fix: datetime formatting for Pandas tables (#3383)

Always include H:M:S.f. When the time is excluded our frontend doesn't
know how to interpret the timestamp.

Fixes #3352 

(Separately I noticed the table component doesn't display nonzero
microseconds. Microseconds are useful for HFT and real-time control.)

<img width="939" alt="image"
src="https://github.com/user-attachments/assets/4e4db1d0-9ff9-4c9e-9a6c-3373f592939c"
/>

marimo/_plugins/ui/_impl/tables/pandas_table.py

@@ -45,9 +45,14 @@ def to_csv(
                 self, format_mapping: Optional[FormatMapping] = None
             ) -> bytes:
                 has_headers = len(self.get_row_headers()) > 0
+                # Pandas omits H:M:S for datetimes when H:M:S is identically
+                # 0; this doesn't play well with our frontend table component,
+                # so we use an explicit date format.
                 return (
                     self.apply_formatting(format_mapping)
-                    ._original_data.to_csv(index=has_headers)
+                    ._original_data.to_csv(
+                        index=has_headers, date_format="%Y-%m-%d %H:%M:%S"
+                    )
                     .encode("utf-8")
                 )
 

tests/_plugins/ui/_impl/charts/test_altair_transformers.py

@@ -201,9 +201,9 @@ def test_data_to_json_string_with_special_characters(
             "float": [1.1, 2.2, 3.3],
             "bool": [True, False, True],
             "datetime": [
-                datetime.datetime(2023, 1, 1),
-                datetime.datetime(2023, 1, 2),
-                datetime.datetime(2023, 1, 3),
+                datetime.datetime(2023, 1, 1, 1),
+                datetime.datetime(2023, 1, 2, 1),
+                datetime.datetime(2023, 1, 3, 1),
             ],
             "category": ["a", "b", "c"],
         },
@@ -213,6 +213,8 @@ def test_data_to_json_string_with_special_characters(
 def test_data_to_csv_string_with_different_dtypes(df: IntoDataFrame):
     result = _data_to_csv_string(df)
     assert isinstance(result, str)
+    # NB: These can differ based on datetime canonicalization --- our table
+    # manager always includes H:M:S:.f for pandas, but narwhals doesn't.
     assert result == dataframe_to_csv(df)
 
 

tests/_plugins/ui/_impl/tables/snapshots/ibis.csv

@@ -1,4 +1,4 @@
 strings,bool,int,float,date,datetime,time,nulls
-a,True,1,1.0,2021-01-01,2021-01-01,01:02:03,
-b,False,2,2.0,2021-01-02,2021-01-02,04:05:06,data
-c,True,3,3.0,2021-01-03,2021-01-03,07:08:09,
+a,True,1,1.0,2021-01-01,2021-01-01 00:00:00,01:02:03,
+b,False,2,2.0,2021-01-02,2021-01-02 00:00:00,04:05:06,data
+c,True,3,3.0,2021-01-03,2021-01-03 00:00:00,07:08:09,

tests/_plugins/ui/_impl/tables/snapshots/pandas.csv

@@ -1,4 +1,4 @@
 strings,bool,int,float,datetime,date,struct,list,nulls,category,set,imaginary,time,duration,mixed_list
-a,True,1,1.0,2021-01-01,2021-01-01,"{'a': 1, 'b': 2}","[1, 2]",,cat,"{1, 2}",(1+2j),12:30:00,1 days,"[1, 'two']"
-b,False,2,2.0,2021-01-02,2021-01-02,"{'a': 3, 'b': 4}","[3, 4]",data,dog,"{3, 4}",(3+4j),13:45:00,2 days,"[3.0, False]"
-c,True,3,3.0,2021-01-03,2021-01-03,"{'a': 5, 'b': 6}","[5, 6]",,mouse,"{5, 6}",(5+6j),14:15:00,3 days,"[None, datetime.datetime(2021, 1, 1, 0, 0)]"
+a,True,1,1.0,2021-01-01 00:00:00,2021-01-01,"{'a': 1, 'b': 2}","[1, 2]",,cat,"{1, 2}",(1+2j),12:30:00,1 days,"[1, 'two']"
+b,False,2,2.0,2021-01-02 00:00:00,2021-01-02,"{'a': 3, 'b': 4}","[3, 4]",data,dog,"{3, 4}",(3+4j),13:45:00,2 days,"[3.0, False]"
+c,True,3,3.0,2021-01-03 00:00:00,2021-01-03,"{'a': 5, 'b': 6}","[5, 6]",,mouse,"{5, 6}",(5+6j),14:15:00,3 days,"[None, datetime.datetime(2021, 1, 1, 0, 0)]"

tests/_plugins/ui/_impl/tables/test_pandas_table.py

@@ -120,9 +120,21 @@ def test_package_name(self) -> None:
         assert self.factory.package_name() == "pandas"
 
     def test_to_csv(self) -> None:
-        expected_csv = self.data.to_csv(index=False).encode("utf-8")
+        expected_csv = self.data.to_csv(
+            index=False, date_format="%Y-%m-%d %H:%M:%S"
+        ).encode("utf-8")
         assert self.manager.to_csv() == expected_csv
 
+    def test_to_csv_datetime(self) -> None:
+        D = pd.to_datetime("2024-12-17", errors="coerce")
+
+        data = {
+            "D timestamp": [D],
+        }
+        df = pd.DataFrame(data)
+        manager = PandasTableManagerFactory.create()(df)
+        assert "2024-12-17 00:00:00" in manager.to_csv().decode("utf-8")
+
     def test_to_csv_complex(self) -> None:
         complex_data = self.get_complex_data()
         data = complex_data.to_csv()