Fix a more serious issue where the object was not being reset to the new query compiler

Author: sfc-gh-jkew

modin/core/storage_formats/pandas/query_compiler_caster.py

@@ -290,11 +290,17 @@ def cast_to_qc(arg):
         if qc is None or qc is type(current_qc):
             return obj(*args, **kwargs)
 
-        # we need to cast current_qc to a new query compiler
+        # we need to cast current_qc to a new query compiler,
+        # then lookup the same function on the new query compiler
+        # we need to also drop the first argument to the original
+        # args since it references self on the original argument
+        # call
         if qc != current_qc:
             data_cls = current_qc._modin_frame
-            return qc.from_pandas(current_qc.to_pandas(), data_cls)
-        # need to find the new function for obj
+            new_qc = qc.from_pandas(current_qc.to_pandas(), data_cls)
+            obj_new = getattr(new_qc, obj.__name__)
+            return obj_new(*args[1:], **kwargs)
+
         return obj(*args, **kwargs)
 
     return cast_args

modin/tests/pandas/native_df_interoperability/test_compiler_caster.py

@@ -275,3 +275,15 @@ def test_qc_default_self_cost(default_df, default2_df):
     assert (
         default_df.qc_engine_switch_cost(type(default_df)) is QCCoercionCost.COST_ZERO
     )
+
+
+def test_qc_casting_moved_data(pico_df, cloud_df):
+    pico_df1 = pd.DataFrame(query_compiler=pico_df)
+    cloud_df1 = pd.DataFrame(query_compiler=cloud_df)
+    native_cdf2 = cloud_df1._to_pandas()
+    native_pdf2 = pico_df1._to_pandas()
+    expected = native_cdf2 + native_pdf2
+    df1 = pico_df1 + cloud_df1
+    df2 = cloud_df1 + pico_df1
+    assert df1._to_pandas().equals(expected)
+    assert df2._to_pandas().equals(expected)