astropy: No longer able to read BinnedTimeSeries with datetime column saved as ECSV after upgrading from 4.2.1 -> 5.0+

Description

Hi, This commit merged in PR #11569 breaks my ability to read an ECSV file created using Astropy v 4.2.1, BinnedTimeSeries class’s write method, which has a datetime64 column. Downgrading astropy back to 4.2.1 fixes the issue because the strict type checking in line 177 of ecsv.py is not there.

Is there a reason why this strict type checking was added to ECSV? Is there a way to preserve reading and writing of ECSV files created with BinnedTimeSeries across versions? I am happy to make a PR on this if the strict type checking is allowed to be scaled back or we can add datetime64 as an allowed type.

Expected behavior

The file is read into a BinnedTimeSeries object from ecsv file without error.

Actual behavior

ValueError is produced and the file is not read because ECSV.py does not accept the datetime64 column. ValueError: datatype 'datetime64' of column 'time_bin_start' is not in allowed values ('bool', 'int8', 'int16', 'int32', 'int64', 'uint8', 'uint16', 'uint32', 'uint64', 'float16', 'float32', 'float64', 'float128', 'string')

Steps to Reproduce

The file is read using:
BinnedTimeSeries.read('<file_path>', format='ascii.ecsv') which gives a long error.

The file in question is a binned time series created by astropy.timeseries.aggregate_downsample. which itself is a binned version of an astropy.timeseries.TimeSeries instance with some TESS data. (loaded via TimeSeries.from_pandas(Tess.set_index(‘datetime’)). I.e., it has a datetime64 index. The file was written using the classes own .write method in Astropy V4.2.1 from an instance of said class:
myBinnedTimeSeries.write('<file_path>',format='ascii.ecsv',overwrite=True)

I’ll attach a concatenated version of the file (as it contains private data). However, the relevant part from the header is on line 4:

# %ECSV 0.9
# ---
# datatype:
# - {name: time_bin_start, datatype: datetime64}

as you can see, the datatype is datetime64. This works fine with ECSV V0.9 but not V1.0 as some sort of strict type checking was added.

Full error log:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
Input In [3], in <module>
---> 49 tsrbin = BinnedTimeSeries.read('../Photometry/tsr_bin.dat', format='ascii.ecsv')

File ~/Apps/miniconda3/envs/py310_latest/lib/python3.10/site-packages/astropy/timeseries/binned.py:285, in BinnedTimeSeries.read(self, filename, time_bin_start_column, time_bin_end_column, time_bin_size_column, time_bin_size_unit, time_format, time_scale, format, *args, **kwargs)
    230 """
    231 Read and parse a file and returns a `astropy.timeseries.BinnedTimeSeries`.
    232 
   (...)
    279 
    280 """
    282 try:
    283 
    284     # First we try the readers defined for the BinnedTimeSeries class
--> 285     return super().read(filename, format=format, *args, **kwargs)
    287 except TypeError:
    288 
    289     # Otherwise we fall back to the default Table readers
    291     if time_bin_start_column is None:

File ~/Apps/miniconda3/envs/py310_latest/lib/python3.10/site-packages/astropy/table/connect.py:62, in TableRead.__call__(self, *args, **kwargs)
     59 units = kwargs.pop('units', None)
     60 descriptions = kwargs.pop('descriptions', None)
---> 62 out = self.registry.read(cls, *args, **kwargs)
     64 # For some readers (e.g., ascii.ecsv), the returned `out` class is not
     65 # guaranteed to be the same as the desired output `cls`.  If so,
     66 # try coercing to desired class without copying (io.registry.read
     67 # would normally do a copy).  The normal case here is swapping
     68 # Table <=> QTable.
     69 if cls is not out.__class__:

File ~/Apps/miniconda3/envs/py310_latest/lib/python3.10/site-packages/astropy/io/registry/core.py:199, in UnifiedInputRegistry.read(self, cls, format, cache, *args, **kwargs)
    195     format = self._get_valid_format(
    196         'read', cls, path, fileobj, args, kwargs)
    198 reader = self.get_reader(format, cls)
--> 199 data = reader(*args, **kwargs)
    201 if not isinstance(data, cls):
    202     # User has read with a subclass where only the parent class is
    203     # registered.  This returns the parent class, so try coercing
    204     # to desired subclass.
    205     try:

File ~/Apps/miniconda3/envs/py310_latest/lib/python3.10/site-packages/astropy/io/ascii/connect.py:18, in io_read(format, filename, **kwargs)
     16     format = re.sub(r'^ascii\.', '', format)
     17     kwargs['format'] = format
---> 18 return read(filename, **kwargs)

File ~/Apps/miniconda3/envs/py310_latest/lib/python3.10/site-packages/astropy/io/ascii/ui.py:376, in read(table, guess, **kwargs)
    374     else:
    375         reader = get_reader(**new_kwargs)
--> 376         dat = reader.read(table)
    377         _read_trace.append({'kwargs': copy.deepcopy(new_kwargs),
    378                             'Reader': reader.__class__,
    379                             'status': 'Success with specified Reader class '
    380                                       '(no guessing)'})
    382 # Static analysis (pyright) indicates `dat` might be left undefined, so just
    383 # to be sure define it at the beginning and check here.

File ~/Apps/miniconda3/envs/py310_latest/lib/python3.10/site-packages/astropy/io/ascii/core.py:1343, in BaseReader.read(self, table)
   1340 self.header.update_meta(self.lines, self.meta)
   1342 # Get the table column definitions
-> 1343 self.header.get_cols(self.lines)
   1345 # Make sure columns are valid
   1346 self.header.check_column_names(self.names, self.strict_names, self.guessing)

File ~/Apps/miniconda3/envs/py310_latest/lib/python3.10/site-packages/astropy/io/ascii/ecsv.py:177, in EcsvHeader.get_cols(self, lines)
    175 col.dtype = header_cols[col.name]['datatype']
    176 if col.dtype not in ECSV_DATATYPES:
--> 177     raise ValueError(f'datatype {col.dtype!r} of column {col.name!r} '
    178                      f'is not in allowed values {ECSV_DATATYPES}')
    180 # Subtype is written like "int64[2,null]" and we want to split this
    181 # out to "int64" and [2, None].
    182 subtype = col.subtype

ValueError: datatype 'datetime64' of column 'time_bin_start' is not in allowed values ('bool', 'int8', 'int16', 'int32', 'int64', 'uint8', 'uint16', 'uint32', 'uint64', 'float16', 'float32', 'float64', 'float128', 'string')

System Details

(For the version that does not work) Python 3.10.2 | packaged by conda-forge | (main, Feb 1 2022, 19:28:35) [GCC 9.4.0] Numpy 1.22.2 pyerfa 2.0.0.1 astropy 5.0.1 Scipy 1.8.0 Matplotlib 3.5.1

(For the version that does work) Python 3.7.11 (default, Jul 27 2021, 14:32:16) [GCC 7.5.0] Numpy 1.20.3 pyerfa 2.0.0.1 astropy 4.2.1 Scipy 1.7.0 Matplotlib 3.4.2