[IR] Steel-thread example for pandas iloc get

src/snowflake/snowpark/_internal/ast.py

@@ -8,9 +8,14 @@
 import json
 import sys
 import uuid
-from typing import Any, Sequence, Tuple
+from decimal import Decimal
+from inspect import signature
+from typing import Any, Callable, Sequence, Tuple
 
 from google.protobuf.json_format import ParseDict
+from numpy import datetime64, float64, int32, int64
+from pandas import Timestamp
+from pandas.core.dtypes.inference import is_list_like
 
 import snowflake.snowpark._internal.proto.ast_pb2 as proto
 from snowflake.connector.arrow_context import ArrowConverterContext
@@ -28,6 +33,60 @@ def expr_to_dataframe_expr(expr):
     return dfe
 
 
+# Map from python type to its corresponding IR entity. The entities below all have the 'v' attribute.
+TYPE_TO_IR_TYPE_NAME = {
+    bytes: "binary_val",
+    bool: "bool_val",
+    datetime64: "date_val",
+    Decimal: "big_decimal_val",
+    float64: "float_64_val",
+    int32: "int_32_val",
+    int64: "int_64_val",
+    str: "string_val",
+    Timestamp: "timestamp_val",
+}
+
+
+def ast_expr_from_python_val(expr, val):
+    """
+    Converts a Python value to an IR expression.
+    This IR expression is set to an attribute of `expr`.
+
+    Parameters
+    ----------
+    expr : IR entity protobuf builder
+    val : Python value that needs to be converted to IR expression.
+    """
+    if val is None:
+        expr.none_val = val
+    val_type = type(val)
+    if val_type not in TYPE_TO_IR_TYPE_NAME:
+        # Modin is imported here to prevent circular import issues.
+        from snowflake.snowpark.modin.pandas import DataFrame, Series
+
+        if isinstance(val, Callable):
+            for item in signature(val).parameters:
+                item_expr = expr.fn_val.params.add()
+                item_expr.v = item
+            expr.fn_val.body = val
+        if isinstance(val, slice):
+            expr.slice_val.start.v = val.start
+            expr.slice_val.stop.v = val.stop
+            expr.slice_val.step.v = val.step
+        elif not isinstance(val, Series) and is_list_like(val):
+            # Checking that val is not a Series since Series objects are considered list-like.
+            for item in val:
+                item_expr = expr.list_val.add()
+                item_expr.v = item
+        elif isinstance(val, Series):
+            expr.series_val.ref = val
+        elif isinstance(val, DataFrame):
+            expr.series_val.ref = val
+    else:
+        ir_type_name = TYPE_TO_IR_TYPE_NAME[val_type]
+        setattr(getattr(expr, ir_type_name), "v", val)  # noqa: B010
+
+
 def check_response(response: Any) -> None:
     # TODO SNOW-1474659: Add logic here to check whether response is a valid result,
     # else raise client-compatible exceptions.

src/snowflake/snowpark/column.py

@@ -4,7 +4,7 @@
 #
 
 import sys
-from typing import Any, Dict, Optional, Union
+from typing import Optional, Union
 
 import snowflake.snowpark
 import snowflake.snowpark._internal.proto.ast_pb2 as proto
@@ -624,20 +624,20 @@ def __invert__(self) -> "Column":
     def _cast(self, to: Union[str, DataType], try_: bool = False) -> "Column":
         if isinstance(to, str):
             to = type_string_to_type_object(to)
-        if try_:
-            ast = Column._create_ast(
-                property="sp_column_try_cast",
-                copy_messages={"col": self._ast},
-            )
-            to._fill_ast(ast.sp_column_try_cast.to)
-        else:
-            ast = Column._create_ast(
-                property="sp_column_cast",
-                copy_messages={"col": self._ast},
-            )
-            to._fill_ast(ast.sp_column_cast.to)
-
-        return Column(Cast(self._expression, to, try_), ast=ast)
+        # if try_:
+        #     ast = Column._create_ast(
+        #         property="sp_column_try_cast",
+        #         copy_messages={"col": self._ast},
+        #     )
+        #     to._fill_ast(ast.sp_column_try_cast.to)
+        # else:
+        #     ast = Column._create_ast(
+        #         property="sp_column_cast",
+        #         copy_messages={"col": self._ast},
+        #     )
+        #     to._fill_ast(ast.sp_column_cast.to)
+
+        return Column(Cast(self._expression, to, try_), ast=None)
 
     def cast(self, to: Union[str, DataType]) -> "Column":
         """Casts the value of the Column to the specified data type.
@@ -989,6 +989,16 @@ def _create_ast(
         Returns:
             proto.SpColumnExpr: Returns fully populated SpColumnExpr AST from given arguments
         """
+        return None
+        # Avoid using mutable values as default values in function signature.
+        if fill_expr_asts is None:
+            fill_expr_asts = {}
+        if copy_messages is None:
+            copy_messages = {}
+        if assign_opt_fields is None:
+            assign_opt_fields = {}
+        if assign_fields is None:
+            assign_fields = {}
 
         ast = proto.Expr()
         if property is not None:

src/snowflake/snowpark/dataframe.py

@@ -90,8 +90,8 @@
     decode_ast_response_from_snowpark,
 )
 from snowflake.snowpark._internal.ast_utils import (
-    set_src_position,
     get_symbol,
+    set_src_position,
     setattr_if_not_none,
 )
 from snowflake.snowpark._internal.error_message import SnowparkClientExceptionMessages

src/snowflake/snowpark/functions.py

@@ -250,11 +250,11 @@ def col(df_alias: str, col_name: str) -> Column:
     ...  # pragma: no cover
 
 
-def col(name1: str, name2: Optional[str] = None) -> Column:
+def col(name1: str, name2: Optional[str] = None, ast=None) -> Column:
     if name2 is None:
-        return Column(name1)
+        return Column(name1, ast=ast)
     else:
-        return Column(name1, name2)
+        return Column(name1, name2, ast=ast)
 
 
 @overload

src/snowflake/snowpark/modin/pandas/dataframe.py

@@ -150,6 +150,7 @@ def __init__(
         dtype=None,
         copy=None,
         query_compiler=None,
+        ast_stmt=None,
     ) -> None:
         # TODO: SNOW-1063346: Modin upgrade - modin.pandas.DataFrame functions
         # Siblings are other dataframes that share the same query compiler. We
@@ -158,6 +159,11 @@ def __init__(
 
         self._siblings = []
 
+        if not ast_stmt:
+            ast_stmt = pd.session._ast_batch.assign()
+        self._ast_id = ast_stmt.var_id.bitfield1
+        self._ast_stmt = ast_stmt
+
         # Engine.subscribe(_update_engine)
         if isinstance(data, (DataFrame, Series)):
             self._query_compiler = data._query_compiler.copy()
@@ -275,6 +281,9 @@ def __repr__(self):
         ) = self._query_compiler.build_repr_df(num_rows, num_cols, "x")
         result = repr(repr_df)
 
+        _, ast = pd.session._pd_ast_batch.flush()
+        print(ast)  # noqa: T201
+
         # if truncated, add shape information
         if is_repr_truncated(row_count, col_count, num_rows, num_cols):
             # The split here is so that we don't repr pandas row lengths.
@@ -2864,7 +2873,10 @@ def __setattr__(self, key, value):
         # - `_siblings`, which Modin initializes before it appears in __dict__
         # - `_cache`, which pandas.cache_readonly uses to cache properties
         #   before it appears in __dict__.
-        if key in ("_query_compiler", "_siblings", "_cache") or key in self.__dict__:
+        if (
+            key in ("_query_compiler", "_siblings", "_cache", "_ast_id", "_ast_stmt")
+            or key in self.__dict__
+        ):
             pass
         elif key in self and key not in dir(self):
             self.__setitem__(key, value)

src/snowflake/snowpark/modin/pandas/indexing.py

@@ -58,6 +58,7 @@
 
 import snowflake.snowpark.modin.pandas as pd
 import snowflake.snowpark.modin.pandas.utils as frontend_utils
+from snowflake.snowpark._internal.ast import ast_expr_from_python_val
 from snowflake.snowpark.modin.pandas.base import BasePandasDataset
 from snowflake.snowpark.modin.pandas.dataframe import DataFrame
 from snowflake.snowpark.modin.pandas.series import (
@@ -1049,6 +1050,15 @@ def __getitem__(
 
         original_row_loc = row_loc  # keep a copy for error message
 
+        # IR changes! Add iloc get nodes to AST.
+        stmt = pd.session._pd_ast_batch.assign()
+        ast = stmt.expr
+        ast.pd_dataframe_i_loc.df.var_id.bitfield1 = self.df._ast_id
+        # Map python built-ins (functions, scalars, lists, slices, etc.) to AST expr and emit Ref nodes for dataframes,
+        # series, and indexes.
+        ast_expr_from_python_val(ast.pd_dataframe_i_loc.rows, row_loc)
+        ast_expr_from_python_val(ast.pd_dataframe_i_loc.columns, col_loc)
+
         # Convert range to slice objects.
         if not isinstance(row_loc, pd.Series) and is_range_like(row_loc):
             row_loc = self._convert_range_to_valid_slice(row_loc)
@@ -1057,7 +1067,7 @@ def __getitem__(
 
         # Convert all scalar, list-like, and indexer row_loc to a Series object to get a query compiler object.
         if is_scalar(row_loc):
-            row_loc = pd.Series([row_loc])
+            row_loc = pd.Series([row_loc], ast_stmt=stmt)
         elif is_list_like(row_loc):
             if hasattr(row_loc, "dtype"):
                 dtype = row_loc.dtype
@@ -1066,7 +1076,7 @@ def __getitem__(
                 dtype = float
             else:
                 dtype = None
-            row_loc = pd.Series(row_loc, dtype=dtype)
+            row_loc = pd.Series(row_loc, dtype=dtype, ast_stmt=stmt)
 
         # Check whether the row and column input is of numeric dtype.
         self._validate_numeric_get_key_values(row_loc, original_row_loc)
@@ -1089,6 +1099,10 @@ def __getitem__(
         if isinstance(result, Series):
             result._parent = self.df
             result._parent_axis = 0
+
+        _, ast = pd.session._ast_batch.flush()
+        print(ast)  # noqa: T201
+
         return result
 
     def _get_pandas_object_from_qc_view(

src/snowflake/snowpark/modin/pandas/series.py

@@ -121,6 +121,7 @@ def __init__(
         copy=False,
         fastpath=False,
         query_compiler=None,
+        ast_stmt=None,
     ) -> None:
         # TODO: SNOW-1063347: Modin upgrade - modin.pandas.Series functions
         # Siblings are other dataframes that share the same query compiler. We
@@ -130,6 +131,11 @@ def __init__(
         # modified:
         # Engine.subscribe(_update_engine)
 
+        if not ast_stmt:
+            ast_stmt = pd.session._ast_batch.assign()
+        self._ast_id = ast_stmt.var_id.bitfield1
+        self._ast_stmt = ast_stmt
+
         if isinstance(data, type(self)):
             query_compiler = data._query_compiler.copy()
             if index is not None:
@@ -196,7 +202,9 @@ def _set_name(self, name):
         else:
             columns = [name]
         self._update_inplace(
-            new_query_compiler=self._query_compiler.set_columns(columns)
+            new_query_compiler=self._query_compiler.set_columns(
+                columns, ast_stmt=self._ast_stmt
+            )
         )
 
     name = property(_get_name, _set_name)
@@ -492,6 +500,10 @@ def __repr__(self):
             if temp_df.empty
             else temp_str.rsplit("dtype: ", 1)[-1]
         )
+
+        _, ast = pd.session._ast_batch.flush()
+        print(ast)  # noqa: T201
+
         if row_count == 0:
             return f"Series([], {freq_str}{name_str}{dtype_str}"
         maxsplit = 1

src/snowflake/snowpark/modin/plugin/_internal/frame.py

@@ -11,6 +11,7 @@
 from pandas._typing import IndexLabel
 from pandas.core.dtypes.common import is_object_dtype
 
+import snowflake.snowpark._internal.proto.ast_pb2 as proto
 from snowflake.snowpark._internal.analyzer.analyzer_utils import (
     quote_name_without_upper_casing,
 )
@@ -802,7 +803,9 @@ def project_columns(
         )
 
     def rename_snowflake_identifiers(
-        self, old_to_new_identifiers: dict[str, str]
+        self,
+        old_to_new_identifiers: dict[str, str],
+        ast_stmt: Optional[proto.Expr] = None,
     ) -> "InternalFrame":
         """
         Rename columns for underlying ordered dataframe.
@@ -847,7 +850,8 @@ def rename_snowflake_identifiers(
                 # retain the original column
                 select_list.append(old_id)
             else:
-                select_list.append(col(old_id).as_(new_id))
+                ast = ast_stmt.expr if ast_stmt is not None else None
+                select_list.append(col(old_id, ast=ast).as_(new_id))
                 # if the old column is part of the ordering or row position columns, retains the column
                 # as part of the projected columns.
                 if old_id in ordering_and_row_position_columns:

src/snowflake/snowpark/modin/plugin/compiler/snowflake_query_compiler.py

@@ -59,6 +59,7 @@
 from pandas.io.formats.format import format_percentiles
 from pandas.io.formats.printing import PrettyDict
 
+import snowflake.snowpark._internal.proto.ast_pb2 as proto
 import snowflake.snowpark.modin.pandas as pd
 from snowflake.snowpark._internal.analyzer.analyzer_utils import (
     quote_name_without_upper_casing,
@@ -1250,7 +1251,11 @@ def columns(self) -> "pd.Index":
         # TODO SNOW-837664: add more tests for df.columns
         return self._modin_frame.data_columns_index
 
-    def set_columns(self, new_pandas_labels: Axes) -> "SnowflakeQueryCompiler":
+    def set_columns(
+        self,
+        new_pandas_labels: Axes,
+        ast_stmt: Optional[proto.Expr] = None,
+    ) -> "SnowflakeQueryCompiler":
         """
         Set pandas column labels with the new column labels
 
@@ -1288,7 +1293,8 @@ def set_columns(self, new_pandas_labels: Axes) -> "SnowflakeQueryCompiler":
         )
 
         renamed_frame = self._modin_frame.rename_snowflake_identifiers(
-            renamed_quoted_identifier_mapping
+            renamed_quoted_identifier_mapping,
+            ast_stmt=ast_stmt,
         )
 
         new_internal_frame = InternalFrame.create(

src/snowflake/snowpark/session.py

@@ -509,6 +509,7 @@ def __init__(
             self._conn.create_coprocessor()
 
         self._ast_batch = AstBatch(self)
+        self._pd_ast_batch = AstBatch(self)
 
         _logger.info("Snowpark Session information: %s", self._session_info)
 

tests/ast/README.md

@@ -27,7 +27,7 @@ pytest --update-expectations tests/ast
 For these tests to work, the Unparser must be built in the monorepo:
 ```bash
 cd my-monorepo-path
-cd Snowflake/unparser
+cd Snowflake/trunk/Snowpark/unparser
 sbt assembly
 ```
 

tests/ast/data/pd-iloc.test

@@ -0,0 +1,18 @@
+## TEST CASE
+
+import modin.pandas as pd
+import snowflake.snowpark.modin.plugin
+from snowflake.snowpark import Session
+
+session = Session.builder.create()
+df = pd.DataFrame(
+    [[1, 2, 3, 4, 5, 6, 7], [8, 9, 10, 11, 12, 13, 14], [15, 16, 17, 18, 19, 20, 21]],
+    columns=["A", "B", "C", "D", "E", "F", "G"],
+)
+result = df.iloc[2, 2]
+print(result)
+
+## EXPECTED OUTPUT
+
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
+17