WIP dtype cast

Author: eicchen

pandas/core/reshape/merge.py

@@ -61,6 +61,7 @@
     is_numeric_dtype,
     is_object_dtype,
     is_string_dtype,
+    is_unsigned_integer_dtype,
     needs_i8_conversion,
 )
 from pandas.core.dtypes.dtypes import (
@@ -1717,6 +1718,20 @@ def _maybe_coerce_merge_keys(self) -> None:
             if is_numeric_dtype(lk.dtype) and is_numeric_dtype(rk.dtype):
                 if lk.dtype.kind == rk.dtype.kind:
                     continue
+                
+                if(is_unsigned_integer_dtype(lk.dtype)):
+                    try:
+                        lk = lk.astype(rk.dtype)
+                        continue
+                    except TypeError as err:
+                        raise err
+                elif(is_unsigned_integer_dtype(rk.dtype)):
+                    try:
+                        rk = rk.astype(lk.dtype)
+                        # continue
+                    except TypeError as err:
+                        raise err
+
 
                 if isinstance(lk.dtype, ExtensionDtype) and not isinstance(
                     rk.dtype, ExtensionDtype

pandas/tests/arrays/integer/test_dtypes.py

@@ -297,3 +297,29 @@ def test_astype_boolean():
     result = a.astype("boolean")
     expected = pd.array([True, False, True, True, None], dtype="boolean")
     tm.assert_extension_array_equal(result, expected)
+
+signed_list = ["int64", "float64", "Int64", "Float64"]
+unsigned_list = ["uint64", "UInt64"]
+@pytest.mark.parametrize(
+    "signed_dtype, unsigned_dtype",
+    [(s_dt, u_dt) for s_dt in signed_list for u_dt in unsigned_list]
+)
+def test_unsigned_1to1_validation(signed_dtype, unsigned_dtype):
+    #61688
+    input1 = [1721088000012322083, 1721088047408560273, 1721088047408560451]
+    input2 = 1
+    df1 = pd.DataFrame()
+    df2 = pd.DataFrame()
+    df1["1"] = pd.Series(input1, dtype=signed_dtype)   # Note different types here
+    df2["1"] = pd.Series(input1, dtype=unsigned_dtype)  # Note different types here
+    df1["2"] = input2
+    df2["2"] = input2
+    
+    result = pd.merge(df1, df2, on=["1", "2"], how="left", validate="1:1")
+    expected = pd.merge(df1, df2, on=["1"], how="left", validate="1:1")
+    # expected = pd.merge(df1, df2, on=["1"], how="left", validate="1:1")
+    
+    print(result)
+    print(expected)
+    #TODO add reverse tesetcase
+    tm.assert_frame_equal(result, expected)

test.py

@@ -0,0 +1,16 @@
+import pandas as pd
+import numpy as np
+
+df1 = pd.DataFrame()
+df2 = pd.DataFrame()
+df1["t"] = np.array([1721088000012322083, 1721088047408560273, 1721088047408560451], dtype=np.int64)   # Note different types here
+df2["t"] = np.array([1721088000012322083, 1721088047408560273, 1721088047408560451], dtype=np.uint64)  # Note different types here
+# df1["t"] = np.array([32765,32766, 32767], dtype=np.int16)   # Note different types here
+# df2["t"] = np.array([32765,32766, 32767], dtype=np.uint16)  # Note different types here
+df1["i"] = 1
+df2["i"] = 1
+df1["p"] = [3, 6, 2]
+df2["q"] = [1, 2, 2]
+
+print(pd.merge(df1, df2, on=["i", "t"], how="left", validate="1:1"))
+print(pd.merge(df1, df2, on=["t"], how="left", validate="1:1"))