BUG: make Series handle dtype='int64' for string array (pandas-dev#4…

…8333)

* make sure conversion is not lossy

* add whatsnew and ref issue in test

* raise earlier

Note the comment at the last line:
# No known cases that get here

* mv whatsnew entry to 1.6.0.rst

doc/source/whatsnew/v1.6.0.rst

@@ -143,7 +143,7 @@ Numeric
 
 Conversion
 ^^^^^^^^^^
--
+- Bug in constructing :class:`Series` with ``int64`` dtype from a string list raising instead of casting (:issue:`44923`)
 -
 
 Strings

pandas/core/dtypes/cast.py

@@ -1851,6 +1851,11 @@ def maybe_cast_to_integer_array(
     # doesn't handle `uint64` correctly.
     arr = np.asarray(arr)
 
+    if np.issubdtype(arr.dtype, str):
+        if (casted.astype(str) == arr).all():
+            return casted
+        raise ValueError(f"string values cannot be losslessly cast to {dtype}")
+
     if is_unsigned_integer_dtype(dtype) and (arr < 0).any():
         raise OverflowError("Trying to coerce negative values to unsigned integers")
 

pandas/tests/series/test_constructors.py

@@ -1881,6 +1881,19 @@ def test_constructor_bool_dtype_missing_values(self):
         expected = Series(True, index=[0], dtype="bool")
         tm.assert_series_equal(result, expected)
 
+    def test_constructor_int64_dtype(self, any_int_dtype):
+        # GH#44923
+        result = Series(["0", "1", "2"], dtype=any_int_dtype)
+        expected = Series([0, 1, 2], dtype=any_int_dtype)
+        tm.assert_series_equal(result, expected)
+
+    def test_constructor_raise_on_lossy_conversion_of_strings(self):
+        # GH#44923
+        with pytest.raises(
+            ValueError, match="string values cannot be losslessly cast to int8"
+        ):
+            Series(["128"], dtype="int8")
+
     def test_constructor_dtype_timedelta_alternative_construct(self):
         # GH#35465
         result = Series([1000000, 200000, 3000000], dtype="timedelta64[ns]")