BUG: fix convert_dtypes dropping values from sliced mixed-dtype DataFrames (#64712)

Co-authored-by: moktamd <moktamd@users.noreply.github.com>

Author: moktamd

doc/source/whatsnew/v3.1.0.rst

@@ -166,6 +166,7 @@ Numeric
 Conversion
 ^^^^^^^^^^
 - Fixed :func:`pandas.array` to preserve mask information when converting NumPy masked arrays, converting masked values to missing values (:issue:`63879`).
+- Fixed bug in :meth:`DataFrame.convert_dtypes` where values were dropped from sliced :class:`DataFrame` objects with mixed dtypes when the internal block structure spanned multiple columns (:issue:`64702`)
 - Fixed bug in :meth:`DataFrame.from_records` where ``exclude`` was ignored when ``data`` was an iterator and ``nrows=0`` (:issue:`63774`)
 -
 

pandas/core/internals/blocks.py

@@ -544,7 +544,7 @@ def convert_dtypes(
         for blk in blks:
             # Determine dtype column by column
             sub_blks = (
-                [blk] if blk.ndim == 1 or self.shape[0] == 1 else list(blk._split())
+                [blk] if blk.ndim == 1 or blk.shape[0] == 1 else list(blk._split())
             )
             dtypes = [
                 convert_dtypes(
@@ -558,7 +558,7 @@ def convert_dtypes(
                 )
                 for b in sub_blks
             ]
-            if all(dtype == self.dtype for dtype in dtypes):
+            if all(dtype == blk.dtype for dtype in dtypes):
                 # Avoid block splitting if no dtype changes
                 rbs.append(blk.copy(deep=False))
                 continue

pandas/tests/frame/methods/test_convert_dtypes.py

@@ -240,3 +240,16 @@ def test_convert_dtypes_complex(self):
         )
         result = df.convert_dtypes()
         tm.assert_frame_equal(result, expected)
+
+    def test_convert_dtypes_mixed_column_after_slice(self):
+        # GH#64702
+        df = pd.DataFrame(data=[[1, "a"], [2, "b"], ["c", 3]], columns=["col1", "col2"])
+        df = df.loc[[0, 1]].copy()
+        result = df.convert_dtypes()
+        expected = pd.DataFrame(
+            {
+                "col1": pd.array([1, 2], dtype="Int64"),
+                "col2": pd.array(["a", "b"], dtype="string"),
+            }
+        )
+        tm.assert_frame_equal(result, expected)