From d412badc5bb88013688234dbc730f62283c132d7 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Sat, 3 Jan 2026 00:54:28 +0000
Subject: [PATCH 1/4] Initial plan


From 5806429a97b243952ca9d921cda9ab5719006912 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Sat, 3 Jan 2026 01:00:28 +0000
Subject: [PATCH 2/4] Fix compound dtype slicing - support numeric indexing

Co-authored-by: rly <310197+rly@users.noreply.github.com>
---
 lindi/LindiH5pyFile/LindiH5pyDataset.py | 14 +++++--
 tests/test_lindi_h5py_file.py           | 53 +++++++++++++++++++++++++
 2 files changed, 64 insertions(+), 3 deletions(-)

diff --git a/lindi/LindiH5pyFile/LindiH5pyDataset.py b/lindi/LindiH5pyFile/LindiH5pyDataset.py
index 1ec9da3..f017d4c 100644
--- a/lindi/LindiH5pyFile/LindiH5pyDataset.py
+++ b/lindi/LindiH5pyFile/LindiH5pyDataset.py
@@ -214,9 +214,17 @@ def _get_item_for_zarr(self, zarr_array: zarr.Array, selection: Any):
                 )
                 return ret
             else:
-                raise TypeError(
-                    f"Compound dataset {self.name} does not support selection with {selection}"
-                )
+                # Numeric slicing (e.g., [:], [0], [0:2])
+                # Get the raw data from zarr (will be object array of lists)
+                raw_data = zarr_array[selection]
+                # Convert to structured array with compound dtype
+                if isinstance(raw_data, np.ndarray):
+                    # Multiple elements - convert each list to tuple
+                    tuples = [tuple(item) for item in raw_data]
+                    return np.array(tuples, dtype=self._compound_dtype)
+                else:
+                    # Single element - raw_data is a list
+                    return np.array(tuple(raw_data), dtype=self._compound_dtype)[()]
 
         # We use zarr's slicing, except in the case of a scalar dataset
         if self.ndim == 0:
diff --git a/tests/test_lindi_h5py_file.py b/tests/test_lindi_h5py_file.py
index 032f8e6..b8058c0 100644
--- a/tests/test_lindi_h5py_file.py
+++ b/tests/test_lindi_h5py_file.py
@@ -357,6 +357,59 @@ def test_create_dataset():
             assert ds.shape == (3,)
 
 
+def test_compound_dtype_slicing():
+    """Test that compound dtype datasets support numeric slicing (e.g., [:], [0], [0:2])"""
+    import numpy as np
+    import zarr
+    
+    with tempfile.TemporaryDirectory() as tmpdir:
+        dirname = f'{tmpdir}/test.zarr'
+        store = zarr.DirectoryStore(dirname)
+        zarr.group(store=store)
+        
+        # Write compound dataset
+        with lindi.LindiH5pyFile.from_zarr_store(store, mode='r+') as h5f:
+            compound_dtype = np.dtype([('x', np.int32), ('y', np.float64)])
+            data = np.array([(1, 2.2), (3, 4.4), (5, 6.6)], dtype=compound_dtype)
+            h5f.create_dataset('dset_compound', data=data)
+        
+        # Read and test slicing
+        with lindi.LindiH5pyFile.from_zarr_store(store, mode='r') as h5f:
+            dset = h5f['dset_compound']
+            
+            # Test full slice
+            result = dset[:]
+            assert result.shape == (3,)
+            assert result.dtype == compound_dtype
+            assert result[0]['x'] == 1
+            assert result[0]['y'] == 2.2
+            assert result[1]['x'] == 3
+            assert result[1]['y'] == 4.4
+            
+            # Test single element access
+            result = dset[0]
+            assert isinstance(result, np.void)
+            assert result['x'] == 1
+            assert result['y'] == 2.2
+            
+            # Test partial slice
+            result = dset[0:2]
+            assert result.shape == (2,)
+            assert result.dtype == compound_dtype
+            assert result[0]['x'] == 1
+            assert result[1]['x'] == 3
+            
+            # Test negative indexing
+            result = dset[-1]
+            assert result['x'] == 5
+            assert result['y'] == 6.6
+            
+            # Test field selection (should still work)
+            result = dset['x'][:]
+            assert result.shape == (3,)
+            assert np.array_equal(result, [1, 3, 5])
+
+
 def create_example_h5_file(fname):
     with h5py.File(fname, 'w') as f:
         f.attrs['attr1'] = 'value1'

From de96c54e18cbf7b445dad0e8117075bc59dcf7b2 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Sat, 3 Jan 2026 01:02:32 +0000
Subject: [PATCH 3/4] Fix linting issues - remove trailing whitespace

Co-authored-by: rly <310197+rly@users.noreply.github.com>
---
 tests/test_lindi_h5py_file.py | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/tests/test_lindi_h5py_file.py b/tests/test_lindi_h5py_file.py
index b8058c0..ea4e4ba 100644
--- a/tests/test_lindi_h5py_file.py
+++ b/tests/test_lindi_h5py_file.py
@@ -361,22 +361,22 @@ def test_compound_dtype_slicing():
     """Test that compound dtype datasets support numeric slicing (e.g., [:], [0], [0:2])"""
     import numpy as np
     import zarr
-    
+
     with tempfile.TemporaryDirectory() as tmpdir:
         dirname = f'{tmpdir}/test.zarr'
         store = zarr.DirectoryStore(dirname)
         zarr.group(store=store)
-        
+
         # Write compound dataset
         with lindi.LindiH5pyFile.from_zarr_store(store, mode='r+') as h5f:
             compound_dtype = np.dtype([('x', np.int32), ('y', np.float64)])
             data = np.array([(1, 2.2), (3, 4.4), (5, 6.6)], dtype=compound_dtype)
             h5f.create_dataset('dset_compound', data=data)
-        
+
         # Read and test slicing
         with lindi.LindiH5pyFile.from_zarr_store(store, mode='r') as h5f:
             dset = h5f['dset_compound']
-            
+
             # Test full slice
             result = dset[:]
             assert result.shape == (3,)
@@ -385,25 +385,25 @@ def test_compound_dtype_slicing():
             assert result[0]['y'] == 2.2
             assert result[1]['x'] == 3
             assert result[1]['y'] == 4.4
-            
+
             # Test single element access
             result = dset[0]
             assert isinstance(result, np.void)
             assert result['x'] == 1
             assert result['y'] == 2.2
-            
+
             # Test partial slice
             result = dset[0:2]
             assert result.shape == (2,)
             assert result.dtype == compound_dtype
             assert result[0]['x'] == 1
             assert result[1]['x'] == 3
-            
+
             # Test negative indexing
             result = dset[-1]
             assert result['x'] == 5
             assert result['y'] == 6.6
-            
+
             # Test field selection (should still work)
             result = dset['x'][:]
             assert result.shape == (3,)

From 1b04ba3c3e98b4b05dd90251f66b4909f3de8886 Mon Sep 17 00:00:00 2001
From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com>
Date: Sat, 3 Jan 2026 01:05:12 +0000
Subject: [PATCH 4/4] Address code review feedback - improve type handling and
 imports

Co-authored-by: rly <310197+rly@users.noreply.github.com>
---
 lindi/LindiH5pyFile/LindiH5pyDataset.py | 11 ++++++++---
 tests/test_lindi_h5py_file.py           |  4 ++--
 2 files changed, 10 insertions(+), 5 deletions(-)

diff --git a/lindi/LindiH5pyFile/LindiH5pyDataset.py b/lindi/LindiH5pyFile/LindiH5pyDataset.py
index f017d4c..e824a3d 100644
--- a/lindi/LindiH5pyFile/LindiH5pyDataset.py
+++ b/lindi/LindiH5pyFile/LindiH5pyDataset.py
@@ -220,11 +220,16 @@ def _get_item_for_zarr(self, zarr_array: zarr.Array, selection: Any):
                 # Convert to structured array with compound dtype
                 if isinstance(raw_data, np.ndarray):
                     # Multiple elements - convert each list to tuple
-                    tuples = [tuple(item) for item in raw_data]
+                    tuples = [tuple(item) if isinstance(item, (list, np.ndarray)) else item for item in raw_data]
                     return np.array(tuples, dtype=self._compound_dtype)
-                else:
-                    # Single element - raw_data is a list
+                elif isinstance(raw_data, (list, tuple)):
+                    # Single element - raw_data is a list or tuple
                     return np.array(tuple(raw_data), dtype=self._compound_dtype)[()]
+                else:
+                    # Unexpected type - this shouldn't happen, but handle gracefully
+                    raise TypeError(
+                        f"Unexpected data type from zarr array for compound dataset {self.name}: {type(raw_data)}"
+                    )
 
         # We use zarr's slicing, except in the case of a scalar dataset
         if self.ndim == 0:
diff --git a/tests/test_lindi_h5py_file.py b/tests/test_lindi_h5py_file.py
index ea4e4ba..17a13f5 100644
--- a/tests/test_lindi_h5py_file.py
+++ b/tests/test_lindi_h5py_file.py
@@ -2,6 +2,8 @@
 import os
 import pytest
 import h5py
+import numpy as np
+import zarr
 import lindi
 from .utils import assert_h5py_files_equal
 
@@ -359,8 +361,6 @@ def test_create_dataset():
 
 def test_compound_dtype_slicing():
     """Test that compound dtype datasets support numeric slicing (e.g., [:], [0], [0:2])"""
-    import numpy as np
-    import zarr
 
     with tempfile.TemporaryDirectory() as tmpdir:
         dirname = f'{tmpdir}/test.zarr'