sqlframe 3.13.4__py3-none-any.whl → 3.14.1__py3-none-any.whl

sqlframe/_version.py

@@ -12,5 +12,5 @@ __version__: str
 __version_tuple__: VERSION_TUPLE
 version_tuple: VERSION_TUPLE
 
-__version__ = version = '3.13.4'
-__version_tuple__ = version_tuple = (3, 13, 4)
+__version__ = version = '3.14.1'
+__version_tuple__ = version_tuple = (3, 14, 1)

sqlframe/base/dataframe.py

@@ -79,6 +79,23 @@ JOIN_HINTS = {
     "SHUFFLE_REPLICATE_NL",
 }
 
+JOIN_TYPE_MAPPING = {
+    "inner": "inner",
+    "cross": "cross",
+    "outer": "full_outer",
+    "full": "full_outer",
+    "fullouter": "full_outer",
+    "left": "left_outer",
+    "leftouter": "left_outer",
+    "right": "right_outer",
+    "rightouter": "right_outer",
+    "semi": "left_semi",
+    "leftsemi": "left_semi",
+    "left_semi": "left_semi",
+    "anti": "left_anti",
+    "leftanti": "left_anti",
+    "left_anti": "left_anti",
+}
 
 DF = t.TypeVar("DF", bound="BaseDataFrame")
 
@@ -872,6 +889,68 @@ class BaseDataFrame(t.Generic[SESSION, WRITER, NA, STAT, GROUP_DATA]):
         """
         return self.join.__wrapped__(self, other, how="cross")  # type: ignore
 
+    def _handle_self_join(self, other_df: DF, join_columns: t.List[Column]):
+        # If the two dataframes being joined come from the same branch, we then check if they have any columns that
+        # were created using the "branch_id" (df["column_name"]). If so, we know that we need to differentiate
+        # the two columns since they would end up with the same table name. We do this by checking for the unique
+        # uuids in the other df and finding columns that have metadata on them that match the uuids. If so, we know
+        # it comes from the other df and we change the table name to the other df's table name.
+        # See `test_self_join` for an example of this.
+        if self.branch_id == other_df.branch_id:
+            other_df_unique_uuids = other_df.known_uuids - self.known_uuids
+            for col in join_columns:
+                for col_expr in col.expression.find_all(exp.Column):
+                    if (
+                        "join_on_uuid" in col_expr.meta
+                        and col_expr.meta["join_on_uuid"] in other_df_unique_uuids
+                    ):
+                        col_expr.set("table", exp.to_identifier(other_df.latest_cte_name))
+
+    @staticmethod
+    def _handle_join_column_names_only(
+        join_columns: t.List[Column],
+        join_expression: exp.Select,
+        other_df: DF,
+        table_names: t.List[str],
+    ):
+        potential_ctes = [
+            cte
+            for cte in join_expression.ctes
+            if cte.alias_or_name in table_names and cte.alias_or_name != other_df.latest_cte_name
+        ]
+        # Determine the table to reference for the left side of the join by checking each of the left side
+        # tables and see if they have the column being referenced.
+        join_column_pairs = []
+        for join_column in join_columns:
+            num_matching_ctes = 0
+            for cte in potential_ctes:
+                if join_column.alias_or_name in cte.this.named_selects:
+                    left_column = join_column.copy().set_table_name(cte.alias_or_name)
+                    right_column = join_column.copy().set_table_name(other_df.latest_cte_name)
+                    join_column_pairs.append((left_column, right_column))
+                    num_matching_ctes += 1
+                    # We only want to match one table to the column and that should be matched left -> right
+                    # so we break after the first match
+                    break
+            if num_matching_ctes == 0:
+                raise ValueError(
+                    f"Column `{join_column.alias_or_name}` does not exist in any of the tables."
+                )
+        join_clause = functools.reduce(
+            lambda x, y: x & y,
+            [left_column == right_column for left_column, right_column in join_column_pairs],
+        )
+        return join_column_pairs, join_clause
+
+    def _normalize_join_clause(
+        self, join_columns: t.List[Column], join_expression: t.Optional[exp.Select]
+    ) -> Column:
+        join_columns = self._ensure_and_normalize_cols(join_columns, join_expression)
+        if len(join_columns) > 1:
+            join_columns = [functools.reduce(lambda x, y: x & y, join_columns)]
+        join_clause = join_columns[0]
+        return join_clause
+
     @operation(Operation.FROM)
     def join(
         self,
@@ -882,37 +961,33 @@ class BaseDataFrame(t.Generic[SESSION, WRITER, NA, STAT, GROUP_DATA]):
     ) -> Self:
         from sqlframe.base.functions import coalesce
 
-        if on is None:
+        if (on is None) and ("cross" not in how):
             logger.warning("Got no value for on. This appears to change the join to a cross join.")
             how = "cross"
+        if (on is not None) and ("cross" in how):
+            # Not a lot of doc, but Spark handles cross with predicate as an inner join
+            # https://learn.microsoft.com/en-us/dotnet/api/microsoft.spark.sql.dataframe.join
+            logger.warning("Got cross join with an 'on' value. This will result in an inner join.")
+            how = "inner"
 
         other_df = other_df._convert_leaf_to_cte()
         join_expression = self._add_ctes_to_expression(self.expression, other_df.expression.ctes)
         # We will determine actual "join on" expression later so we don't provide it at first
-        join_expression = join_expression.join(
-            join_expression.ctes[-1].alias, join_type=how.replace("_", " ")
-        )
+        join_type = JOIN_TYPE_MAPPING.get(how, how).replace("_", " ")
+        join_expression = join_expression.join(join_expression.ctes[-1].alias, join_type=join_type)
         self_columns = self._get_outer_select_columns(join_expression)
         other_columns = self._get_outer_select_columns(other_df.expression)
         join_columns = self._ensure_and_normalize_cols(on)
-        # If the two dataframes being joined come from the same branch, we then check if they have any columns that
-        # were created using the "branch_id" (df["column_name"]). If so, we know that we need to differentiate
-        # the two columns since they would end up with the same table name. We do this by checking for the unique
-        # uuids in the other df and finding columns that have metadata on them that match the uuids. If so, we know
-        # it comes from the other df and we change the table name to the other df's table name.
-        # See `test_self_join` for an example of this.
-        if self.branch_id == other_df.branch_id:
-            other_df_unique_uuids = other_df.known_uuids - self.known_uuids
-            for col in join_columns:
-                for col_expr in col.expression.find_all(exp.Column):
-                    if (
-                        "join_on_uuid" in col_expr.meta
-                        and col_expr.meta["join_on_uuid"] in other_df_unique_uuids
-                    ):
-                        col_expr.set("table", exp.to_identifier(other_df.latest_cte_name))
+        self._handle_self_join(other_df, join_columns)
+
         # Determines the join clause and select columns to be used passed on what type of columns were provided for
         # the join. The columns returned changes based on how the on expression is provided.
-        if how != "cross":
+        select_columns = (
+            self_columns
+            if join_type in ["left anti", "left semi"]
+            else self_columns + other_columns
+        )
+        if join_type != "cross":
             if isinstance(join_columns[0].expression, exp.Column):
                 """
                 Unique characteristics of join on column names only:
@@ -923,38 +998,9 @@ class BaseDataFrame(t.Generic[SESSION, WRITER, NA, STAT, GROUP_DATA]):
                     table.alias_or_name
                     for table in get_tables_from_expression_with_join(join_expression)
                 ]
-                potential_ctes = [
-                    cte
-                    for cte in join_expression.ctes
-                    if cte.alias_or_name in table_names
-                    and cte.alias_or_name != other_df.latest_cte_name
-                ]
-                # Determine the table to reference for the left side of the join by checking each of the left side
-                # tables and see if they have the column being referenced.
-                join_column_pairs = []
-                for join_column in join_columns:
-                    num_matching_ctes = 0
-                    for cte in potential_ctes:
-                        if join_column.alias_or_name in cte.this.named_selects:
-                            left_column = join_column.copy().set_table_name(cte.alias_or_name)
-                            right_column = join_column.copy().set_table_name(
-                                other_df.latest_cte_name
-                            )
-                            join_column_pairs.append((left_column, right_column))
-                            num_matching_ctes += 1
-                            # We only want to match one table to the column and that should be matched left -> right
-                            # so we break after the first match
-                            break
-                    if num_matching_ctes == 0:
-                        raise ValueError(
-                            f"Column `{join_column.alias_or_name}` does not exist in any of the tables."
-                        )
-                join_clause = functools.reduce(
-                    lambda x, y: x & y,
-                    [
-                        left_column == right_column
-                        for left_column, right_column in join_column_pairs
-                    ],
+
+                join_column_pairs, join_clause = self._handle_join_column_names_only(
+                    join_columns, join_expression, other_df, table_names
                 )
                 join_column_names = [
                     coalesce(
@@ -972,7 +1018,7 @@ class BaseDataFrame(t.Generic[SESSION, WRITER, NA, STAT, GROUP_DATA]):
                         if not isinstance(column.expression.this, exp.Star)
                         else column.sql()
                     )
-                    for column in self_columns + other_columns
+                    for column in select_columns
                 ]
                 select_column_names = [
                     column_name
@@ -989,17 +1035,12 @@ class BaseDataFrame(t.Generic[SESSION, WRITER, NA, STAT, GROUP_DATA]):
                 * There is no deduplication of the results.
                 * The left join dataframe columns go first and right come after. No sort preference is given to join columns
                 """
-                join_columns = self._ensure_and_normalize_cols(join_columns, join_expression)
-                if len(join_columns) > 1:
-                    join_columns = [functools.reduce(lambda x, y: x & y, join_columns)]
-                join_clause = join_columns[0]
-                select_column_names = [
-                    column.alias_or_name for column in self_columns + other_columns
-                ]
+                join_clause = self._normalize_join_clause(join_columns, join_expression)
+                select_column_names = [column.alias_or_name for column in select_columns]
 
             # Update the on expression with the actual join clause to replace the dummy one from before
         else:
-            select_column_names = [column.alias_or_name for column in self_columns + other_columns]
+            select_column_names = [column.alias_or_name for column in select_columns]
             join_clause = None
         join_expression.args["joins"][-1].set("on", join_clause.expression if join_clause else None)
         new_df = self.copy(expression=join_expression)

sqlframe/base/mixins/table_mixins.py

@@ -0,0 +1,335 @@
+import functools
+import logging
+import typing as t
+
+from sqlglot import exp
+
+try:
+    from sqlglot.expressions import Whens
+except ImportError:
+    Whens = None  # type: ignore
+from sqlglot.helper import object_to_dict
+
+from sqlframe.base.column import Column
+from sqlframe.base.table import (
+    DF,
+    Clause,
+    LazyExpression,
+    WhenMatched,
+    WhenNotMatched,
+    WhenNotMatchedBySource,
+    _BaseTable,
+)
+
+if t.TYPE_CHECKING:
+    from sqlframe.base._typing import ColumnOrLiteral
+
+
+logger = logging.getLogger(__name__)
+
+
+def ensure_cte() -> t.Callable[[t.Callable], t.Callable]:
+    def decorator(func: t.Callable) -> t.Callable:
+        @functools.wraps(func)
+        def wrapper(self: _BaseTable, *args, **kwargs) -> t.Any:
+            if len(self.expression.ctes) > 0:
+                return func(self, *args, **kwargs)  # type: ignore
+            self_class = self.__class__
+            self = self._convert_leaf_to_cte()
+            self = self_class(**object_to_dict(self))
+            return func(self, *args, **kwargs)  # type: ignore
+
+        wrapper.__wrapped__ = func  # type: ignore
+        return wrapper
+
+    return decorator
+
+
+class _BaseTableMixins(_BaseTable, t.Generic[DF]):
+    def _ensure_where_condition(
+        self, where: t.Optional[t.Union[Column, str, bool]] = None
+    ) -> exp.Expression:
+        self_name = self.expression.ctes[0].this.args["from"].this.alias_or_name
+
+        if where is None:
+            logger.warning("Empty value for `where`clause. Defaults to `True`.")
+            condition: exp.Expression = exp.Boolean(this=True)
+        else:
+            condition_list = self._ensure_and_normalize_cols(where, self.expression)
+            if len(condition_list) > 1:
+                condition_list = [functools.reduce(lambda x, y: x & y, condition_list)]
+            for col_expr in condition_list[0].expression.find_all(exp.Column):
+                if col_expr.table == self.expression.args["from"].this.alias_or_name:
+                    col_expr.set("table", exp.to_identifier(self_name))
+            condition = condition_list[0].expression
+            if isinstance(condition, exp.Alias):
+                condition = condition.this
+        return condition
+
+
+class UpdateSupportMixin(_BaseTableMixins, t.Generic[DF]):
+    @ensure_cte()
+    def update(
+        self,
+        set_: t.Dict[t.Union[Column, str], t.Union[Column, "ColumnOrLiteral", exp.Expression]],
+        where: t.Optional[t.Union[Column, str, bool]] = None,
+    ) -> LazyExpression:
+        self_expr = self.expression.ctes[0].this.args["from"].this
+
+        condition = self._ensure_where_condition(where)
+        update_set = self._ensure_and_normalize_update_set(set_)
+        update_expr = exp.Update(
+            this=self_expr,
+            expressions=[
+                exp.EQ(
+                    this=key,
+                    expression=val,
+                )
+                for key, val in update_set.items()
+            ],
+            where=exp.Where(this=condition),
+        )
+
+        return LazyExpression(update_expr, self.session)
+
+    def _ensure_and_normalize_update_set(
+        self,
+        set_: t.Dict[t.Union[Column, str], t.Union[Column, "ColumnOrLiteral", exp.Expression]],
+    ) -> t.Dict[str, exp.Expression]:
+        self_name = self.expression.ctes[0].this.args["from"].this.alias_or_name
+        update_set = {}
+        for key, val in set_.items():
+            key_column: Column = self._ensure_and_normalize_col(key)
+            key_expr = list(key_column.expression.find_all(exp.Column))
+            if len(key_expr) > 1:
+                raise ValueError(f"Can only update one a single column at a time.")
+            key = key_expr[0].alias_or_name
+
+            val_column: Column = self._ensure_and_normalize_col(val)
+            for col_expr in val_column.expression.find_all(exp.Column):
+                if col_expr.table == self.expression.args["from"].this.alias_or_name:
+                    col_expr.set("table", exp.to_identifier(self_name))
+                else:
+                    raise ValueError(
+                        f"Column `{col_expr.alias_or_name}` does not exist in the table."
+                    )
+
+            update_set[key] = val_column.expression
+        return update_set
+
+
+class DeleteSupportMixin(_BaseTableMixins, t.Generic[DF]):
+    @ensure_cte()
+    def delete(
+        self,
+        where: t.Optional[t.Union[Column, str, bool]] = None,
+    ) -> LazyExpression:
+        self_expr = self.expression.ctes[0].this.args["from"].this
+
+        condition = self._ensure_where_condition(where)
+        delete_expr = exp.Delete(
+            this=self_expr,
+            where=exp.Where(this=condition),
+        )
+
+        return LazyExpression(delete_expr, self.session)
+
+
+class MergeSupportMixin(_BaseTable, t.Generic[DF]):
+    _merge_supported_clauses: t.Iterable[
+        t.Union[t.Type[WhenMatched], t.Type[WhenNotMatched], t.Type[WhenNotMatchedBySource]]
+    ]
+    _merge_support_star: bool
+
+    @ensure_cte()
+    def merge(
+        self,
+        other_df: DF,
+        condition: t.Union[str, t.List[str], Column, t.List[Column], bool],
+        clauses: t.Iterable[t.Union[WhenMatched, WhenNotMatched, WhenNotMatchedBySource]],
+    ) -> LazyExpression:
+        self_name = self.expression.ctes[0].this.args["from"].this.alias_or_name
+        self_expr = self.expression.ctes[0].this.args["from"].this
+
+        other_df = other_df._convert_leaf_to_cte()
+
+        if condition is None:
+            raise ValueError("condition cannot be None")
+
+        condition_columns: Column = self._ensure_and_normalize_condition(condition, other_df)
+        other_name = self._create_hash_from_expression(other_df.expression)
+        other_expr = exp.Subquery(
+            this=other_df.expression, alias=exp.TableAlias(this=exp.to_identifier(other_name))
+        )
+
+        for col_expr in condition_columns.expression.find_all(exp.Column):
+            if col_expr.table == self.expression.args["from"].this.alias_or_name:
+                col_expr.set("table", exp.to_identifier(self_name))
+            if col_expr.table == other_df.latest_cte_name:
+                col_expr.set("table", exp.to_identifier(other_name))
+
+        merge_expressions = []
+        for clause in clauses:
+            if not isinstance(clause, tuple(self._merge_supported_clauses)):
+                raise ValueError(
+                    f"Unsupported clause type {type(clause.clause)} for merge operation"
+                )
+            expression = None
+
+            if clause.clause.condition is not None:
+                cond_clause = self._ensure_and_normalize_condition(
+                    clause.clause.condition, other_df, True
+                )
+                for col_expr in cond_clause.expression.find_all(exp.Column):
+                    if col_expr.table == self.expression.args["from"].this.alias_or_name:
+                        col_expr.set("table", exp.to_identifier(self_name))
+                    if col_expr.table == other_df.latest_cte_name:
+                        col_expr.set("table", exp.to_identifier(other_name))
+            else:
+                cond_clause = None
+            if clause.clause.clause_type == Clause.UPDATE:
+                update_set = self._ensure_and_normalize_assignments(
+                    clause.clause.assignments, other_df
+                )
+                expression = exp.When(
+                    matched=clause.clause.matched,
+                    source=clause.clause.by_source,
+                    condition=cond_clause.expression if cond_clause else None,
+                    then=exp.Update(
+                        expressions=[
+                            exp.EQ(
+                                this=key,
+                                expression=val,
+                            )
+                            for key, val in update_set.items()
+                        ]
+                    ),
+                )
+            if clause.clause.clause_type == Clause.UPDATE_ALL:
+                if not self._support_star:
+                    raise ValueError("Merge operation does not support UPDATE_ALL")
+                expression = exp.When(
+                    matched=clause.clause.matched,
+                    source=clause.clause.by_source,
+                    condition=cond_clause.expression if cond_clause else None,
+                    then=exp.Update(expressions=[exp.Star()]),
+                )
+            elif clause.clause.clause_type == Clause.INSERT:
+                insert_values = self._ensure_and_normalize_assignments(
+                    clause.clause.assignments, other_df
+                )
+                expression = exp.When(
+                    matched=clause.clause.matched,
+                    source=clause.clause.by_source,
+                    condition=cond_clause.expression if cond_clause else None,
+                    then=exp.Insert(
+                        this=exp.Tuple(expressions=[key for key in insert_values.keys()]),
+                        expression=exp.Tuple(expressions=[val for val in insert_values.values()]),
+                    ),
+                )
+            elif clause.clause.clause_type == Clause.INSERT_ALL:
+                if not self._support_star:
+                    raise ValueError("Merge operation does not support INSERT_ALL")
+                expression = exp.When(
+                    matched=clause.clause.matched,
+                    source=clause.clause.by_source,
+                    condition=cond_clause.expression if cond_clause else None,
+                    then=exp.Insert(expression=exp.Star()),
+                )
+            elif clause.clause.clause_type == Clause.DELETE:
+                expression = exp.When(
+                    matched=clause.clause.matched,
+                    source=clause.clause.by_source,
+                    condition=cond_clause.expression if cond_clause else None,
+                    then=exp.var("DELETE"),
+                )
+
+            if expression:
+                merge_expressions.append(expression)
+
+        if Whens is None:
+            merge_expr = exp.merge(
+                *merge_expressions,
+                into=self_expr,
+                using=other_expr,
+                on=condition_columns.expression,
+            )
+        else:
+            merge_expr = exp.merge(
+                Whens(expressions=merge_expressions),
+                into=self_expr,
+                using=other_expr,
+                on=condition_columns.expression,
+            )
+
+        return LazyExpression(merge_expr, self.session)
+
+    def _ensure_and_normalize_condition(
+        self,
+        condition: t.Union[str, t.List[str], Column, t.List[Column], bool],
+        other_df: DF,
+        clause: t.Optional[bool] = False,
+    ):
+        join_expression = self._add_ctes_to_expression(
+            self.expression, other_df.expression.copy().ctes
+        )
+        condition = self._ensure_and_normalize_cols(condition, self.expression)
+        self._handle_self_join(other_df, condition)
+
+        if isinstance(condition[0].expression, exp.Column) and not clause:
+            table_names = [
+                table.alias_or_name
+                for table in [
+                    self.expression.args["from"].this,
+                    other_df.expression.args["from"].this,
+                ]
+            ]
+
+            join_column_pairs, join_clause = self._handle_join_column_names_only(
+                condition, join_expression, other_df, table_names
+            )
+        else:
+            join_clause = self._normalize_join_clause(condition, join_expression)
+        return join_clause
+
+    def _ensure_and_normalize_assignments(
+        self,
+        assignments: t.Dict[
+            t.Union[Column, str], t.Union[Column, "ColumnOrLiteral", exp.Expression]
+        ],
+        other_df,
+    ) -> t.Dict[exp.Column, exp.Expression]:
+        self_name = self.expression.ctes[0].this.args["from"].this.alias_or_name
+        other_name = self._create_hash_from_expression(other_df.expression)
+        update_set = {}
+        for key, val in assignments.items():
+            key_column: Column = self._ensure_and_normalize_col(key)
+            key_expr = list(key_column.expression.find_all(exp.Column))
+            if len(key_expr) > 1:
+                raise ValueError(f"Target expression `{key_expr}` should be a single column.")
+            column_key = exp.column(key_expr[0].alias_or_name)
+
+            val = self._ensure_and_normalize_col(val)
+            val = self._ensure_and_normalize_cols(val, other_df.expression)[0]
+            if self.branch_id == other_df.branch_id:
+                other_df_unique_uuids = other_df.known_uuids - self.known_uuids
+                for col_expr in val.expression.find_all(exp.Column):
+                    if (
+                        "join_on_uuid" in col_expr.meta
+                        and col_expr.meta["join_on_uuid"] in other_df_unique_uuids
+                    ):
+                        col_expr.set("table", exp.to_identifier(other_df.latest_cte_name))
+
+            for col_expr in val.expression.find_all(exp.Column):
+                if not col_expr.table or col_expr.table == other_df.latest_cte_name:
+                    col_expr.set("table", exp.to_identifier(other_name))
+                elif col_expr.table == self.expression.args["from"].this.alias_or_name:
+                    col_expr.set("table", exp.to_identifier(self_name))
+                else:
+                    raise ValueError(
+                        f"Column `{col_expr.alias_or_name}` does not exist in any of the tables."
+                    )
+            if isinstance(val.expression, exp.Alias):
+                val.expression = val.expression.this
+            update_set[column_key] = val.expression
+        return update_set

sqlframe/base/readerwriter.py

@@ -21,19 +21,20 @@ else:
 if t.TYPE_CHECKING:
     from sqlframe.base._typing import OptionalPrimitiveType, PathOrPaths
     from sqlframe.base.column import Column
-    from sqlframe.base.session import DF, _BaseSession
+    from sqlframe.base.session import DF, TABLE, _BaseSession
     from sqlframe.base.types import StructType
 
     SESSION = t.TypeVar("SESSION", bound=_BaseSession)
 else:
     SESSION = t.TypeVar("SESSION")
     DF = t.TypeVar("DF")
+    TABLE = t.TypeVar("TABLE")
 
 
 logger = logging.getLogger(__name__)
 
 
-class _BaseDataFrameReader(t.Generic[SESSION, DF]):
+class _BaseDataFrameReader(t.Generic[SESSION, DF, TABLE]):
     def __init__(self, spark: SESSION):
         self._session = spark
         self.state_format_to_read: t.Optional[str] = None
@@ -42,7 +43,7 @@ class _BaseDataFrameReader(t.Generic[SESSION, DF]):
     def session(self) -> SESSION:
         return self._session
 
-    def table(self, tableName: str) -> DF:
+    def table(self, tableName: str) -> TABLE:
         tableName = normalize_string(tableName, from_dialect="input", is_table=True)
         if df := self.session.temp_views.get(tableName):
             return df
@@ -50,7 +51,7 @@ class _BaseDataFrameReader(t.Generic[SESSION, DF]):
         self.session.catalog.add_table(table)
         columns = self.session.catalog.get_columns_from_schema(table)
 
-        return self.session._create_df(
+        return self.session._create_table(
             exp.Select()
             .from_(tableName, dialect=self.session.input_dialect)
             .select(*columns, dialect=self.session.input_dialect)

sqlframe/base/session.py

@@ -27,6 +27,7 @@ from sqlframe.base.catalog import _BaseCatalog
 from sqlframe.base.dataframe import BaseDataFrame
 from sqlframe.base.normalize import normalize_dict
 from sqlframe.base.readerwriter import _BaseDataFrameReader, _BaseDataFrameWriter
+from sqlframe.base.table import _BaseTable
 from sqlframe.base.udf import _BaseUDFRegistration
 from sqlframe.base.util import (
     get_column_mapping_from_schema_input,
@@ -65,17 +66,19 @@ CATALOG = t.TypeVar("CATALOG", bound=_BaseCatalog)
 READER = t.TypeVar("READER", bound=_BaseDataFrameReader)
 WRITER = t.TypeVar("WRITER", bound=_BaseDataFrameWriter)
 DF = t.TypeVar("DF", bound=BaseDataFrame)
+TABLE = t.TypeVar("TABLE", bound=_BaseTable)
 UDF_REGISTRATION = t.TypeVar("UDF_REGISTRATION", bound=_BaseUDFRegistration)
 
 _MISSING = "MISSING"
 
 
-class _BaseSession(t.Generic[CATALOG, READER, WRITER, DF, CONN, UDF_REGISTRATION]):
+class _BaseSession(t.Generic[CATALOG, READER, WRITER, DF, TABLE, CONN, UDF_REGISTRATION]):
     _instance = None
     _reader: t.Type[READER]
     _writer: t.Type[WRITER]
     _catalog: t.Type[CATALOG]
     _df: t.Type[DF]
+    _table: t.Type[TABLE]
     _udf_registration: t.Type[UDF_REGISTRATION]
 
     SANITIZE_COLUMN_NAMES = False
@@ -158,12 +161,15 @@ class _BaseSession(t.Generic[CATALOG, READER, WRITER, DF, CONN, UDF_REGISTRATION
             return name.replace("(", "_").replace(")", "_")
         return name
 
-    def table(self, tableName: str) -> DF:
+    def table(self, tableName: str) -> TABLE:
         return self.read.table(tableName)
 
     def _create_df(self, *args, **kwargs) -> DF:
         return self._df(self, *args, **kwargs)
 
+    def _create_table(self, *args, **kwargs) -> TABLE:
+        return self._table(self, *args, **kwargs)
+
     def __new__(cls, *args, **kwargs):
         if _BaseSession._instance is None:
             _BaseSession._instance = super().__new__(cls)