Fix pandas-dev#25099 set na_rep values before converting to string to prevent data truncation

Author: kinow

pandas/core/internals/blocks.py

@@ -749,7 +749,10 @@ def to_native_types(self, slicer=None, na_rep="nan", quoting=None, **kwargs):
         mask = isna(values)
 
         if not self.is_object and not quoting:
-            values = values.astype(str)
+            if na_rep and isinstance(na_rep, str):
+                values = values.astype("<U{length}".format(length=len(na_rep)))
+            else:
+                values = values.astype(str)
         else:
             values = np.array(values, dtype="object")
 

pandas/tests/io/formats/test_to_csv.py

@@ -515,3 +515,19 @@ def test_to_csv_compression(self, compression_only, read_infer, to_infer):
             df.to_csv(path, compression=to_compression)
             result = pd.read_csv(path, index_col=0, compression=read_compression)
             tm.assert_frame_equal(result, df)
+
+    def test_to_csv_na_rep_long_string(self, capsys):
+        # see gh-25099
+        df = pd.DataFrame({"c": [float('nan')] * 3})
+        df = df.astype("Int64")
+        expected_rows = ['c',
+                         'mynull',
+                         'mynull',
+                         'mynull']
+        expected_ascii = tm.convert_rows_list_to_csv_str(expected_rows)
+
+        df.to_csv(sys.stdout, index=False, na_rep='mynull', encoding='ascii')
+        captured = capsys.readouterr()
+
+        assert captured.out == expected_ascii
+        assert not sys.stdout.closed