/django/db/models/sql/compiler.py

import re
from itertools import chain

from django.core.exceptions import FieldError
from django.db.models.constants import LOOKUP_SEP
from django.db.models.expressions import OrderBy, Random, RawSQL, Ref
from django.db.models.query_utils import QueryWrapper, select_related_descend
from django.db.models.sql.constants import (
    CURSOR, GET_ITERATOR_CHUNK_SIZE, MULTI, NO_RESULTS, ORDER_DIR, SINGLE,
)
from django.db.models.sql.datastructures import EmptyResultSet
from django.db.models.sql.query import Query, get_order_dir
from django.db.transaction import TransactionManagementError
from django.db.utils import DatabaseError
from django.utils.six.moves import zip


class SQLCompiler(object):
    def __init__(self, query, connection, using):
        self.query = query
        self.connection = connection
        self.using = using
        self.quote_cache = {'*': '*'}
        # The select, klass_info, and annotations are needed by QuerySet.iterator()
        # these are set as a side-effect of executing the query. Note that we calculate
        # separately a list of extra select columns needed for grammatical correctness
        # of the query, but these columns are not included in self.select.
        self.select = None
        self.annotation_col_map = None
        self.klass_info = None
        self.ordering_parts = re.compile(r'(.*)\s(ASC|DESC)(.*)')
        self.subquery = False

    def setup_query(self):
        if all(self.query.alias_refcount[a] == 0 for a in self.query.tables):
            self.query.get_initial_alias()
        self.select, self.klass_info, self.annotation_col_map = self.get_select()
        self.col_count = len(self.select)

    def pre_sql_setup(self):
        """
        Does any necessary class setup immediately prior to producing SQL. This
        is for things that can't necessarily be done in __init__ because we
        might not have all the pieces in place at that time.
        """
        self.setup_query()
        order_by = self.get_order_by()
        self.where, self.having = self.query.where.split_having()
        extra_select = self.get_extra_select(order_by, self.select)
        group_by = self.get_group_by(self.select + extra_select, order_by)
        return extra_select, order_by, group_by

    def get_group_by(self, select, order_by):
        """
        Returns a list of 2-tuples of form (sql, params).

        The logic of what exactly the GROUP BY clause contains is hard
        to describe in other words than "if it passes the test suite,
        then it is correct".
        """
        # Some examples:
        #     SomeModel.objects.annotate(Count('somecol'))
        #     GROUP BY: all fields of the model
        #
        #    SomeModel.objects.values('name').annotate(Count('somecol'))
        #    GROUP BY: name
        #
        #    SomeModel.objects.annotate(Count('somecol')).values('name')
        #    GROUP BY: all cols of the model
        #
        #    SomeModel.objects.values('name', 'pk').annotate(Count('somecol')).values('pk')
        #    GROUP BY: name, pk
        #
        #    SomeModel.objects.values('name').annotate(Count('somecol')).values('pk')
        #    GROUP BY: name, pk
        #
        # In fact, the self.query.group_by is the minimal set to GROUP BY. It
        # can't be ever restricted to a smaller set, but additional columns in
        # HAVING, ORDER BY, and SELECT clauses are added to it. Unfortunately
        # the end result is that it is impossible to force the query to have
        # a chosen GROUP BY clause - you can almost do this by using the form:
        #     .values(*wanted_cols).annotate(AnAggregate())
        # but any later annotations, extra selects, values calls that
        # refer some column outside of the wanted_cols, order_by, or even
        # filter calls can alter the GROUP BY clause.

        # The query.group_by is either None (no GROUP BY at all), True
        # (group by select fields), or a list of expressions to be added
        # to the group by.
        if self.query.group_by is None:
            return []
        expressions = []
        if self.query.group_by is not True:
            # If the group by is set to a list (by .values() call most likely),
            # then we need to add everything in it to the GROUP BY clause.
            # Backwards compatibility hack for setting query.group_by. Remove
            # when  we have public API way of forcing the GROUP BY clause.
            # Converts string references to expressions.
            for expr in self.query.group_by:
                if not hasattr(expr, 'as_sql'):
                    expressions.append(self.query.resolve_ref(expr))
                else:
                    expressions.append(expr)
        # Note that even if the group_by is set, it is only the minimal
        # set to group by. So, we need to add cols in select, order_by, and
        # having into the select in any case.
        for expr, _, _ in select:
            cols = expr.get_group_by_cols()
            for col in cols:
                expressions.append(col)
        for expr, (sql, params, is_ref) in order_by:
            if expr.contains_aggregate:
                continue
            # We can skip References to select clause, as all expressions in
            # the select clause are already part of the group by.
            if is_ref:
                continue
            expressions.extend(expr.get_source_expressions())
        having_group_by = self.having.get_group_by_cols() if self.having else ()
        for expr in having_group_by:
            expressions.append(expr)
        result = []
        seen = set()
        expressions = self.collapse_group_by(expressions, having_group_by)

        for expr in expressions:
            sql, params = self.compile(expr)
            if (sql, tuple(params)) not in seen:
                result.append((sql, params))
                seen.add((sql, tuple(params)))
        return result

    def collapse_group_by(self, expressions, having):
        # If the DB can group by primary key, then group by the primary key of
        # query's main model. Note that for PostgreSQL the GROUP BY clause must
        # include the primary key of every table, but for MySQL it is enough to
        # have the main table's primary key.
        if self.connection.features.allows_group_by_pk:
            # The logic here is: if the main model's primary key is in the
            # query, then set new_expressions to that field. If that happens,
            # then also add having expressions to group by.
            pk = None
            for expr in expressions:
                # Is this a reference to query's base table primary key? If the
                # expression isn't a Col-like, then skip the expression.
                if (getattr(expr, 'target', None) == self.query.model._meta.pk and
                        getattr(expr, 'alias', None) == self.query.tables[0]):
                    pk = expr
                    break
            if pk:
                # MySQLism: Columns in HAVING clause must be added to the GROUP BY.
                expressions = [pk] + [expr for expr in expressions if expr in having]
        elif self.connection.features.allows_group_by_selected_pks:
            # Filter out all expressions associated with a table's primary key
            # present in the grouped columns. This is done by identifying all
            # tables that have their primary key included in the grouped
            # columns and removing non-primary key columns referring to them.
            pks = {expr for expr in expressions if hasattr(expr, 'target') and expr.target.primary_key}
            aliases = {expr.alias for expr in pks}
            expressions = [
                expr for expr in expressions if expr in pks or getattr(expr, 'alias', None) not in aliases
            ]
        return expressions

    def get_select(self):
        """
        Returns three values:
        - a list of 3-tuples of (expression, (sql, params), alias)
        - a klass_info structure,
        - a dictionary of annotations

        The (sql, params) is what the expression will produce, and alias is the
        "AS alias" for the column (possibly None).

        The klass_info structure contains the following information:
        - Which model to instantiate
        - Which columns for that model are present in the query (by
          position of the select clause).
        - related_klass_infos: [f, klass_info] to descent into

        The annotations is a dictionary of {'attname': column position} values.
        """
        select = []
        klass_info = None
        annotations = {}
        select_idx = 0
        for alias, (sql, params) in self.query.extra_select.items():
            annotations[alias] = select_idx
            select.append((RawSQL(sql, params), alias))
            select_idx += 1
        assert not (self.query.select and self.query.default_cols)
        if self.query.default_cols:
            select_list = []
            for c in self.get_default_columns():
                select_list.append(select_idx)
                select.append((c, None))
                select_idx += 1
            klass_info = {
                'model': self.query.model,
                'select_fields': select_list,
            }
        # self.query.select is a special case. These columns never go to
        # any model.
        for col in self.query.select:
            select.append((col, None))
            select_idx += 1
        for alias, annotation in self.query.annotation_select.items():
            annotations[alias] = select_idx
            select.append((annotation, alias))
            select_idx += 1

        if self.query.select_related:
            related_klass_infos = self.get_related_selections(select)
            klass_info['related_klass_infos'] = related_klass_infos

            def get_select_from_parent(klass_info):
                for ki in klass_info['related_klass_infos']:
                    if ki['from_parent']:
                        ki['select_fields'] = (klass_info['select_fields'] +
                                               ki['select_fields'])
                    get_select_from_parent(ki)
            get_select_from_parent(klass_info)

        ret = []
        for col, alias in select:
            ret.append((col, self.compile(col, select_format=True), alias))
        return ret, klass_info, annotations

    def get_order_by(self):
        """
        Returns a list of 2-tuples of form (expr, (sql, params, is_ref)) for the
        ORDER BY clause.

        The order_by clause can alter the select clause (for example it
        can add aliases to clauses that do not yet have one, or it can
        add totally new select clauses).
        """
        if self.query.extra_order_by:
            ordering = self.query.extra_order_by
        elif not self.query.default_ordering:
            ordering = self.query.order_by
        else:
            ordering = (self.query.order_by or self.query.get_meta().ordering or [])
        if self.query.standard_ordering:
            asc, desc = ORDER_DIR['ASC']
        else:
            asc, desc = ORDER_DIR['DESC']

        order_by = []
        for pos, field in enumerate(ordering):
            if hasattr(field, 'resolve_expression'):
                if not isinstance(field, OrderBy):
                    field = field.asc()
                if not self.query.standard_ordering:
                    field.reverse_ordering()
                order_by.append((field, False))
                continue
            if field == '?':  # random
                order_by.append((OrderBy(Random()), False))
                continue

            col, order = get_order_dir(field, asc)
            descending = True if order == 'DESC' else False

            if col in self.query.annotation_select:
                # Reference to expression in SELECT clause
                order_by.append((
                    OrderBy(Ref(col, self.query.annotation_select[col]), descending=descending),
                    True))
                continue
            if col in self.query.annotations:
                # References to an expression which is masked out of the SELECT clause
                order_by.append((
                    OrderBy(self.query.annotations[col], descending=descending),
                    False))
                continue

            if '.' in field:
                # This came in through an extra(order_by=...) addition. Pass it
                # on verbatim.
                table, col = col.split('.', 1)
                order_by.append((
                    OrderBy(
                        RawSQL('%s.%s' % (self.quote_name_unless_alias(table), col), []),
                        descending=descending
                    ), False))
                continue

            if not self.query._extra or col not in self.query._extra:
                # 'col' is of the form 'field' or 'field1__field2' or
                # '-field1__field2__field', etc.
                order_by.extend(self.find_ordering_name(
                    field, self.query.get_meta(), default_order=asc))
            else:
                if col not in self.query.extra_select:
                    order_by.append((
                        OrderBy(RawSQL(*self.query.extra[col]), descending=descending),
                        False))
                else:
                    order_by.append((
                        OrderBy(Ref(col, RawSQL(*self.query.extra[col])), descending=descending),
                        True))
        result = []
        seen = set()

        for expr, is_ref in order_by:
            resolved = expr.resolve_expression(
                self.query, allow_joins=True, reuse=None)
            sql, params = self.compile(resolved)
            # Don't add the same column twice, but the order direction is
            # not taken into account so we strip it. When this entire method
            # is refactored into expressions, then we can check each part as we
            # generate it.
            without_ordering = self.ordering_parts.search(sql).group(1)
            if (without_ordering, tuple(params)) in seen:
                continue
            seen.add((without_ordering, tuple(params)))
            result.append((resolved, (sql, params, is_ref)))
        return result

    def get_extra_select(self, order_by, select):
        extra_select = []
        select_sql = [t[1] for t in select]
        if self.query.distinct and not self.query.distinct_fields:
            for expr, (sql, params, is_ref) in order_by:
                without_ordering = self.ordering_parts.search(sql).group(1)
                if not is_ref and (without_ordering, params) not in select_sql:
                    extra_select.append((expr, (without_ordering, params), None))
        return extra_select

    def quote_name_unless_alias(self, name):
        """
        A wrapper around connection.ops.quote_name that doesn't quote aliases
        for table names. This avoids problems with some SQL dialects that treat
        quoted strings specially (e.g. PostgreSQL).
        """
        if name in self.quote_cache:
            return self.quote_cache[name]
        if ((name in self.query.alias_map and name not in self.query.table_map) or
                name in self.query.extra_select or (
                    name in self.query.external_aliases and name not in self.query.table_map)):
            self.quote_cache[name] = name
            return name
        r = self.connection.ops.quote_name(name)
        self.quote_cache[name] = r
        return r

    def compile(self, node, select_format=False):
        vendor_impl = getattr(node, 'as_' + self.connection.vendor, None)
        if vendor_impl:
            sql, params = vendor_impl(self, self.connection)
        else:
            sql, params = node.as_sql(self, self.connection)
        if select_format and not self.subquery:
            return node.output_field.select_format(self, sql, params)
        return sql, params

    def as_sql(self, with_limits=True, with_col_aliases=False, subquery=False):
        """
        Creates the SQL for this query. Returns the SQL string and list of
        parameters.

        If 'with_limits' is False, any limit/offset information is not included
        in the query.
        """
        self.subquery = subquery
        refcounts_before = self.query.alias_refcount.copy()
        try:
            extra_select, order_by, group_by = self.pre_sql_setup()
            distinct_fields = self.get_distinct()

            # This must come after 'select', 'ordering', and 'distinct' -- see
            # docstring of get_from_clause() for details.
            from_, f_params = self.get_from_clause()

            where, w_params = self.compile(self.where) if self.where is not None else ("", [])
            having, h_params = self.compile(self.having) if self.having is not None else ("", [])
            params = []
            result = ['SELECT']

            if self.query.distinct:
                result.append(self.connection.ops.distinct_sql(distinct_fields))

            out_cols = []
            col_idx = 1
            for _, (s_sql, s_params), alias in self.select + extra_select:
                if alias:
                    s_sql = '%s AS %s' % (s_sql, self.connection.ops.quote_name(alias))
                elif with_col_aliases:
                    s_sql = '%s AS %s' % (s_sql, 'Col%d' % col_idx)
                    col_idx += 1
                params.extend(s_params)
                out_cols.append(s_sql)

            result.append(', '.join(out_cols))

            result.append('FROM')
            result.extend(from_)
            params.extend(f_params)

            if where:
                result.append('WHERE %s' % where)
                params.extend(w_params)

            grouping = []
            for g_sql, g_params in group_by:
                grouping.append(g_sql)
                params.extend(g_params)
            if grouping:
                if distinct_fields:
                    raise NotImplementedError(
                        "annotate() + distinct(fields) is not implemented.")
                if not order_by:
                    order_by = self.connection.ops.force_no_ordering()
                result.append('GROUP BY %s' % ', '.join(grouping))

            if having:
                result.append('HAVING %s' % having)
                params.extend(h_params)

            if order_by:
                ordering = []
                for _, (o_sql, o_params, _) in order_by:
                    ordering.append(o_sql)
                    params.extend(o_params)
                result.append('ORDER BY %s' % ', '.join(ordering))

            if with_limits:
                if self.query.high_mark is not None:
                    result.append('LIMIT %d' % (self.query.high_mark - self.query.low_mark))
                if self.query.low_mark:
                    if self.query.high_mark is None:
                        val = self.connection.ops.no_limit_value()
                        if val:
                            result.append('LIMIT %d' % val)
                    result.append('OFFSET %d' % self.query.low_mark)

            if self.query.select_for_update and self.connection.features.has_select_for_update:
                if self.connection.get_autocommit():
                    raise TransactionManagementError(
                        "select_for_update cannot be used outside of a transaction."
                    )

                # If we've been asked for a NOWAIT query but the backend does
                # not support it, raise a DatabaseError otherwise we could get
                # an unexpected deadlock.
                nowait = self.query.select_for_update_nowait
                if nowait and not self.connection.features.has_select_for_update_nowait:
                    raise DatabaseError('NOWAIT is not supported on this database backend.')
                result.append(self.connection.ops.for_update_sql(nowait=nowait))

            return ' '.join(result), tuple(params)
        finally:
            # Finally do cleanup - get rid of the joins we created above.
            self.query.reset_refcounts(refcounts_before)

    def as_nested_sql(self):
        """
        Perform the same functionality as the as_sql() method, returning an
        SQL string and parameters. However, the alias prefixes are bumped
        beforehand (in a copy -- the current query isn't changed), and any
        ordering is removed if the query is unsliced.

        Used when nesting this query inside another.
        """
        obj = self.query.clone()
        if obj.low_mark == 0 and obj.high_mark is None and not self.query.distinct_fields:
            # If there is no slicing in use, then we can safely drop all ordering
            obj.clear_ordering(True)
        nested_sql = obj.get_compiler(connection=self.connection).as_sql(subquery=True)
        if nested_sql == ('', ()):
            raise EmptyResultSet
        return nested_sql

    def get_default_columns(self, start_alias=None, opts=None, from_parent=None):
        """
        Computes the default columns for selecting every field in the base
        model. Will sometimes be called to pull in related models (e.g. via
        select_related), in which case "opts" and "start_alias" will be given
        to provide a starting point for the traversal.

        Returns a list of strings, quoted appropriately for use in SQL
        directly, as well as a set of aliases used in the select statement (if
        'as_pairs' is True, returns a list of (alias, col_name) pairs instead
        of strings as the first component and None as the second component).
        """
        result = []
        if opts is None:
            opts = self.query.get_meta()
        only_load = self.deferred_to_columns()
        if not start_alias:
            start_alias = self.query.get_initial_alias()
        # The 'seen_models' is used to optimize checking the needed parent
        # alias for a given field. This also includes None -> start_alias to
        # be used by local fields.
        seen_models = {None: start_alias}

        for field in opts.concrete_fields:
            model = field.model._meta.concrete_model
            # A proxy model will have a different model and concrete_model. We
            # will assign None if the field belongs to this model.
            if model == opts.model:
                model = None
            if from_parent and model is not None and issubclass(
                    from_parent._meta.concrete_model, model._meta.concrete_model):
                # Avoid loading data for already loaded parents.
                # We end up here in the case select_related() resolution
                # proceeds from parent model to child model. In that case the
                # parent model data is already present in the SELECT clause,
                # and we want to avoid reloading the same data again.
                continue
            if field.model in only_load and field.attname not in only_load[field.model]:
                continue
            alias = self.query.join_parent_model(opts, model, start_alias,
                                                 seen_models)
            column = field.get_col(alias)
            result.append(column)
        return result

    def get_distinct(self):
        """
        Returns a quoted list of fields to use in DISTINCT ON part of the query.

        Note that this method can alter the tables in the query, and thus it
        must be called before get_from_clause().
        """
        qn = self.quote_name_unless_alias
        qn2 = self.connection.ops.quote_name
        result = []
        opts = self.query.get_meta()

        for name in self.query.distinct_fields:
            parts = name.split(LOOKUP_SEP)
            _, targets, alias, joins, path, _ = self._setup_joins(parts, opts, None)
            targets, alias, _ = self.query.trim_joins(targets, joins, path)
            for target in targets:
                if name in self.query.annotation_select:
                    result.append(name)
                else:
                    result.append("%s.%s" % (qn(alias), qn2(target.column)))
        return result

    def find_ordering_name(self, name, opts, alias=None, default_order='ASC',
                           already_seen=None):
        """
        Returns the table alias (the name might be ambiguous, the alias will
        not be) and column name for ordering by the given 'name' parameter.
        The 'name' is of the form 'field1__field2__...__fieldN'.
        """
        name, order = get_order_dir(name, default_order)
        descending = True if order == 'DESC' else False
        pieces = name.split(LOOKUP_SEP)
        field, targets, alias, joins, path, opts = self._setup_joins(pieces, opts, alias)

        # If we get to this point and the field is a relation to another model,
        # append the default ordering for that model unless the attribute name
        # of the field is specified.
        if field.is_relation and path and opts.ordering and name != field.attname:
            # Firstly, avoid infinite loops.
            if not already_seen:
                already_seen = set()
            join_tuple = tuple(getattr(self.query.alias_map[j], 'join_cols', None) for j in joins)
            if join_tuple in already_seen:
                raise FieldError('Infinite loop caused by ordering.')
            already_seen.add(join_tuple)

            results = []
            for item in opts.ordering:
                results.extend(self.find_ordering_name(item, opts, alias,
                                                       order, already_seen))
            return results
        targets, alias, _ = self.query.trim_joins(targets, joins, path)
        return [(OrderBy(t.get_col(alias), descending=descending), False) for t in targets]

    def _setup_joins(self, pieces, opts, alias):
        """
        A helper method for get_order_by and get_distinct.

        Note that get_ordering and get_distinct must produce same target
        columns on same input, as the prefixes of get_ordering and get_distinct
        must match. Executing SQL where this is not true is an error.
        """
        if not alias:
            alias = self.query.get_initial_alias()
        field, targets, opts, joins, path = self.query.setup_joins(
            pieces, opts, alias)
        alias = joins[-1]
        return field, targets, alias, joins, path, opts

    def get_from_clause(self):
        """
        Returns a list of strings that are joined together to go after the
        "FROM" part of the query, as well as a list any extra parameters that
        need to be included. Sub-classes, can override this to create a
        from-clause via a "select".

        This should only be called after any SQL construction methods that
        might change the tables we need. This means the select columns,
        ordering and distinct must be done first.
        """
        result = []
        params = []
        for alias in self.query.tables:
            if not self.query.alias_refcount[alias]:
                continue
            try:
                from_clause = self.query.alias_map[alias]
            except KeyError:
                # Extra tables can end up in self.tables, but not in the
                # alias_map if they aren't in a join. That's OK. We skip them.
                continue
            clause_sql, clause_params = self.compile(from_clause)
            result.append(clause_sql)
            params.extend(clause_params)
        for t in self.query.extra_tables:
            alias, _ = self.query.table_alias(t)
            # Only add the alias if it's not already present (the table_alias()
            # call increments the refcount, so an alias refcount of one means
            # this is the only reference).
            if alias not in self.query.alias_map or self.query.alias_refcount[alias] == 1:
                result.append(', %s' % self.quote_name_unless_alias(alias))
        return result, params

    def get_related_selections(self, select, opts=None, root_alias=None, cur_depth=1,
                               requested=None, restricted=None):
        """
        Fill in the information needed for a select_related query. The current
        depth is measured as the number of connections away from the root model
        (for example, cur_depth=1 means we are looking at models with direct
        connections to the root model).
        """
        def _get_field_choices():
            direct_choices = (f.name for f in opts.fields if f.is_relation)
            reverse_choices = (
                f.field.related_query_name()
                for f in opts.related_objects if f.field.unique
            )
            return chain(direct_choices, reverse_choices)

        related_klass_infos = []
        if not restricted and self.query.max_depth and cur_depth > self.query.max_depth:
            # We've recursed far enough; bail out.
            return related_klass_infos

        if not opts:
            opts = self.query.get_meta()
            root_alias = self.query.get_initial_alias()
        only_load = self.query.get_loaded_field_names()

        # Setup for the case when only particular related fields should be
        # included in the related selection.
        fields_found = set()
        if requested is None:
            if isinstance(self.query.select_related, dict):
                requested = self.query.select_related
                restricted = True
            else:
                restricted = False

        def get_related_klass_infos(klass_info, related_klass_infos):
            klass_info['related_klass_infos'] = related_klass_infos

        for f in opts.fields:
            field_model = f.model._meta.concrete_model
            fields_found.add(f.name)

            if restricted:
                next = requested.get(f.name, {})
                if not f.is_relation:
                    # If a non-related field is used like a relation,
                    # or if a single non-relational field is given.
                    if next or f.name in requested:
                        raise FieldError(
                            "Non-relational field given in select_related: '%s'. "
                            "Choices are: %s" % (
                                f.name,
                                ", ".join(_get_field_choices()) or '(none)',
                            )
                        )
            else:
                next = False

            if not select_related_descend(f, restricted, requested,
                                          only_load.get(field_model)):
                continue
            klass_info = {
                'model': f.remote_field.model,
                'field': f,
                'reverse': False,
                'from_parent': False,
            }
            related_klass_infos.append(klass_info)
            select_fields = []
            _, _, _, joins, _ = self.query.setup_joins(
                [f.name], opts, root_alias)
            alias = joins[-1]
            columns = self.get_default_columns(start_alias=alias, opts=f.remote_field.model._meta)
            for col in columns:
                select_fields.append(len(select))
                select.append((col, None))
            klass_info['select_fields'] = select_fields
            next_klass_infos = self.get_related_selections(
                select, f.remote_field.model._meta, alias, cur_depth + 1, next, restricted)
            get_related_klass_infos(klass_info, next_klass_infos)

        if restricted:
            related_fields = [
                (o.field, o.related_model)
                for o in opts.related_objects
                if o.field.unique and not o.many_to_many
            ]
            for f, model in related_fields:
                if not select_related_descend(f, restricted, requested,
                                              only_load.get(model), reverse=True):
                    continue

                related_field_name = f.related_query_name()
                fields_found.add(related_field_name)

                _, _, _, joins, _ = self.query.setup_joins([related_field_name], opts, root_alias)
                alias = joins[-1]
                from_parent = issubclass(model, opts.model)
                klass_info = {
                    'model': model,
                    'field': f,
                    'reverse': True,
                    'from_parent': from_parent,
                }
                related_klass_infos.append(klass_info)
                select_fields = []
                columns = self.get_default_columns(
                    start_alias=alias, opts=model._meta, from_parent=opts.model)
                for col in columns:
                    select_fields.append(len(select))
                    select.append((col, None))
                klass_info['select_fields'] = select_fields
                next = requested.get(f.related_query_name(), {})
                next_klass_infos = self.get_related_selections(
                    select, model._meta, alias, cur_depth + 1,
                    next, restricted)
                get_related_klass_infos(klass_info, next_klass_infos)
            fields_not_found = set(requested.keys()).difference(fields_found)
            if fields_not_found:
                invalid_fields = ("'%s'" % s for s in fields_not_found)
                raise FieldError(
                    'Invalid field name(s) given in select_related: %s. '
                    'Choices are: %s' % (
                        ', '.join(invalid_fields),
                        ', '.join(_get_field_choices()) or '(none)',
                    )
                )
        return related_klass_infos

    def deferred_to_columns(self):
        """
        Converts the self.deferred_loading data structure to mapping of table
        names to sets of column names which are to be loaded. Returns the
        dictionary.
        """
        columns = {}
        self.query.deferred_to_data(columns, self.query.get_loaded_field_names_cb)
        return columns

    def get_converters(self, expressions):
        converters = {}
        for i, expression in enumerate(expressions):
            if expression:
                backend_converters = self.connection.ops.get_db_converters(expression)
                field_converters = expression.get_db_converters(self.connection)
                if backend_converters or field_converters:
                    converters[i] = (backend_converters + field_converters, expression)
        return converters

    def apply_converters(self, row, converters):
        row = list(row)
        for pos, (convs, expression) in converters.items():
            value = row[pos]
            for converter in convs:
                value = converter(value, expression, self.connection, self.query.context)
            row[pos] = value
        return tuple(row)

    def results_iter(self, results=None):
        """
        Returns an iterator over the results from executing this query.
        """
        converters = None
        if results is None:
            results = self.execute_sql(MULTI)
        fields = [s[0] for s in self.select[0:self.col_count]]
        converters = self.get_converters(fields)
        for rows in results:
            for row in rows:
                if converters:
                    row = self.apply_converters(row, converters)
                yield row

    def has_results(self):
        """
        Backends (e.g. NoSQL) can override this in order to use optimized
        versions of "query has any results."
        """
        # This is always executed on a query clone, so we can modify self.query
        self.query.add_extra({'a': 1}, None, None, None, None, None)
        self.query.set_extra_mask(['a'])
        return bool(self.execute_sql(SINGLE))

    def execute_sql(self, result_type=MULTI):
        """
        Run the query against the database and returns the result(s). The
        return value is a single data item if result_type is SINGLE, or an
        iterator over the results if the result_type is MULTI.

        result_type is either MULTI (use fetchmany() to retrieve all rows),
        SINGLE (only retrieve a single row), or None. In this last case, the
        cursor is returned if any query is executed, since it's used by
        subclasses such as InsertQuery). It's possible, however, that no query
        is needed, as the filters describe an empty set. In that case, None is
        returned, to avoid any unnecessary database interaction.
        """
        if not result_type:
            result_type = NO_RESULTS
        try:
            sql, params = self.as_sql()
            if not sql:
                raise EmptyResultSet
        except EmptyResultSet:
            if result_type == MULTI:
                return iter([])
            else:
                return

        cursor = self.connection.cursor()
        try:
            cursor.execute(sql, params)
        except Exception:
            cursor.close()
            raise

        if result_type == CURSOR:
            # Caller didn't specify a result_type, so just give them back the
            # cursor to process (and close).
            return cursor
        if result_type == SINGLE:
            try:
                val = cursor.fetchone()
                if val:
                    return val[0:self.col_count]
                return val
            finally:
                # done with the cursor
                cursor.close()
        if result_type == NO_RESULTS:
            cursor.close()
            return

        result = cursor_iter(
            cursor, self.connection.features.empty_fetchmany_value,
            self.col_count
        )
        if not self.connection.features.can_use_chunked_reads:
            try:
                # If we are using non-chunked reads, we return the same data
                # structure as normally, but ensure it is all read into memory
                # before going any further.
                return list(result)
            finally:
                # done with the cursor
                cursor.close()
        return result

    def as_subquery_condition(self, alias, columns, compiler):
        qn = compiler.quote_name_unless_alias
        qn2 = self.connection.ops.quote_name
        if len(columns) == 1:
            sql, params = self.as_sql()
            return '%s.%s IN (%s)' % (qn(alias), qn2(columns[0]), sql), params

        for index, select_col in enumerate(self.query.select):
            lhs_sql, lhs_params = self.compile(select_col)
            rhs = '%s.%s' % (qn(alias), qn2(columns[index]))
            self.query.where.add(
                QueryWrapper('%s = %s' % (lhs_sql, rhs), lhs_params), 'AND')

        sql, params = self.as_sql()
        return 'EXISTS (%s)' % sql, params


class SQLInsertCompiler(SQLCompiler):

    def __init__(self, *args, **kwargs):
        self.return_id = False
        super(SQLInsertCompiler, self).__init__(*args, **kwargs)

    def field_as_sql(self, field, val):
        """
        Take a field and a value intended to be saved on that field, and
        return placeholder SQL and accompanying params. Checks for raw values,
        expressions and fields with get_placeholder() defined in that order.

        When field is None, the value is considered raw and is used as the
        placeholder, with no corresponding parameters returned.
        """
        if field is None:
            # A field value of None means the value is raw.
            sql, params = val, []
        elif hasattr(val, 'as_sql'):
            # This is an expression, let's compile it.
            sql, params = self.compile(val)
        elif hasattr(field, 'get_placeholder'):
            # Some fields (e.g. geo fields) need special munging before
            # they can be inserted.
            sql, params = field.get_placeholder(val, self, self.connection), [val]
        else:
            # Return the common case for the placeholder
            sql, params = '%s', [val]

        # The following hook is only used by Oracle Spatial, which sometimes
        # needs to yield 'NULL' and [] as its placeholder and params instead
        # of '%s' and [None]. The 'NULL' placeholder is produced earlier by
        # OracleOperations.get_geom_placeholder(). The following line removes
        # the corresponding None parameter. See ticket #10888.
        params = self.connection.ops.modify_insert_params(sql, params)

        return sql, params

    def prepare_value(self, field, value):
        """
        Prepare a value to be used in a query by resolving it if it is an
        expression and otherwise calling the field's get_db_prep_save().
        """
        if hasattr(value, 'resolve_expression'):
            value = value.resolve_expression(self.query, allow_joins=False, for_save=True)
            # Don't allow values containing Col expressions. They refer to
            # existing columns on a row, but in the case of insert the row
            # doesn't exist yet.
            if value.contains_column_references:
                raise ValueError(
                    'Failed to insert expression "%s" on %s. F() expressions '
                    'can only be used to update, not to insert.' % (value, field)
                )
            if value.contains_aggregate:
                raise FieldError("Aggregate functions are not allowed in this query")
        else:
            value = field.get_db_prep_save(value, connection=self.connection)
        return value

    def pre_save_val(self, field, obj):
        """
        Get the given field's value off the given obj. pre_save() is used for
        things like auto_now on DateTimeField. Skip it if this is a raw query.
        """
        if self.query.raw:
            return getattr(obj, field.attname)
        return field.pre_save(obj, add=True)

    def assemble_as_sql(self, fields, value_rows):
        """
        Take a sequence of N fields and a sequence of M rows of values,
        generate placeholder SQL and parameters for each field and value, and
        return a pair containing:
         * a sequence of M rows of N SQL placeholder strings, and
         * a sequence of M rows of corresponding parameter values.

        Each placeholder string may contain any number of '%s' interpolation
        strings, and each parameter row will contain exactly as many params
        as the total number of '%s's in the corresponding placeholder row.
        """
        if not value_rows:
            return [], []

        # list of (sql, [params]) tuples for each object to be saved
        # Shape: [n_objs][n_fields][2]
        rows_of_fields_as_sql = (
            (self.field_as_sql(field, v) for field, v in zip(fields, row))
            for row in value_rows
        )

        # tuple like ([sqls], [[params]s]) for each object to be saved
        # Shape: [n_objs][2][n_fields]
        sql_and_param_pair_rows = (zip(*row) for row in rows_of_fields_as_sql)

        # Extract separate lists for placeholders and params.
        # Each of these has shape [n_objs][n_fields]
        placeholder_rows, param_rows = zip(*sql_and_param_pair_rows)

        # Params for each field are still lists, and need to be flattened.
        param_rows = [[p for ps in row for p in ps] for row in param_rows]

        return placeholder_rows, param_rows

    def as_sql(self):
        # We don't need quote_name_unless_alias() here, since these are all
        # going to be column names (so we can avoid the extra overhead).
        qn = self.connection.ops.quote_name
        opts = self.query.get_meta()
        result = ['INSERT INTO %s' % qn(opts.db_table)]

        has_fields = bool(self.query.fields)
        fields = self.query.fields if has_fields else [opts.pk]
        result.append('(%s)' % ', '.join(qn(f.column) for f in fields))

        if has_fields:
            value_rows = [
                [self.prepare_value(field, self.pre_save_val(field, obj)) for field in fields]
                for obj in self.query.objs
            ]
        else:
            # An empty object.
            value_rows = [[self.connection.ops.pk_default_value()] for _ in self.query.objs]
            fields = [None]

        # Currently the backends just accept values when generating bulk
        # queries and generate their own placeholders. Doing that isn't
        # necessary and it should be possible to use placeholders and
        # expressions in bulk inserts too.
        can_bulk = (not self.return_id and self.connection.features.has_bulk_insert)

        placeholder_rows, param_rows = self.assemble_as_sql(fields, value_rows)

        if self.return_id and self.connection.features.can_return_id_from_insert:
            params = param_rows[0]
            col = "%s.%s" % (qn(opts.db_table), qn(opts.pk.column))
            result.append("VALUES (%s)" % ", ".join(placeholder_rows[0]))
            r_fmt, r_params = self.connection.ops.return_insert_id()
            # Skip empty r_fmt to allow subclasses to customize behavior for
            # 3rd party backends. Refs #19096.
            if r_fmt:
                result.append(r_fmt % col)
                params += r_params
            return [(" ".join(result), tuple(params))]

        if can_bulk:
            result.append(self.connection.ops.bulk_insert_sql(fields, placeholder_rows))
            return [(" ".join(result), tuple(p for ps in param_rows for p in ps))]
        else:
            return [
                (" ".join(result + ["VALUES (%s)" % ", ".join(p)]), vals)
                for p, vals in zip(placeholder_rows, param_rows)
            ]

    def execute_sql(self, return_id=False):
        assert not (return_id and len(self.query.objs) != 1)
        self.return_id = return_id
        with self.connection.cursor() as cursor:
            for sql, params in self.as_sql():
                cursor.execute(sql, params)
            if not (return_id and cursor):
                return
            if self.connection.features.can_return_id_from_insert:
                return self.connection.ops.fetch_returned_insert_id(cursor)
            return self.connection.ops.last_insert_id(cursor,
                    self.query.get_meta().db_table, self.query.get_meta().pk.column)


class SQLDeleteCompiler(SQLCompiler):
    def as_sql(self):
        """
        Creates the SQL for this query. Returns the SQL string and list of
        parameters.
        """
        assert len([t for t in self.query.tables if self.query.alias_refcount[t] > 0]) == 1, \
            "Can only delete from one table at a time."
        qn = self.quote_name_unless_alias
        result = ['DELETE FROM %s' % qn(self.query.tables[0])]
        where, params = self.compile(self.query.where)
        if where:
            result.append('WHERE %s' % where)
        return ' '.join(result), tuple(params)


class SQLUpdateCompiler(SQLCompiler):
    def as_sql(self):
        """
        Creates the SQL for this query. Returns the SQL string and list of
        parameters.
        """
        self.pre_sql_setup()
        if not self.query.values:
            return '', ()
        qn = self.quote_name_unless_alias
        values, update_params = [], []
        for field, model, val in self.query.values:
            if hasattr(val, 'resolve_expression'):
                val = val.resolve_expression(self.query, allow_joins=False, for_save=True)
                if val.contains_aggregate:
                    raise FieldError("Aggregate functions are not allowed in this query")
            elif hasattr(val, 'prepare_database_save'):
                if field.remote_field:
                    val = field.get_db_prep_save(
                        val.prepare_database_save(field),
                        connection=self.connection,
                    )
                else:
                    raise TypeError(
                        "Tried to update field %s with a model instance, %r. "
                        "Use a value compatible with %s."
                        % (field, val, field.__class__.__name__)
                    )
            else:
                val = field.get_db_prep_save(val, connection=self.connection)

            # Getting the placeholder for the field.
            if hasattr(field, 'get_placeholder'):
                placeholder = field.get_placeholder(val, self, self.connection)
            else:
                placeholder = '%s'
            name = field.column
            if hasattr(val, 'as_sql'):
                sql, params = self.compile(val)
                values.append('%s = %s' % (qn(name), sql))
                update_params.extend(params)
            elif val is not None:
                values.append('%s = %s' % (qn(name), placeholder))
                update_params.append(val)
            else:
                values.append('%s = NULL' % qn(name))
        if not values:
            return '', ()
        table = self.query.tables[0]
        result = [
            'UPDATE %s SET' % qn(table),
            ', '.join(values),
        ]
        where, params = self.compile(self.query.where)
        if where:
            result.append('WHERE %s' % where)
        return ' '.join(result), tuple(update_params + params)

    def execute_sql(self, result_type):
        """
        Execute the specified update. Returns the number of rows affected by
        the primary update query. The "primary update query" is the first
        non-empty query that is executed. Row counts for any subsequent,
        related queries are not available.
        """