fix struct field op

Author: chelsea-lin

bigframes/core/compile/sqlglot/expressions/unary_compiler.py

@@ -16,6 +16,8 @@
 
 import typing
 
+import pandas as pd
+import pyarrow as pa
 import sqlglot
 import sqlglot.expressions as sge
 
@@ -561,7 +563,17 @@ def _(op: ops.StrftimeOp, expr: TypedExpr) -> sge.Expression:
 
 @UNARY_OP_REGISTRATION.register(ops.StructFieldOp)
 def _(op: ops.StructFieldOp, expr: TypedExpr) -> sge.Expression:
-    return sge.StructExtract(this=expr.expr, expression=sge.convert(op.name_or_index))
+    if isinstance(op.name_or_index, str):
+        name = op.name_or_index
+    else:
+        pa_type = typing.cast(pd.ArrowDtype, expr.dtype)
+        pa_struct_type = typing.cast(pa.StructType, pa_type.pyarrow_dtype)
+        name = pa_struct_type.field(op.name_or_index).name
+
+    return sge.Column(
+        this=sge.to_identifier(name, quoted=True),
+        catalog=expr.expr,
+    )
 
 
 @UNARY_OP_REGISTRATION.register(ops.tan_op)

tests/unit/core/compile/sqlglot/expressions/snapshots/test_unary_compiler/test_struct_field/out.sql

@@ -0,0 +1,13 @@
+WITH `bfcte_0` AS (
+  SELECT
+    `people` AS `bfcol_0`
+  FROM `bigframes-dev`.`sqlglot_test`.`nested_structs_types`
+), `bfcte_1` AS (
+  SELECT
+    *,
+    `bfcol_0`.`name` AS `bfcol_1`
+  FROM `bfcte_0`
+)
+SELECT
+  `bfcol_1` AS `people`
+FROM `bfcte_1`

tests/unit/core/compile/sqlglot/expressions/test_unary_compiler.py

@@ -468,14 +468,16 @@ def test_strftime(scalar_types_df: bpd.DataFrame, snapshot):
     snapshot.assert_match(sql, "out.sql")
 
 
-def test_struct_field(scalar_types_df: bpd.DataFrame, snapshot):
-    bf_df = scalar_types_df[["int64_col"]]
-    bf_df["struct_col"] = bpd.DataFrame(
-        {"field1": bf_df["int64_col"], "field2": bf_df["int64_col"] * 2}
-    ).to_struct()
-    sql = _apply_unary_op(bf_df, ops.StructFieldOp("field1"), "struct_col")
+def test_struct_field(nested_structs_types_df: bpd.DataFrame, snapshot):
+    bf_df = nested_structs_types_df[["people"]]
+
+    # When a name string is provided.
+    sql = _apply_unary_op(bf_df, ops.StructFieldOp("name"), "people")
+    snapshot.assert_match(sql, "out.sql")
 
-    snapshot.assert_match(sql, "out_sql")
+    # When an index integer is provided.
+    sql = _apply_unary_op(bf_df, ops.StructFieldOp(0), "people")
+    snapshot.assert_match(sql, "out.sql")
 
 
 def test_str_contains(scalar_types_df: bpd.DataFrame, snapshot):