feat: integrated duckdb casting into data contract and added initial spark casting

Author: stevenhsd

src/dve/core_engine/backends/implementations/duckdb/contract.py

@@ -28,6 +28,7 @@
     generate_error_casting_entity_message,
 )
 from dve.core_engine.backends.implementations.duckdb.duckdb_helpers import (
+    get_duckdb_cast_statement_from_annotation,
     duckdb_read_parquet,
     duckdb_record_index,
     duckdb_write_parquet,
@@ -101,17 +102,6 @@ def create_entity_from_py_iterator(  # pylint: disable=unused-argument
         _lazy_df = pl.LazyFrame(records, polars_schema)  # type: ignore # pylint: disable=unused-variable
         return self._connection.sql("select * from _lazy_df")
 
-    @staticmethod
-    def generate_ddb_cast_statement(
-        column_name: str, dtype: DuckDBPyType, null_flag: bool = False
-    ) -> str:
-        """Helper method to generate sql statements for casting datatypes (permissively).
-        Current duckdb python API doesn't play well with this currently.
-        """
-        if not null_flag:
-            return f'try_cast("{column_name}" AS {dtype}) AS "{column_name}"'
-        return f'cast(NULL AS {dtype}) AS "{column_name}"'
-
     # pylint: disable=R0914
     def apply_data_contract(
         self,
@@ -180,12 +170,13 @@ def apply_data_contract(
 
                 casting_statements = [
                     (
-                        self.generate_ddb_cast_statement(column, dtype)
+                        get_duckdb_cast_statement_from_annotation(column, mdl_fld.annotation) + f""" AS "{column}" """
                         if column in relation.columns
-                        else self.generate_ddb_cast_statement(column, dtype, null_flag=True)
+                        else f"CAST(NULL AS {ddb_schema[column]}) AS {column}"
                     )
-                    for column, dtype in ddb_schema.items()
+                    for column, mdl_fld in entity_fields.items()
                 ]
+                casting_statements.append(f"CAST({RECORD_INDEX_COLUMN_NAME} AS {get_duckdb_type_from_annotation(int)}) AS {RECORD_INDEX_COLUMN_NAME}")
                 try:
                     relation = relation.project(", ".join(casting_statements))
                 except Exception as err:  # pylint: disable=broad-except

src/dve/core_engine/backends/implementations/duckdb/duckdb_helpers.py

@@ -315,30 +315,39 @@ def duckdb_record_index(cls):
     return cls
 
 def _cast_as_ddb_type(field_expr:str, type_annotation:Any) -> str:
-    return f"try_cast({field_expr} as {get_duckdb_type_from_annotation(type_annotation)})"
+    return f"""try_cast({field_expr} as {get_duckdb_type_from_annotation(type_annotation)})"""
+
+def _ddb_safely_quote_name(field_name:str) -> str:
+    try:
+        sep_idx = field_name.rindex(".")
+        return field_name[:sep_idx + 1] + f"\"{field_name[sep_idx + 1:]}\""
+    except ValueError:
+        return f"\"{field_name}\""
 
 
 def get_duckdb_cast_statement_from_annotation(element_name:str,
                                               type_annotation: Any,
                                               date_regex: str = r"^[0-9]{4}-[0-9]{2}-[0-9]{2}$",
                                               timestamp_regex:str = r"^[0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}:[0-9]{2}$",
-                                              include_cast: bool = True) -> DuckDBPyType:
+                                              parent_element: bool = True) -> str:
     type_origin = get_origin(type_annotation)
+    
+    quoted_name = _ddb_safely_quote_name(element_name)
 
     # An `Optional` or `Union` type, check to ensure non-heterogenity.
     if type_origin is Union:
         python_type = _get_non_heterogenous_type(get_args(type_annotation))
-        return get_duckdb_cast_statement_from_annotation(element_name, python_type, date_regex, timestamp_regex, include_cast)
+        return get_duckdb_cast_statement_from_annotation(element_name, python_type, date_regex, timestamp_regex, parent_element)
 
     # Type hint is e.g. `List[str]`, check to ensure non-heterogenity.
     if type_origin is list or (isinstance(type_origin, type) and issubclass(type_origin, list)):
         element_type = _get_non_heterogenous_type(get_args(type_annotation))
-        stmt = f"list_transform({element_name}, x -> {get_duckdb_cast_statement_from_annotation('x',element_type, date_regex, timestamp_regex, False)})"
-        return stmt if not include_cast else _cast_as_ddb_type(stmt, type_annotation)
+        stmt = f"list_transform({quoted_name}, x -> {get_duckdb_cast_statement_from_annotation('x',element_type, date_regex, timestamp_regex, False)})"
+        return stmt if not parent_element else _cast_as_ddb_type(stmt, type_annotation)
 
     if type_origin is Annotated:
         python_type, *other_args = get_args(type_annotation)  # pylint: disable=unused-variable
-        return get_duckdb_cast_statement_from_annotation(element_name, python_type, date_regex, timestamp_regex, include_cast) # add other expected params here
+        return get_duckdb_cast_statement_from_annotation(element_name, python_type, date_regex, timestamp_regex, parent_element) # add other expected params here
     # Ensure that we have a concrete type at this point.
     if not isinstance(type_annotation, type):
         raise ValueError(f"Unsupported type annotation {type_annotation!r}")
@@ -372,9 +381,9 @@ def get_duckdb_cast_statement_from_annotation(element_name:str,
             raise ValueError(
                 f"No type annotations in dict/dataclass type (got {type_annotation!r})"
             )
-        cast_exprs = ",".join([f'{nme}:= {stmt}' for nme, stmt in fields.items()])
+        cast_exprs = ",".join([f"\"{nme}\":= {stmt}" for nme, stmt in fields.items()])
         stmt = f"struct_pack({cast_exprs})"
-        return stmt if not include_cast else _cast_as_ddb_type(stmt, type_annotation)
+        return stmt if not parent_element else _cast_as_ddb_type(stmt, type_annotation)
 
     if type_annotation is list:
         raise ValueError(
@@ -385,13 +394,13 @@ def get_duckdb_cast_statement_from_annotation(element_name:str,
 
     for type_ in type_annotation.mro():
         if issubclass(type_, datetime):
-            stmt = f"CASE WHEN REGEXP_MATCHES(TRIM({element_name}), '{timestamp_regex}') THEN TRY_CAST(TRIM({element_name}) as TIMESTAMP) ELSE NULL END"
+            stmt = f"CASE WHEN REGEXP_MATCHES(TRIM({quoted_name}), '{timestamp_regex}') THEN TRY_CAST(TRIM({quoted_name}) as TIMESTAMP) ELSE NULL END"
             return stmt
         if issubclass(type_, date):
-            stmt = f"CASE WHEN REGEXP_MATCHES(TRIM({element_name}), '{date_regex}') THEN TRY_CAST(TRIM({element_name}) as DATE) ELSE NULL END"
+            stmt = f"CASE WHEN REGEXP_MATCHES(TRIM({quoted_name}), '{date_regex}') THEN TRY_CAST(TRIM({quoted_name}) as DATE) ELSE NULL END"
             return stmt
         duck_type = get_duckdb_type_from_annotation(type_)   
         if duck_type:
-            stmt =  f"trim({element_name})"
-            return _cast_as_ddb_type(stmt, type_) if include_cast else stmt
+            stmt =  f"trim({quoted_name})"
+            return _cast_as_ddb_type(stmt, type_) if parent_element else stmt
     raise ValueError(f"No equivalent DuckDB type for {type_annotation!r}")

src/dve/core_engine/backends/implementations/spark/spark_helpers.py

@@ -439,3 +439,76 @@ def spark_record_index(cls):
     setattr(cls, "add_record_index", _add_spark_record_index)
     setattr(cls, "drop_record_index", _drop_spark_record_index)
     return cls
+
+
+def _cast_as_spark_type(field_expr:str, field_type: st.DataType) -> Column:
+    return sf.expr(field_expr).cast(field_type)
+
+
+def get_spark_cast_statement_from_annotation(element_name:str,
+                                              type_annotation: Any,
+                                              include_cast: bool = True) -> st.DataType:
+    type_origin = get_origin(type_annotation)
+
+    # An `Optional` or `Union` type, check to ensure non-heterogenity.
+    if type_origin is Union:
+        python_type = _get_non_heterogenous_type(get_args(type_annotation))
+        return get_spark_cast_statement_from_annotation(element_name, include_cast)
+
+    # Type hint is e.g. `List[str]`, check to ensure non-heterogenity.
+    if type_origin is list or (isinstance(type_origin, type) and issubclass(type_origin, list)):
+        element_type = _get_non_heterogenous_type(get_args(type_annotation))
+        stmt = f"transform({element_name}, x -> {get_spark_cast_statement_from_annotation('x',element_type, False)})"
+        return stmt if not include_cast else _cast_as_spark_type(stmt, type_annotation)
+
+    if type_origin is Annotated:
+        python_type, *other_args = get_args(type_annotation)  # pylint: disable=unused-variable
+        return get_spark_cast_statement_from_annotation(element_name, python_type, include_cast) # add other expected params here
+    # Ensure that we have a concrete type at this point.
+    if not isinstance(type_annotation, type):
+        raise ValueError(f"Unsupported type annotation {type_annotation!r}")
+
+    if (
+        # Type hint is a dict subclass, but not dict. Possibly a `TypedDict`.
+        (issubclass(type_annotation, dict) and type_annotation is not dict)
+        # Type hint is a dataclass.
+        or is_dataclass(type_annotation)
+        # Type hint is a `pydantic` model.
+        or (type_origin is None and issubclass(type_annotation, BaseModel))
+    ):
+        fields: dict[str, str] = {}
+        for field_name, field_annotation in get_type_hints(type_annotation).items():
+            # Technically non-string keys are disallowed, but people are bad.
+            if not isinstance(field_name, str):
+                raise ValueError(
+                    f"Dictionary/Dataclass keys must be strings, got {type_annotation!r}"
+                )  # pragma: no cover
+            if get_origin(field_annotation) is ClassVar:
+                continue
+
+            fields[field_name] = get_spark_cast_statement_from_annotation(
+                f"{element_name}.{field_name}",
+                field_annotation,
+                False)
+
+        if not fields:
+            raise ValueError(
+                f"No type annotations in dict/dataclass type (got {type_annotation!r})"
+            )
+        cast_exprs = ",".join([f'{nme}:= {stmt}' for nme, stmt in fields.items()])
+        stmt = f"struct_pack({cast_exprs})"
+        return stmt if not include_cast else _cast_as_spark_type(stmt, type_annotation)
+
+    if type_annotation is list:
+        raise ValueError(
+            f"List must have type annotation (e.g. `List[str]`), got {type_annotation!r}"
+        )
+    if type_annotation is dict or type_origin is dict:
+        raise ValueError(f"dict must be `typing.TypedDict` subclass, got {type_annotation!r}")
+
+    for type_ in type_annotation.mro():
+        duck_type = get_type_from_annotation(type_)   
+        if duck_type:
+            stmt =  f"trim({element_name})"
+            return _cast_as_spark_type(stmt, type_) if include_cast else stmt
+    raise ValueError(f"No equivalent DuckDB type for {type_annotation!r}")

tests/test_core_engine/test_backends/test_implementations/test_duckdb/test_duckdb_helpers.py

@@ -154,13 +154,13 @@ def test_duckdb_rel_to_dictionaries(temp_ddb_conn: DuckDBPyConnection,
 
 # add decimal check
 @pytest.mark.parametrize("field_name,field_type,cast_statement",
-                         [("str_test", str, "try_cast(trim(str_test) as VARCHAR)"),
-                          ("int_test", int, "try_cast(trim(int_test) as BIGINT)"),
-                          ("date_test", datetime.date,"CASE WHEN REGEXP_MATCHES(TRIM(date_test), '^[0-9]{4}-[0-9]{2}-[0-9]{2}$') THEN TRY_CAST(TRIM(date_test) as DATE) ELSE NULL END"),
-                          ("timestamp_test", datetime.datetime,"CASE WHEN REGEXP_MATCHES(TRIM(timestamp_test), '^[0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}:[0-9]{2}$') THEN TRY_CAST(TRIM(timestamp_test) as TIMESTAMP) ELSE NULL END"),
-                          ("list_int_field", list[int], "try_cast(list_transform(list_int_field, x -> trim(x)) as BIGINT[])"),
-                          ("basic_model", BasicModel, "try_cast(struct_pack(str_field:= trim(basic_model.str_field),date_field:= CASE WHEN REGEXP_MATCHES(TRIM(basic_model.date_field), '^[0-9]{4}-[0-9]{2}-[0-9]{2}$') THEN TRY_CAST(TRIM(basic_model.date_field) as DATE) ELSE NULL END) as STRUCT(str_field VARCHAR, date_field DATE))"),
-                          ("another_model", AnotherModel, "try_cast(struct_pack(unique_id:= trim(another_model.unique_id),basic_models:= list_transform(another_model.basic_models, x -> struct_pack(str_field:= trim(x.str_field),date_field:= CASE WHEN REGEXP_MATCHES(TRIM(x.date_field), '^[0-9]{4}-[0-9]{2}-[0-9]{2}$') THEN TRY_CAST(TRIM(x.date_field) as DATE) ELSE NULL END))) as STRUCT(unique_id BIGINT, basic_models STRUCT(str_field VARCHAR, date_field DATE)[]))")])
+                         [("str_test", str, "try_cast(trim(\"str_test\") as VARCHAR)"),
+                          ("int_test", int, "try_cast(trim(\"int_test\") as BIGINT)"),
+                          ("date_test", datetime.date,"CASE WHEN REGEXP_MATCHES(TRIM(\"date_test\"), '^[0-9]{4}-[0-9]{2}-[0-9]{2}$') THEN TRY_CAST(TRIM(\"date_test\") as DATE) ELSE NULL END"),
+                          ("timestamp_test", datetime.datetime,"CASE WHEN REGEXP_MATCHES(TRIM(\"timestamp_test\"), '^[0-9]{4}-[0-9]{2}-[0-9]{2} [0-9]{2}:[0-9]{2}:[0-9]{2}$') THEN TRY_CAST(TRIM(\"timestamp_test\") as TIMESTAMP) ELSE NULL END"),
+                          ("list_int_field", list[int], "try_cast(list_transform(\"list_int_field\", x -> trim(\"x\")) as BIGINT[])"),
+                          ("basic_model", BasicModel, "try_cast(struct_pack(\"str_field\":= trim(basic_model.\"str_field\"),\"date_field\":= CASE WHEN REGEXP_MATCHES(TRIM(basic_model.\"date_field\"), '^[0-9]{4}-[0-9]{2}-[0-9]{2}$') THEN TRY_CAST(TRIM(basic_model.\"date_field\") as DATE) ELSE NULL END) as STRUCT(str_field VARCHAR, date_field DATE))"),
+                          ("another_model", AnotherModel, "try_cast(struct_pack(\"unique_id\":= trim(another_model.\"unique_id\"),\"basic_models\":= list_transform(another_model.\"basic_models\", x -> struct_pack(\"str_field\":= trim(x.\"str_field\"),\"date_field\":= CASE WHEN REGEXP_MATCHES(TRIM(x.\"date_field\"), '^[0-9]{4}-[0-9]{2}-[0-9]{2}$') THEN TRY_CAST(TRIM(x.\"date_field\") as DATE) ELSE NULL END))) as STRUCT(unique_id BIGINT, basic_models STRUCT(str_field VARCHAR, date_field DATE)[]))")])
 def test_get_duckdb_cast_statement_from_annotation(field_name, field_type, cast_statement):
     assert get_duckdb_cast_statement_from_annotation(field_name, field_type) == cast_statement
 
@@ -172,5 +172,8 @@ def test_use_cast_statements(casting_test_table):
     test_rel = test_rel.project(",".join(casting_statements))
     assert dict(zip(test_rel.columns, test_rel.dtypes)) == {fld.name: get_duckdb_type_from_annotation(fld.annotation) for fld in CastingRecord.__fields__.values()}
     dodgy_date_rec = test_rel.pl()[1].to_dicts()[0]
-    assert not dodgy_date_rec.get("date_test") and not dodgy_date_rec.get("basic_model",{}).get("date_field") and all(not val.get("date_field") for val in dodgy_date_rec.get("another_model",{}).get("basic_models",[]))
+    assert (not dodgy_date_rec.get("date_test") and 
+            not dodgy_date_rec.get("basic_model",{}).get("date_field")
+            and all(not val.get("date_field") for val in dodgy_date_rec.get("another_model",{}).get("basic_models",[]))
+    )