Fix chunking bug with compound dtypes

.github/workflows/dev-testing.yml

@@ -1,6 +1,11 @@
 name: Dev Branch Testing
 on:
   workflow_dispatch:
+    inputs:
+      python-versions:
+        description: 'List of Python versions to use in matrix, as JSON string'
+        required: true
+        type: string
   workflow_call:
     inputs:
       python-versions:

src/neuroconv/tools/nwb_helpers/_configuration_models/_base_dataset_io.py

@@ -7,6 +7,7 @@
 import h5py
 import numcodecs
 import numpy as np
+import pynwb
 import zarr
 from hdmf import Container
 from hdmf.utils import get_data_shape
@@ -258,9 +259,24 @@ def from_neurodata_object(cls, neurodata_object: Container, dataset_name: Litera
             and `timestamps`, each of which can be configured separately.
         """
         location_in_file = _find_location_in_memory_nwbfile(neurodata_object=neurodata_object, field_name=dataset_name)
-
         candidate_dataset = getattr(neurodata_object, dataset_name)
-        full_shape = get_data_shape(data=candidate_dataset)
+
+        manager = pynwb.get_manager()
+        namespace_catalog = manager.type_map.namespace_catalog
+        for namespace in namespace_catalog.namespaces:
+            try:
+                spec = namespace_catalog.get_spec(namespace, neurodata_object.parent.neurodata_type)
+                break
+            except ValueError:
+                continue
+        spec = spec.get_dataset(neurodata_object.name)
+        spec = spec if spec is not None else {}
+        dtype = spec.get("dtype", None)
+        if isinstance(dtype, list):  # compound dtype
+            full_shape = (len(candidate_dataset),)
+        else:
+            full_shape = get_data_shape(data=candidate_dataset)
+
         dtype = _infer_dtype(dataset=candidate_dataset)
 
         if isinstance(candidate_dataset, GenericDataChunkIterator):