raise OperationFailure(
                        "$switch requires each branch have a 'case' expression"
                    )
                if 'then' not in branch:
                    raise OperationFailure(
                        "$switch requires each branch have a 'then' expression."
                    )

            for branch in branches:
                if self._parse_to_bool(branch['case']):
                    return self.parse(branch['then'])

            if 'default' not in values:
                raise OperationFailure(
                    '$switch could not find a matching branch for an input, '
                    'and no default was specified.'
                )
            return self.parse(values['default'])

        # This should never happen: it is only a safe fallback if something went wrong.
        raise NotImplementedError(  # pragma: no cover
            "Although '%s' is a valid control flow operator for the "
            'aggregation pipeline, it is currently not implemented '
            'in Mongomock.' % operator)

    def _handle_set_operator(self, operator, values):
        if operator == '$in':
            expression, array = values
            return self.parse(expression) in self.parse(array)
        if operator == '$setUnion':
            result = []
            for set_value in values:
                for value in self.parse(set_value):
                    if value not in result:
                        result.append(value)
            return result
        if operator == '$setEquals':
            set_values = [set(self.parse(value)) for value in values]
            for set1, set2 in itertools.combinations(set_values, 2):
                if set1 != set2:
                    return False
            return True
        raise NotImplementedError(
            "Although '%s' is a valid set operator for the aggregation "
            'pipeline, it is currently not implemented in Mongomock.' % operator)


def _parse_expression(expression, doc_dict, ignore_missing_keys=False):
    """Parse an expression.

    Args:
        expression: an Aggregate Expression, see
            https://docs.mongodb.com/manual/meta/aggregation-quick-reference/#aggregation-expressions.
        doc_dict: the document on which to evaluate the expression.
        ignore_missing_keys: if True, missing keys evaluated by the expression are ignored silently
            if it is possible.
    """
    return _Parser(doc_dict, ignore_missing_keys=ignore_missing_keys).parse(expression)


filtering.register_parse_expression(_parse_expression)


def _accumulate_group(output_fields, group_list):
    doc_dict = {}
    for field, value in output_fields.items():
        if field == '_id':
            continue
        for operator, key in value.items():
            values = []
            for doc in group_list:
                try:
                    values.append(_parse_expression(key, doc))
                except KeyError:
                    continue
            if operator in _GROUPING_OPERATOR_MAP:
                doc_dict[field] = _GROUPING_OPERATOR_MAP[operator](values)
            elif operator == '$addToSet':
                value = []
                val_it = (val or None for val in values)
                # Don't use set in case elt in not hashable (like dicts).
                for elt in val_it:
                    if elt not in value:
                        value.append(elt)
                doc_dict[field] = value
            elif operator == '$push':
                if field not in doc_dict:
                    doc_dict[field] = values
                else:
                    doc_dict[field].extend(values)
            elif operator in group_operators:
                raise NotImplementedError(
                    'Although %s is a valid group operator for the '
                    'aggregation pipeline, it is currently not implemented '
                    'in Mongomock.' % operator)
            else:
                raise NotImplementedError(
                    '%s is not a valid group operator for the aggregation '
                    'pipeline. See http://docs.mongodb.org/manual/meta/'
                    'aggregation-quick-reference/ for a complete list of '
                    'valid operators.' % operator)
    return doc_dict


def _fix_sort_key(key_getter):
    def fixed_getter(doc):
        key = key_getter(doc)
        # Convert dictionaries to make sorted() work in Python 3.
        if isinstance(key, dict):
            return [(k, v) for (k, v) in sorted(key.items())]
        return key
    return fixed_getter


def _handle_lookup_stage(in_collection, database, options):
    for operator in ('let', 'pipeline'):
        if operator in options:
            raise NotImplementedError(
                "Although '%s' is a valid lookup operator for the "
                'aggregation pipeline, it is currently not '
                'implemented in Mongomock.' % operator)
    for operator in ('from', 'localField', 'foreignField', 'as'):
        if operator not in options:
            raise OperationFailure(
                "Must specify '%s' field for a $lookup" % operator)
        if not isinstance(options[operator], str):
            raise OperationFailure(
                'Arguments to $lookup must be strings')
        if operator in ('as', 'localField', 'foreignField') and \
                options[operator].startswith('$'):
            raise OperationFailure(
                "FieldPath field names may not start with '$'")
        if operator == 'as' and \
                '.' in options[operator]:
            raise NotImplementedError(
                "Although '.' is valid in the 'as' "
                'parameters for the lookup stage of the aggregation '
                'pipeline, it is currently not implemented in Mongomock.')

    foreign_name = options['from']
    local_field = options['localField']
    foreign_field = options['foreignField']
    local_name = options['as']
    foreign_collection = database.get_collection(foreign_name)
    for doc in in_collection:
        try:
            query = helpers.get_value_by_dot(doc, local_field)
        except KeyError:
            query = None
        if isinstance(query, list):
            query = {'$in': query}
        matches = foreign_collection.find({foreign_field: query})
        doc[local_name] = [foreign_doc for foreign_doc in matches]

    return in_collection


def _recursive_get(match, nested_fields):
    head = match.get(nested_fields[0])
    remaining_fields = nested_fields[1:]
    if not remaining_fields:
        # Final/last field reached.
        yield head
        return
    # More fields to go, must be list, tuple, or dict.
    if isinstance(head, (list, tuple)):
        for m in head:
            # Yield from _recursive_get(m, remaining_fields).
            yield from _recursive_get(m, remaining_fields)
    elif isinstance(head, dict):
        # Yield from _recursive_get(head, remaining_fields).
        yield from _recursive_get(head, remaining_fields)


def _handle_graph_lookup_stage(in_collection, database, options):
    if not isinstance(options.get('maxDepth', 0), int):
        raise OperationFailure(
            "Argument 'maxDepth' to $graphLookup must be a number")
    if not isinstance(options.get('restrictSearchWithMatch', {}), dict):
        raise OperationFailure(
            "Argument 'restrictSearchWithMatch' to $graphLookup must be a Dictionary")
    if not isinstance(options.get('depthField', ''), str):
        raise OperationFailure(
            "Argument 'depthField' to $graphlookup must be a string")
    if 'startWith' not in options:
        raise OperationFailure(
            "Must specify 'startWith' field for a $graphLookup")
    for operator in ('as', 'connectFromField', 'connectToField', 'from'):
        if operator not in options:
            raise OperationFailure(
                "Must specify '%s' field for a $graphLookup" % operator)
        if not isinstance(options[operator], str):
            raise OperationFailure(
                "Argument '%s' to $graphLookup must be string" % operator)
        if options[operator].startswith('$'):
            raise OperationFailure("FieldPath field names may not start with '$'")
        if operator == 'as' and '.' in options[operator]:
            raise NotImplementedError(
                "Although '.' is valid in the '%s' "
                'parameter for the $graphLookup stage of the aggregation '
                'pipeline, it is currently not implemented in Mongomock.' % operator)

    foreign_name = options['from']
    start_with = options['startWith']
    connect_from_field = options['connectFromField']
    connect_to_field = options['connectToField']
    local_name = options['as']
    max_depth = options.get('maxDepth', None)
    depth_field = options.get('depthField', None)
    restrict_search_with_match = options.get('restrictSearchWithMatch', {})
    foreign_collection = database.get_collection(foreign_name)
    out_doc = copy.deepcopy(in_collection)  # TODO(pascal): speed the deep copy

    def _find_matches_for_depth(query):
        if isinstance(query, list):
            query = {'$in': query}
        matches = foreign_collection.find({connect_to_field: query})
        new_matches = []
        for new_match in matches:
            if filtering.filter_applies(restrict_search_with_match, new_match) \
                    and new_match['_id'] not in found_items:
                if depth_field is not None:
                    new_match = collections.OrderedDict(new_match, **{depth_field: depth})
                new_matches.append(new_match)
                found_items.add(new_match['_id'])
        return new_matches

    for doc in out_doc:
        found_items = set()
        depth = 0
        try:
            result = _parse_expression(start_with, doc)
        except KeyError:
            continue
        origin_matches = doc[local_name] = _find_matches_for_depth(result)
        while origin_matches and (max_depth is None or depth < max_depth):
            depth += 1
            newly_discovered_matches = []
            for match in origin_matches:
                nested_fields = connect_from_field.split('.')
                for match_target in _recursive_get(match, nested_fields):
                    newly_discovered_matches += _find_matches_for_depth(match_target)
            doc[local_name] += newly_discovered_matches
            origin_matches = newly_discovered_matches
    return out_doc


def _handle_group_stage(in_collection, unused_database, options):
    grouped_collection = []
    _id = options['_id']
    if _id:

        def _key_getter(doc):
            try:
                return _parse_expression(_id, doc, ignore_missing_keys=True)
            except KeyError:
                return None

        def _sort_key_getter(doc):
            return filtering.BsonComparable(_key_getter(doc))

        # Sort the collection only for the itertools.groupby.
        # $group does not order its output document.
        sorted_collection = sorted(in_collection, key=_sort_key_getter)
        grouped = itertools.groupby(sorted_collection, _key_getter)
    else:
        grouped = [(None, in_collection)]

    for doc_id, group in grouped:
        group_list = ([x for x in group])
        doc_dict = _accumulate_group(options, group_list)
        doc_dict['_id'] = doc_id
        grouped_collection.append(doc_dict)

    return grouped_collection


def _handle_bucket_stage(in_collection, unused_database, options):
    unknown_options = set(options) - {'groupBy', 'boundaries', 'output', 'default'}
    if unknown_options:
        raise OperationFailure(
            'Unrecognized option to $bucket: %s.' % unknown_options.pop())
    if 'groupBy' not in options or 'boundaries' not in options:
        raise OperationFailure(
            "$bucket requires 'groupBy' and 'boundaries' to be specified.")
    group_by = options['groupBy']
    boundaries = options['boundaries']
    if not isinstance(boundaries, list):
        raise OperationFailure(
            "The $bucket 'boundaries' field must be an array, but found type: %s"
            % type(boundaries))
    if len(boundaries) < 2:
        raise OperationFailure(
            "The $bucket 'boundaries' field must have at least 2 values, but "
            'found %d value(s).' % len(boundaries))
    if sorted(boundaries) != boundaries:
        raise OperationFailure(
            "The 'boundaries' option to $bucket must be sorted in ascending order")
    output_fields = options.get('output', {'count': {'$sum': 1}})
    default_value = options.get('default', None)
    try:
        is_default_last = default_value >= boundaries[-1]
    except TypeError:
        is_default_last = True

    def _get_default_bucket():
        try:
            return options['default']
        except KeyError as err:
            raise OperationFailure(
                '$bucket could not find a matching branch for '
                'an input, and no default was specified.') from err

    def _get_bucket_id(doc):
        """Get the bucket ID for a document.

        Note that it actually returns a tuple with the first
        param being a sort key to sort the default bucket even
        if it's not the same type as the boundaries.
        """
        try:
            value = _parse_expression(group_by, doc)
        except KeyError:
            return (is_default_last, _get_default_bucket())
        index = bisect.bisect_right(boundaries, value)
        if index and index < len(boundaries):
            return (False, boundaries[index - 1])
        return (is_default_last, _get_default_bucket())

    in_collection = ((_get_bucket_id(doc), doc) for doc in in_collection)
    out_collection = sorted(in_collection, key=lambda kv: kv[0])
    grouped = itertools.groupby(out_collection, lambda kv: kv[0])

    out_collection = []
    for (unused_key, doc_id), group in grouped:
        group_list = [kv[1] for kv in group]
        doc_dict = _accumulate_group(output_fields, group_list)
        doc_dict['_id'] = doc_id
        out_collection.append(doc_dict)
    return out_collection


def _handle_sample_stage(in_collection, unused_database, options):
    if not isinstance(options, dict):
        raise OperationFailure('the $sample stage specification must be an object')
    size = options.pop('size', None)
    if size is None:
        raise OperationFailure('$sample stage must specify a size')
    if options:
        raise OperationFailure('unrecognized option to $sample: %s' % set(options).pop())
    shuffled = list(in_collection)
    _random.shuffle(shuffled)
    return shuffled[:size]


def _handle_sort_stage(in_collection, unused_database, options):
    sort_array = reversed([{x: y} for x, y in options.items()])
    sorted_collection = in_collection
    for sort_pair in sort_array:
        for sortKey, sortDirection in sort_pair.items():
            sorted_collection = sorted(
                sorted_collection,
                key=lambda x: filtering.resolve_sort_key(sortKey, x),
                reverse=sortDirection < 0)
    return sorted_collection


def _handle_unwind_stage(in_collection, unused_database, options):
    if not isinstance(options, dict):
        options = {'path': options}
    path = options['path']
    if not isinstance(path, str) or path[0] != '$':
        raise ValueError(
            '$unwind failed: exception: field path references must be prefixed '
            "with a '$' '%s'" % path)
    path = path[1:]
    should_preserve_null_and_empty = options.get('preserveNullAndEmptyArrays')
    include_array_index = options.get('includeArrayIndex')
    unwound_collection = []
    for doc in in_collection:
        try:
            array_value = helpers.get_value_by_dot(doc, path)
        except KeyError:
            if should_preserve_null_and_empty:
                unwound_collection.append(doc)
            continue
        if array_value is None:
            if should_preserve_null_and_empty:
                unwound_collection.append(doc)
            continue
        if array_value == []:
            if should_preserve_null_and_empty:
                new_doc = copy.deepcopy(doc)
                # We just ran a get_value_by_dot so we know the value exists.
                helpers.delete_value_by_dot(new_doc, path)
                unwound_collection.append(new_doc)
            continue
        if isinstance(array_value, list):
            iter_array = enumerate(array_value)
        else:
            iter_array = [(None, array_value)]
        for index, field_item in iter_array:
            new_doc = copy.deepcopy(doc)
            new_doc = helpers.set_value_by_dot(new_doc, path, field_item)
            if include_array_index:
                new_doc = helpers.set_value_by_dot(new_doc, include_array_index, index)
            unwound_collection.append(new_doc)

    return unwound_collection


# TODO(pascal): Combine with the equivalent function in collection but check
# what are the allowed overriding.
def _combine_projection_spec(filter_list, original_filter, prefix=''):
    """Re-format a projection fields spec into a nested dictionary.

    e.g: ['a', 'b.c', 'b.d'] => {'a': 1, 'b': {'c': 1, 'd': 1}}
    """
    if not isinstance(filter_list, list):
        return filter_list

    filter_dict = collections.OrderedDict()

    for key in filter_list:
        field, separator, subkey = key.partition('.')
        if not separator:
            if isinstance(filter_dict.get(field), list):
                other_key = field + '.' + filter_dict[field][0]
                raise OperationFailure(
                    'Invalid $project :: caused by :: specification contains two conflicting paths.'
                    ' Cannot specify both %s and %s: %s' % (
                        repr(prefix + field), repr(prefix + other_key), original_filter))
            filter_dict[field] = 1
            continue
        if not isinstance(filter_dict.get(field, []), list):
            raise OperationFailure(
                'Invalid $project :: caused by :: specification contains two conflicting paths.'
                ' Cannot specify both %s and %s: %s' % (
                    repr(prefix + field), repr(prefix + key), original_filter))
        filter_dict[field] = filter_dict.get(field, []) + [subkey]

    return collections.OrderedDict(
        (k, _combine_projection_spec(v, original_filter, prefix='{}{}.'.format(prefix, k)))
        for k, v in filter_dict.items()
    )


def _project_by_spec(doc, proj_spec, is_include):
    output = {}
    for key, value in doc.items():
        if key not in proj_spec:
            if not is_include:
                output[key] = value
            continue

        if not isinstance(proj_spec[key], dict):
            if is_include:
                output[key] = value
            continue

        if isinstance(value, dict):
            output[key] = _project_by_spec(value, proj_spec[key], is_include)
        elif isinstance(value, list):
            output[key] = [_project_by_spec(array_value, proj_spec[key], is_include)
                           for array_value in value if isinstance(array_value, dict)]
        elif not is_include:
            output[key] = value

    return output


def _handle_replace_root_stage(in_collection, unused_database, options):
    if 'newRoot' not in options:
        raise OperationFailure("Parameter 'newRoot' is missing for $replaceRoot operation.")
    new_root = options['newRoot']
    out_collection = []
    for doc in in_collection:
        try:
            new_doc = _parse_expression(new_root, doc, ignore_missing_keys=True)
        except KeyError:
            new_doc = NOTHING
        if not isinstance(new_doc, dict):
            raise OperationFailure(
                "'newRoot' expression must evaluate to an object, but resulting value was: {}"
                .format(new_doc))
        out_collection.append(new_doc)
    return out_collection


def _handle_project_stage(in_collection, unused_database, options):
    filter_list = []
    method = None
    include_id = options.get('_id')
    # Compute new values for each field, except inclusion/exclusions that are
    # handled in one final step.
    new_fields_collection = None
    for field, value in options.items():
        if method is None and (field != '_id' or value):
            method = 'include' if value else 'exclude'
        elif method == 'include' and not value and field != '_id':
            raise OperationFailure(
                'Bad projection specification, cannot exclude fields '
                "other than '_id' in an inclusion projection: %s" % options)
        elif method == 'exclude' and value:
            raise OperationFailure(
                'Bad projection specification, cannot include fields '
                'or add computed fields during an exclusion projection: %s' % options)
        if value in (0, 1, True, False):
            if field != '_id':
                filter_list.append(field)
            continue
        if not new_fields_collection:
            new_fields_collection = [{} for unused_doc in in_collection]

        for in_doc, out_doc in zip(in_collection, new_fields_collection):
            try:
                out_doc[field] = _parse_expression(value, in_doc, ignore_missing_keys=True)
            except KeyError:
                # Ignore missing key.
                pass
    if (method == 'include') == (include_id is not False and include_id != 0):
        filter_list.append('_id')

    if not filter_list:
        return new_fields_collection

    # Final steps: include or exclude fields and merge with newly created fields.
    projection_spec = _combine_projection_spec(filter_list, original_filter=options)
    out_collection = [
        _project_by_spec(doc, projection_spec, is_include=method == 'include')
        for doc in in_collection
    ]
    if new_fields_collection:
        return [
            dict(a, **b)
            for a, b in zip(out_collection, new_fields_collection)
        ]
    return out_collection


def _handle_add_fields_stage(in_collection, unused_database, options):
    if not options:
        raise OperationFailure(
            'Invalid $addFields :: caused by :: specification must have at least one field')
    out_collection = [dict(doc) for doc in in_collection]
    for field, value in options.items():
        for in_doc, out_doc in zip(in_collection, out_collection):
            try:
                out_value = _parse_expression(value, in_doc, ignore_missing_keys=True)
            except KeyError:
                continue
            parts = field.split('.')
            for subfield in parts[:-1]:
                out_doc[subfield] = out_doc.get(subfield, {})
                if not isinstance(out_doc[subfield], dict):
                    out_doc[subfield] = {}
                out_doc = out_doc[subfield]
            out_doc[parts[-1]] = out_value
    return out_collection


def _handle_out_stage(in_collection, database, options):
    # TODO(MetrodataTeam): should leave the origin collection unchanged
    out_collection = database.get_collection(options)
    if out_collection.find_one():
        out_collection.drop()
    if in_collection:
        out_collection.insert_many(in_collection)
    return in_collection


def _handle_count_stage(in_collection, database, options):
    if not isinstance(options, str) or options == '':
        raise OperationFailure('the count field must be a non-empty string')
    elif options.startswith('$'):
        raise OperationFailure('the count field cannot be a $-prefixed path')
    elif '.' in options:
        raise OperationFailure("the count field cannot contain '.'")
    return [{options: len(in_collection)}]


def _handle_facet_stage(in_collection, database, options):
    out_collection_by_pipeline = {}
    for pipeline_title, pipeline in options.items():
        out_collection_by_pipeline[pipeline_title] = list(process_pipeline(
            in_collection, database, pipeline, None))
    return [out_collection_by_pipeline]


def _handle_match_stage(in_collection, database, options):
    spec = helpers.patch_datetime_awareness_in_document(options)
    return [
        doc for doc in in_collection
        if filtering.filter_applies(spec, helpers.patch_datetime_awareness_in_document(doc))
    ]


_PIPELINE_HANDLERS = {
    '$addFields': _handle_add_fields_stage,
    '$bucket': _handle_bucket_stage,
    '$bucketAuto': None,
    '$collStats': None,
    '$count': _handle_count_stage,
    '$currentOp': None,
    '$facet': _handle_facet_stage,
    '$geoNear': None,
    '$graphLookup': _handle_graph_lookup_stage,
    '$group': _handle_group_stage,
    '$indexStats': None,
    '$limit': lambda c, d, o: c[:o],
    '$listLocalSessions': None,
    '$listSessions': None,
    '$lookup': _handle_lookup_stage,
    '$match': _handle_match_stage,
    '$merge': None,
    '$out': _handle_out_stage,
    '$planCacheStats': None,
    '$project': _handle_project_stage,
    '$redact': None,
    '$replaceRoot': _handle_replace_root_stage,
    '$replaceWith': None,
    '$sample': _handle_sample_stage,
    '$set': _handle_add_fields_stage,
    '$skip': lambda c, d, o: c[o:],
    '$sort': _handle_sort_stage,
    '$sortByCount': None,
    '$unset': None,
    '$unwind': _handle_unwind_stage,
}


def process_pipeline(collection, database, pipeline, session):
    if session:
        raise NotImplementedError('Mongomock does not handle sessions yet')

    for stage in pipeline:
        for operator, options in stage.items():
            try:
                handler = _PIPELINE_HANDLERS[operator]
            except KeyError as err:
                raise NotImplementedError(
                    '%s is not a valid operator for the aggregation pipeline. '
                    'See http://docs.mongodb.org/manual/meta/aggregation-quick-reference/ '
                    'for a complete list of valid operators.' % operator) from err
            if not handler:
                raise NotImplementedError(
                    "Although '%s' is a valid operator for the aggregation pipeline, it is "
                    'currently not implemented in Mongomock.' % operator)
            collection = handler(collection, database, options)

    return command_cursor.CommandCursor(collection)


def validate_stage_name(name):
    if name not in _PIPELINE_HANDLERS:
        raise OperationFailure("Unrecognized pipeline stage name: '%s'" % name)