[WIP] Fix building datetime datasets with data_type_inc

CHANGELOG.md

@@ -11,6 +11,7 @@
 ### Fixed
 - Added missing validation for dataset reference target types to ensure correct `RefSpec.target_type` matching. @sejalpunwatkar [#1429](https://github.com/hdmf-dev/hdmf/pull/1429)
 - Fixed reading a `DynamicTable` that contains a named link to a `VectorData` (e.g., `MeaningsTable.target`). The link target was being picked up as an extra column, causing a `"Columns must be the same length"` error when the target column's row count differed from the table's own row count. @rly [#1445](https://github.com/hdmf-dev/hdmf/pull/1445)
+- Fixed building datasets with `dtype: isodatetime` when the spec uses `data_type_inc` (e.g., `VectorData`-extending columns). `datetime`/`date` values are now converted to ISO strings before dtype resolution instead of failing with `"Expected unicode or ascii string"`. @rly [#1314](https://github.com/hdmf-dev/hdmf/pull/1314)
 
 
 ## HDMF 5.1.0 (March 24, 2026)

src/hdmf/build/objectmapper.py

@@ -627,7 +627,11 @@ def get_attr_value(self, **kwargs):
             if isinstance(attr_val, TermSetWrapper):
                 attr_val = attr_val.value
             if attr_val is not None:
-                attr_val = self.__convert_string(attr_val, spec)
+                if not isinstance(attr_val, Data):
+                    # Data containers are unwrapped and string-converted later in build(),
+                    # after the Data sub-builder is created. Converting here would operate
+                    # on the Data wrapper itself and produce the wrong result.
+                    attr_val = self.__convert_string(attr_val, spec)
                 spec_dt = self.__get_data_type(spec)
                 if spec_dt is not None:
                     try:
@@ -660,6 +664,7 @@ def __apply_string_type(value, string_type):
             # NOTE: if a user passes a h5py.Dataset that is not wrapped with a hdmf.utils.StrDataset,
             # then this conversion may not be correct. Users should unpack their string h5py.Datasets
             # into a numpy array (or wrap them in StrDataset) before passing them to a container object.
+            # NOTE: this will convert datasets and arrays to lists of lists.
             if hasattr(value, '__iter__') and not isinstance(value, (str, bytes)):
                 return [__apply_string_type(item, string_type) for item in value]
             else:
@@ -673,8 +678,7 @@ def __apply_string_type(value, string_type):
                 else:
                     ret = str(value)
         elif isinstance(spec, DatasetSpec):
-            # TODO: make sure we can handle specs with data_type_inc set
-            if spec.data_type_inc is None and spec.dtype is not None:
+            if spec.dtype is not None:
                 string_type = None
                 if 'text' in spec.dtype:
                     string_type = str
@@ -849,6 +853,7 @@ def build(self, **kwargs):
                                 data = container.data.value
                             else:
                                 data = container.data
+                            data = self.__convert_string(data, spec)
                             bldr_data, dtype = self.convert_dtype(spec, data, spec_dtype=spec_dtype)
                         except Exception as ex:
                             msg = f"could not resolve dtype for {type(container).__name__} '{container.name}'"

tests/unit/build_tests/mapper_tests/test_build_datetime.py

@@ -1,5 +1,5 @@
 from hdmf.utils import docval, getargs
-from hdmf import Container
+from hdmf import Container, Data
 from hdmf.spec import GroupSpec, DatasetSpec
 from hdmf.testing import TestCase
 from datetime import datetime, date
@@ -25,6 +25,38 @@ def data(self):
         return self.__data
 
 
+class Column(Data):
+
+    @docval({'name': 'name', 'type': str, 'doc': 'the name of this Column'},
+            {'name': 'data', 'type': ('data', 'array_data'), 'doc': 'some data'})
+    def __init__(self, **kwargs):
+        name, data = getargs('name', 'data', kwargs)
+        super().__init__(name=name, data=data)
+
+    @property
+    def data_type(self):
+        return 'Column'
+
+
+class BarWithColumnData(Container):
+
+    @docval({'name': 'name', 'type': str, 'doc': 'the name of this Bar'},
+            {'name': 'data', 'type': Column, 'doc': 'some data'})
+    def __init__(self, **kwargs):
+        name, data = getargs('name', 'data', kwargs)
+        super().__init__(name=name)
+        self.__data = data
+        data.parent = self
+
+    @property
+    def data_type(self):
+        return 'Bar'
+
+    @property
+    def data(self):
+        return self.__data
+
+
 class TestBuildDatasetDateTime(TestCase):
     """Test that building a dataset with dtype isodatetime works with datetime and date objects."""
 
@@ -83,3 +115,55 @@ def test_date_array(self):
         ret = builder.get('data')
         assert ret.data == [b'2023-07-09', b'2023-07-10']
         assert ret.dtype == 'ascii'
+
+
+    def test_datetime_array_ext_spec(self):
+        column_spec = DatasetSpec(data_type_def='Column', doc='an example dataset', dims=(None,))
+        bar_spec = GroupSpec(
+            doc='A test group specification with a data type',
+            data_type_def='Bar',
+            datasets=[
+                DatasetSpec(
+                    data_type_inc='Column',
+                    doc='an example dataset',
+                    name='data',
+                    dtype='isodatetime',
+                ),
+            ],
+        )
+        type_map = create_test_type_map([bar_spec, column_spec], {'Bar': BarWithColumnData, 'Column': Column})
+
+        bar_inst = BarWithColumnData(
+            name='my_bar',
+            data=Column(name='data', data=[datetime(2023, 7, 9), datetime(2023, 7, 10)]),
+        )
+        builder = type_map.build(bar_inst)
+        ret = builder.get('data')
+        assert ret.data == [b'2023-07-09T00:00:00', b'2023-07-10T00:00:00']
+        assert ret.dtype == 'ascii'
+
+    def test_date_array_ext_spec(self):
+        # This mirrors the traceback in issue #1311, which used datetime.date on a VectorData extension.
+        column_spec = DatasetSpec(data_type_def='Column', doc='an example dataset', dims=(None,))
+        bar_spec = GroupSpec(
+            doc='A test group specification with a data type',
+            data_type_def='Bar',
+            datasets=[
+                DatasetSpec(
+                    data_type_inc='Column',
+                    doc='an example dataset',
+                    name='data',
+                    dtype='isodatetime',
+                ),
+            ],
+        )
+        type_map = create_test_type_map([bar_spec, column_spec], {'Bar': BarWithColumnData, 'Column': Column})
+
+        bar_inst = BarWithColumnData(
+            name='my_bar',
+            data=Column(name='data', data=[date(2023, 7, 9), date(2023, 7, 10)]),
+        )
+        builder = type_map.build(bar_inst)
+        ret = builder.get('data')
+        assert ret.data == [b'2023-07-09', b'2023-07-10']
+        assert ret.dtype == 'ascii'