BUG: to_csv output formatting for datetimes

pandas/core/generic.py

@@ -3146,6 +3146,7 @@ def to_csv(
         line_terminator: Optional[str] = None,
         chunksize: Optional[int] = None,
         date_format: Optional[str] = None,
+        round_milliseconds: bool_t = False,
         doublequote: bool_t = True,
         escapechar: Optional[str] = None,
         decimal: Optional[str] = ".",
@@ -3286,6 +3287,7 @@ def to_csv(
             chunksize=chunksize,
             quotechar=quotechar,
             date_format=date_format,
+            round_milliseconds=round_milliseconds,
             doublequote=doublequote,
             escapechar=escapechar,
             decimal=decimal,

pandas/core/internals/blocks.py

@@ -2189,7 +2189,13 @@ def _can_hold_element(self, element: Any) -> bool:
         return is_valid_nat_for_dtype(element, self.dtype)
 
     def to_native_types(
-        self, slicer=None, na_rep=None, date_format=None, quoting=None, **kwargs
+        self,
+        slicer=None,
+        na_rep=None,
+        date_format=None,
+        quoting=None,
+        round_milliseconds=None,
+        **kwargs,
     ):
         """ convert to our native types format, slicing if desired """
 
@@ -2210,6 +2216,14 @@ def to_native_types(
             format=fmt,
             na_rep=na_rep,
         ).reshape(i8values.shape)
+
+        if round_milliseconds:
+            for i in range(len(result)):
+                milliseconds = result[i][0].split(".")[1]
+                can_round = milliseconds == len(milliseconds) * "0"
+                if can_round:
+                    result[i] = result[i][0].split(".")[0]
+
         return np.atleast_2d(result)
 
     def should_store(self, value):

pandas/io/formats/csvs.py

@@ -49,6 +49,7 @@ def __init__(
         chunksize=None,
         quotechar='"',
         date_format=None,
+        round_milliseconds=None,
         doublequote=True,
         escapechar=None,
         decimal=".",
@@ -94,6 +95,7 @@ def __init__(
         self.line_terminator = line_terminator or os.linesep
 
         self.date_format = date_format
+        self.round_milliseconds = round_milliseconds
 
         self.has_mi_columns = isinstance(obj.columns, ABCMultiIndex)
 
@@ -109,6 +111,7 @@ def __init__(
                     float_format=float_format,
                     date_format=date_format,
                     quoting=self.quoting,
+                    round_milliseconds=round_milliseconds,
                 )
             else:
                 cols = list(cols)
@@ -123,6 +126,7 @@ def __init__(
                 float_format=float_format,
                 date_format=date_format,
                 quoting=self.quoting,
+                round_milliseconds=round_milliseconds,
             )
         else:
             cols = list(cols)
@@ -342,6 +346,7 @@ def _save_chunk(self, start_i: int, end_i: int):
                 decimal=self.decimal,
                 date_format=self.date_format,
                 quoting=self.quoting,
+                round_milliseconds=self.round_milliseconds,
             )
 
             for col_loc, col in zip(b.mgr_locs, d):

pandas/tests/frame/test_to_csv.py

@@ -1,4 +1,5 @@
 import csv
+import datetime
 from io import StringIO
 import os
 
@@ -1356,3 +1357,62 @@ def test_gz_lineend(self):
                 result = f.read().decode("utf-8")
 
         assert result == expected
+
+    def test_to_csv_dropna_format(self):
+        # see gh-29711
+        date_example_string = "1911180945"
+        ts = datetime.datetime.strptime(date_example_string, "%y%m%d%H%M%S")
+        test_json = [{"created_at": "2019-11-18 16:28:42.932887", "foo": "bar",}]
+
+        df = pd.DataFrame(test_json)
+        df["baz"] = ts
+
+        e = ",created_at,foo,baz\n0,2019-11-18 16:28:42.932887,bar,2019-11-18 09:04:05\n"
+        result = df.to_csv()
+        assert result == e
+
+        df["created_at"] = pd.to_datetime(
+            df["created_at"], infer_datetime_format=True, errors="coerce"
+        )
+        result2 = df.to_csv()
+        assert result2 == e
+
+        df = df.dropna(subset=["created_at"])
+        result3 = df.to_csv()
+        assert result3 == e
+
+    def test_to_csv_round_milliseconds(self):
+        # see gh-29711
+        date_example_string = "1911180945"
+        ts = datetime.datetime.strptime(date_example_string, "%y%m%d%H%M%S")
+        test_json = [{"created_at": "2019-11-18 16:28:42.932887", "foo": "bar",}]
+
+        df = pd.DataFrame(test_json)
+        df["baz"] = ts
+
+        df["created_at"] = pd.to_datetime(
+            df["created_at"], infer_datetime_format=True, errors="coerce"
+        )
+
+        df = df.dropna(subset=["created_at"])
+
+        e = ",created_at,foo,baz\n0,2019-11-18 16:28:42.932887,bar,2019-11-18 09:04:05\n"
+        result = df.to_csv(date_format="%Y-%m-%d %H:%M:%S.%f", round_milliseconds=True)
+        assert result == e
+
+        date_example_string = "1911180945"
+        ts = datetime.datetime.strptime(date_example_string, "%y%m%d%H%M%S")
+        test_json = [{"created_at": "2019-11-18 16:28:42.000001", "foo": "bar",}]
+
+        df = pd.DataFrame(test_json)
+        df["baz"] = ts
+
+        df["created_at"] = pd.to_datetime(
+            df["created_at"], infer_datetime_format=True, errors="coerce"
+        )
+
+        df = df.dropna(subset=["created_at"])
+
+        e = ",created_at,foo,baz\n0,2019-11-18 16:28:42.000001,bar,2019-11-18 09:04:05\n"
+        result = df.to_csv(date_format="%Y-%m-%d %H:%M:%S.%f", round_milliseconds=True)
+        assert result == e