From 632c7f040a4e10c8312e65762c08f3e535c91743 Mon Sep 17 00:00:00 2001
From: Aimee Barciauskas <aimee@developmentseed.org>
Date: Fri, 17 Apr 2026 09:28:38 -0700
Subject: [PATCH 1/8] Add failing test

---
 virtualizarr/tests/test_parsers/conftest.py       | 15 +++++++++++++++
 .../tests/test_parsers/test_hdf/test_hdf.py       | 12 ++++++++++++
 2 files changed, 27 insertions(+)

diff --git a/virtualizarr/tests/test_parsers/conftest.py b/virtualizarr/tests/test_parsers/conftest.py
index 23d30a4f..ee5d1490 100644
--- a/virtualizarr/tests/test_parsers/conftest.py
+++ b/virtualizarr/tests/test_parsers/conftest.py
@@ -479,6 +479,21 @@ def big_endian_dtype_hdf5_file(tmpdir):
     return filepath
 
 
+@pytest.fixture
+def singleton_padded_dimension_hdf5_file(tmp_path: Path) -> str:
+    """HDF5 file mimicking MATLAB layout: a 2D array (N, M) plus coordinate
+    arrays shaped (N, 1) and (1, M)."""
+    N, M = 10, 5
+    filepath = str(tmp_path / "singleton_dimension_layout.nc")
+
+    with h5py.File(filepath, "w") as f:
+        f.create_dataset(name="data", data=np.random.random((N, M)))
+        f.create_dataset(name="row_coord", data=np.random.random((N, 1)))
+        f.create_dataset(name="col_coord", data=np.random.random((1, M)))
+
+    return f"file://{filepath}"
+
+
 @pytest.fixture()
 def dmrpp_xml_simple():
     """Return a minimal valid DMR++ XML string for testing."""
diff --git a/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py b/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py
index 9854174f..4a10b101 100644
--- a/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py
+++ b/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py
@@ -41,6 +41,18 @@ def test_chunked_roundtrip(self, chunked_roundtrip_hdf5_url):
         manifest_store = manifest_store_from_hdf_url(chunked_roundtrip_hdf5_url)
         assert manifest_store._group.arrays["var2"].manifest.shape_chunk_grid == (2, 8)
 
+    def test_singleton_dimensions_squeezed(self, singleton_padded_dimension_hdf5_file):
+        manifest_store = manifest_store_from_hdf_url(
+            singleton_padded_dimension_hdf5_file
+        )
+        assert manifest_store._group.arrays["data"].manifest.shape_chunk_grid == (10, 5)
+        assert manifest_store._group.arrays["row_coord"].manifest.shape_chunk_grid == (
+            10,
+        )
+        assert manifest_store._group.arrays["col_coord"].manifest.shape_chunk_grid == (
+            5,
+        )
+
 
 @requires_hdf5plugin
 @requires_imagecodecs

From 696c05107aa08622630581891f2cc7369c1d3605 Mon Sep 17 00:00:00 2001
From: Aimee Barciauskas <aimee@developmentseed.org>
Date: Fri, 17 Apr 2026 11:14:55 -0700
Subject: [PATCH 2/8] Test should test shape not chunk grid

---
 virtualizarr/tests/test_parsers/test_hdf/test_hdf.py | 10 +++-------
 1 file changed, 3 insertions(+), 7 deletions(-)

diff --git a/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py b/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py
index 4a10b101..2543c8bf 100644
--- a/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py
+++ b/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py
@@ -45,13 +45,9 @@ def test_singleton_dimensions_squeezed(self, singleton_padded_dimension_hdf5_fil
         manifest_store = manifest_store_from_hdf_url(
             singleton_padded_dimension_hdf5_file
         )
-        assert manifest_store._group.arrays["data"].manifest.shape_chunk_grid == (10, 5)
-        assert manifest_store._group.arrays["row_coord"].manifest.shape_chunk_grid == (
-            10,
-        )
-        assert manifest_store._group.arrays["col_coord"].manifest.shape_chunk_grid == (
-            5,
-        )
+        assert manifest_store._group.arrays["data"].shape == (10, 5)
+        assert manifest_store._group.arrays["row_coord"].shape == (10,)
+        assert manifest_store._group.arrays["col_coord"].shape == (5,)
 
 
 @requires_hdf5plugin

From 03e3a1ca126d364e89ddaf8f7aac4a0ce5358a25 Mon Sep 17 00:00:00 2001
From: Aimee Barciauskas <aimee@developmentseed.org>
Date: Fri, 17 Apr 2026 11:15:54 -0700
Subject: [PATCH 3/8] Add function for squeeze indices and use it to squeeze
 path, offset, lenght

---
 virtualizarr/parsers/hdf/hdf.py | 26 ++++++++++++++++++++++----
 1 file changed, 22 insertions(+), 4 deletions(-)

diff --git a/virtualizarr/parsers/hdf/hdf.py b/virtualizarr/parsers/hdf/hdf.py
index a180a502..1b371603 100644
--- a/virtualizarr/parsers/hdf/hdf.py
+++ b/virtualizarr/parsers/hdf/hdf.py
@@ -35,6 +35,11 @@
     from h5py import Group as H5Group
 
 
+def _squeeze_indices(chunks: tuple) -> list[int]:
+    """Return indices of dimensions where chunk size is not 1."""
+    return [i for i, s in enumerate(chunks) if s != 1]
+
+
 def _construct_manifest_array(
     filepath: str,
     dataset: H5Dataset,
@@ -57,6 +62,9 @@ def _construct_manifest_array(
     ManifestArray
     """
     chunks = dataset.chunks or dataset.shape
+    keep_indices = _squeeze_indices(chunks)
+    keep_chunks = tuple(chunks[i] for i in keep_indices)
+    keep_shape = tuple(dataset.shape[i] for i in keep_indices)
     codecs = codecs_from_dataset(dataset)
     attrs = _extract_attrs(dataset)
     dtype = dataset.dtype
@@ -79,13 +87,14 @@ def _construct_manifest_array(
 
     fill_value = dataset.fillvalue.item()
     dims = tuple(_dataset_dims(dataset, group=group))
+    keep_dims = tuple(dims[i] for i in keep_indices)
     metadata = create_v3_array_metadata(
-        shape=dataset.shape,
+        shape=keep_shape,
         data_type=dtype,
-        chunk_shape=chunks,
+        chunk_shape=keep_chunks,
         fill_value=fill_value,
         codecs=codec_configs,
-        dimension_names=dims,
+        dimension_names=keep_dims,
         attributes=attrs,
     )
     manifest = _dataset_chunk_manifest(filepath, dataset)
@@ -218,6 +227,7 @@ def _dataset_chunk_manifest(
         A Virtualizarr ChunkManifest
     """
     dsid = dataset.id
+    keep_indices = _squeeze_indices(dataset.chunks or dataset.shape)
     if dataset.chunks is None:
         if dsid.get_offset() is None:
             chunk_manifest = ChunkManifest(entries={}, shape=dataset.shape)
@@ -228,7 +238,7 @@ def _dataset_chunk_manifest(
                 lengths=np.array(dsid.get_storage_size(), dtype=np.uint64),
             )
         else:
-            key_list = [0] * (len(dataset.shape) or 1)
+            key_list = [0] * (len(keep_indices) or 1)
             key = ".".join(map(str, key_list))
 
             chunk_entry: ChunkEntry = ChunkEntry.with_validation(  # type: ignore[attr-defined]
@@ -265,6 +275,14 @@ def add_chunk_info(blob):
                 for index in range(num_chunks):
                     add_chunk_info(dsid.get_chunk_info(index))
 
+            squeeze_axes = tuple(
+                i for i in range(len(dataset.chunks)) if i not in set(keep_indices)
+            )
+            if squeeze_axes:
+                paths = np.squeeze(paths, axis=squeeze_axes)
+                offsets = np.squeeze(offsets, axis=squeeze_axes)
+                lengths = np.squeeze(lengths, axis=squeeze_axes)
+
             chunk_manifest = ChunkManifest.from_arrays(
                 paths=paths,  # type: ignore
                 offsets=offsets,

From 553ab4546caa091ceb6091cb17216e0f3f29b1bf Mon Sep 17 00:00:00 2001
From: Aimee Barciauskas <aimee@developmentseed.org>
Date: Fri, 17 Apr 2026 11:37:46 -0700
Subject: [PATCH 4/8] Add squeeze boolean parameter to HDFParser.__init__

---
 virtualizarr/parsers/hdf/hdf.py               | 24 +++++++++++++++----
 .../tests/test_parsers/test_hdf/test_hdf.py   | 12 +++++++++-
 virtualizarr/tests/utils.py                   |  4 ++--
 3 files changed, 33 insertions(+), 7 deletions(-)

diff --git a/virtualizarr/parsers/hdf/hdf.py b/virtualizarr/parsers/hdf/hdf.py
index 1b371603..80b8bd69 100644
--- a/virtualizarr/parsers/hdf/hdf.py
+++ b/virtualizarr/parsers/hdf/hdf.py
@@ -44,6 +44,7 @@ def _construct_manifest_array(
     filepath: str,
     dataset: H5Dataset,
     group: str,
+    squeeze: bool = False,
 ) -> ManifestArray:
     """
     Construct a ManifestArray from an h5py dataset
@@ -62,7 +63,7 @@ def _construct_manifest_array(
     ManifestArray
     """
     chunks = dataset.chunks or dataset.shape
-    keep_indices = _squeeze_indices(chunks)
+    keep_indices = _squeeze_indices(chunks) if squeeze else list(range(len(chunks)))
     keep_chunks = tuple(chunks[i] for i in keep_indices)
     keep_shape = tuple(dataset.shape[i] for i in keep_indices)
     codecs = codecs_from_dataset(dataset)
@@ -97,7 +98,7 @@ def _construct_manifest_array(
         dimension_names=keep_dims,
         attributes=attrs,
     )
-    manifest = _dataset_chunk_manifest(filepath, dataset)
+    manifest = _dataset_chunk_manifest(filepath, dataset, squeeze=squeeze)
     return ManifestArray(metadata=metadata, chunkmanifest=manifest)
 
 
@@ -107,6 +108,7 @@ def _construct_manifest_group(
     *,
     group: str | None = None,
     drop_variables: Iterable[str] | None = None,
+    squeeze: bool = False,
 ) -> ManifestGroup:
     """
     Construct a virtual Group from a HDF dataset.
@@ -126,7 +128,9 @@ def _construct_manifest_group(
         drop_variables = set(drop_variables or ()) | set(non_coordinate_dimension_vars)
         group_name = str(g.name)  # NOTE: this will always include leading "/"
         arrays = {
-            key: _construct_manifest_array(filepath, dataset, group_name)
+            key: _construct_manifest_array(
+                filepath, dataset, group_name, squeeze=squeeze
+            )
             for key in g.keys()
             if key not in drop_variables
             if isinstance(dataset := g[key], h5py.Dataset)
@@ -136,6 +140,7 @@ def _construct_manifest_group(
                 filepath,
                 reader,
                 group=str(Path(group) / key) if group is not None else key,
+                squeeze=squeeze,
             )
             for key in g.keys()
             if key not in drop_variables
@@ -152,6 +157,7 @@ def __init__(
         group: str | None = None,
         drop_variables: Iterable[str] | None = None,
         reader_factory: ReaderFactory = BlockStoreReader,
+        squeeze: bool = False,
     ):
         """
         Instantiate a parser that can be used to virtualize HDF5/NetCDF4 files using the
@@ -169,10 +175,13 @@ def __init__(
             Must return an object implementing the
             [ReadableFile][obspec_utils.protocols.ReadableFile] protocol.
             Default is [BlockStoreReader][obspec_utils.readers.BlockStoreReader].
+        squeeze
+            If `True`, remove dimensions of size 1 from arrays (default: `False`).
         """
         self.group = group
         self.drop_variables = drop_variables
         self.reader_factory = reader_factory
+        self.squeeze = squeeze
 
     def __call__(
         self,
@@ -202,6 +211,7 @@ def __call__(
             reader=reader,
             group=self.group,
             drop_variables=self.drop_variables,
+            squeeze=self.squeeze,
         )
         # Convert to a manifest store
         return ManifestStore(registry=registry, group=manifest_group)
@@ -210,6 +220,7 @@ def __call__(
 def _dataset_chunk_manifest(
     filepath: str,
     dataset: H5Dataset,
+    squeeze: bool = False,
 ) -> ChunkManifest:
     """
     Generate ChunkManifest for HDF5 dataset.
@@ -227,7 +238,12 @@ def _dataset_chunk_manifest(
         A Virtualizarr ChunkManifest
     """
     dsid = dataset.id
-    keep_indices = _squeeze_indices(dataset.chunks or dataset.shape)
+    chunks_or_shape = dataset.chunks or dataset.shape
+    keep_indices = (
+        _squeeze_indices(chunks_or_shape)
+        if squeeze
+        else list(range(len(chunks_or_shape)))
+    )
     if dataset.chunks is None:
         if dsid.get_offset() is None:
             chunk_manifest = ChunkManifest(entries={}, shape=dataset.shape)
diff --git a/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py b/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py
index 2543c8bf..ea289715 100644
--- a/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py
+++ b/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py
@@ -43,12 +43,22 @@ def test_chunked_roundtrip(self, chunked_roundtrip_hdf5_url):
 
     def test_singleton_dimensions_squeezed(self, singleton_padded_dimension_hdf5_file):
         manifest_store = manifest_store_from_hdf_url(
-            singleton_padded_dimension_hdf5_file
+            singleton_padded_dimension_hdf5_file, squeeze=True
         )
         assert manifest_store._group.arrays["data"].shape == (10, 5)
         assert manifest_store._group.arrays["row_coord"].shape == (10,)
         assert manifest_store._group.arrays["col_coord"].shape == (5,)
 
+    def test_singleton_dimensions_not_squeezed(
+        self, singleton_padded_dimension_hdf5_file
+    ):
+        manifest_store = manifest_store_from_hdf_url(
+            singleton_padded_dimension_hdf5_file,
+        )
+        assert manifest_store._group.arrays["data"].shape == (10, 5)
+        assert manifest_store._group.arrays["row_coord"].shape == (10, 1)
+        assert manifest_store._group.arrays["col_coord"].shape == (1, 5)
+
 
 @requires_hdf5plugin
 @requires_imagecodecs
diff --git a/virtualizarr/tests/utils.py b/virtualizarr/tests/utils.py
index 4ce2fece..7d6c1814 100644
--- a/virtualizarr/tests/utils.py
+++ b/virtualizarr/tests/utils.py
@@ -40,8 +40,8 @@ def obstore_http(url: str) -> ObjectStore:
     return store
 
 
-def manifest_store_from_hdf_url(url, group: str | None = None):
+def manifest_store_from_hdf_url(url, group: str | None = None, squeeze: bool = False):
     registry: ObjectStoreRegistry = ObjectStoreRegistry()
     registry.register(url, obstore_local(url=url))
-    parser = HDFParser(group=group)
+    parser = HDFParser(group=group, squeeze=squeeze)
     return parser(url=url, registry=registry)

From 01a000113b9a4c8001030d2886c5d74eacf191ef Mon Sep 17 00:00:00 2001
From: Aimee Barciauskas <aimee@developmentseed.org>
Date: Sun, 19 Apr 2026 10:01:40 -0700
Subject: [PATCH 5/8] Add test case for when chunks include a singleton axis

---
 virtualizarr/parsers/hdf/hdf.py               |  2 +-
 virtualizarr/tests/test_parsers/conftest.py   | 41 +++++++++++++++----
 .../tests/test_parsers/test_hdf/test_hdf.py   | 30 +++++++++-----
 3 files changed, 54 insertions(+), 19 deletions(-)

diff --git a/virtualizarr/parsers/hdf/hdf.py b/virtualizarr/parsers/hdf/hdf.py
index 80b8bd69..4c5f939c 100644
--- a/virtualizarr/parsers/hdf/hdf.py
+++ b/virtualizarr/parsers/hdf/hdf.py
@@ -290,7 +290,7 @@ def add_chunk_info(blob):
             else:
                 for index in range(num_chunks):
                     add_chunk_info(dsid.get_chunk_info(index))
-
+            # we squeeze here rather than in get_key
             squeeze_axes = tuple(
                 i for i in range(len(dataset.chunks)) if i not in set(keep_indices)
             )
diff --git a/virtualizarr/tests/test_parsers/conftest.py b/virtualizarr/tests/test_parsers/conftest.py
index ee5d1490..d7082235 100644
--- a/virtualizarr/tests/test_parsers/conftest.py
+++ b/virtualizarr/tests/test_parsers/conftest.py
@@ -479,19 +479,46 @@ def big_endian_dtype_hdf5_file(tmpdir):
     return filepath
 
 
-@pytest.fixture
-def singleton_padded_dimension_hdf5_file(tmp_path: Path) -> str:
+@pytest.fixture(
+    params=[
+        # {"N": 50, "M": 100, "chunked": True, "chunks": (5, 25)},
+        # {"N": 50, "M": 100, "chunked": False, "chunks": None},
+        {"N": 1, "M": 100, "chunked": True, "chunks": (1, 25)},
+    ],
+    # ids=["chunked", "not_chunked", "singleton_chunked"],
+    ids=["singleton_chunked"],
+)
+def singleton_padded_dimension_hdf5_file(tmp_path: Path, request) -> tuple:
     """HDF5 file mimicking MATLAB layout: a 2D array (N, M) plus coordinate
     arrays shaped (N, 1) and (1, M)."""
-    N, M = 10, 5
+    N = request.param["N"]
+    M = request.param["M"]
+    chunked = request.param["chunked"]
+    chunks = request.param["chunks"]
     filepath = str(tmp_path / "singleton_dimension_layout.nc")
 
+    dataset_args = {
+        "data": {
+            "name": "data",
+            "data": np.random.random((N, M)),
+        },
+        "row_coord": {
+            "name": "row_coord",
+            "data": np.random.random((N, 1)),
+        },
+        "col_coord": {
+            "name": "col_coord",
+            "data": np.random.random((1, M)),
+        },
+    }
+    if chunks is not None:
+        dataset_args["data"]["chunks"] = chunks
+
     with h5py.File(filepath, "w") as f:
-        f.create_dataset(name="data", data=np.random.random((N, M)))
-        f.create_dataset(name="row_coord", data=np.random.random((N, 1)))
-        f.create_dataset(name="col_coord", data=np.random.random((1, M)))
+        for v in dataset_args.values():
+            f.create_dataset(**v)
 
-    return f"file://{filepath}"
+    return f"file://{filepath}", N, M, chunked, chunks
 
 
 @pytest.fixture()
diff --git a/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py b/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py
index ea289715..7bd270fe 100644
--- a/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py
+++ b/virtualizarr/tests/test_parsers/test_hdf/test_hdf.py
@@ -42,22 +42,30 @@ def test_chunked_roundtrip(self, chunked_roundtrip_hdf5_url):
         assert manifest_store._group.arrays["var2"].manifest.shape_chunk_grid == (2, 8)
 
     def test_singleton_dimensions_squeezed(self, singleton_padded_dimension_hdf5_file):
-        manifest_store = manifest_store_from_hdf_url(
-            singleton_padded_dimension_hdf5_file, squeeze=True
+        url, N, M, chunked, chunks = singleton_padded_dimension_hdf5_file
+        manifest_store = manifest_store_from_hdf_url(url, squeeze=True)
+        expected_data_shape = (M,) if N == 1 else (N, M)
+        assert manifest_store._group.arrays["data"].shape == expected_data_shape
+        if chunked:
+            expected_chunks = tuple(c for c, n in zip(chunks, (N, M)) if n != 1)
+        else:
+            expected_chunks = expected_data_shape
+        assert manifest_store._group.arrays["data"].chunks == expected_chunks
+        assert manifest_store._group.arrays["row_coord"].shape == (
+            () if N == 1 else (N,)
         )
-        assert manifest_store._group.arrays["data"].shape == (10, 5)
-        assert manifest_store._group.arrays["row_coord"].shape == (10,)
-        assert manifest_store._group.arrays["col_coord"].shape == (5,)
+        assert manifest_store._group.arrays["col_coord"].shape == (M,)
 
     def test_singleton_dimensions_not_squeezed(
         self, singleton_padded_dimension_hdf5_file
     ):
-        manifest_store = manifest_store_from_hdf_url(
-            singleton_padded_dimension_hdf5_file,
-        )
-        assert manifest_store._group.arrays["data"].shape == (10, 5)
-        assert manifest_store._group.arrays["row_coord"].shape == (10, 1)
-        assert manifest_store._group.arrays["col_coord"].shape == (1, 5)
+        url, N, M, chunked, chunks = singleton_padded_dimension_hdf5_file
+        manifest_store = manifest_store_from_hdf_url(url)
+        assert manifest_store._group.arrays["data"].shape == (N, M)
+        expected_chunks = chunks if chunked else (N, M)
+        assert manifest_store._group.arrays["data"].chunks == expected_chunks
+        assert manifest_store._group.arrays["row_coord"].shape == (N, 1)
+        assert manifest_store._group.arrays["col_coord"].shape == (1, M)
 
 
 @requires_hdf5plugin

From a7589205ade1628c7e59e64261ab4c02e1c4644d Mon Sep 17 00:00:00 2001
From: Aimee Barciauskas <aimee@developmentseed.org>
Date: Sun, 19 Apr 2026 16:04:13 -0700
Subject: [PATCH 6/8] Add integration test

---
 virtualizarr/parsers/hdf/hdf.py               |  1 +
 virtualizarr/tests/test_parsers/conftest.py   |  7 +++--
 .../test_hdf/test_hdf_integration.py          | 26 +++++++++++++++++++
 3 files changed, 30 insertions(+), 4 deletions(-)

diff --git a/virtualizarr/parsers/hdf/hdf.py b/virtualizarr/parsers/hdf/hdf.py
index 4c5f939c..61bd4fff 100644
--- a/virtualizarr/parsers/hdf/hdf.py
+++ b/virtualizarr/parsers/hdf/hdf.py
@@ -290,6 +290,7 @@ def add_chunk_info(blob):
             else:
                 for index in range(num_chunks):
                     add_chunk_info(dsid.get_chunk_info(index))
+
             # we squeeze here rather than in get_key
             squeeze_axes = tuple(
                 i for i in range(len(dataset.chunks)) if i not in set(keep_indices)
diff --git a/virtualizarr/tests/test_parsers/conftest.py b/virtualizarr/tests/test_parsers/conftest.py
index d7082235..78fb1aff 100644
--- a/virtualizarr/tests/test_parsers/conftest.py
+++ b/virtualizarr/tests/test_parsers/conftest.py
@@ -481,12 +481,11 @@ def big_endian_dtype_hdf5_file(tmpdir):
 
 @pytest.fixture(
     params=[
-        # {"N": 50, "M": 100, "chunked": True, "chunks": (5, 25)},
-        # {"N": 50, "M": 100, "chunked": False, "chunks": None},
+        {"N": 50, "M": 100, "chunked": True, "chunks": (5, 25)},
+        {"N": 50, "M": 100, "chunked": False, "chunks": None},
         {"N": 1, "M": 100, "chunked": True, "chunks": (1, 25)},
     ],
-    # ids=["chunked", "not_chunked", "singleton_chunked"],
-    ids=["singleton_chunked"],
+    ids=["chunked", "not_chunked", "singleton_chunked"],
 )
 def singleton_padded_dimension_hdf5_file(tmp_path: Path, request) -> tuple:
     """HDF5 file mimicking MATLAB layout: a 2D array (N, M) plus coordinate
diff --git a/virtualizarr/tests/test_parsers/test_hdf/test_hdf_integration.py b/virtualizarr/tests/test_parsers/test_hdf/test_hdf_integration.py
index e779317c..b195549b 100644
--- a/virtualizarr/tests/test_parsers/test_hdf/test_hdf_integration.py
+++ b/virtualizarr/tests/test_parsers/test_hdf/test_hdf_integration.py
@@ -96,6 +96,32 @@ def test_non_coord_dim_roundtrip(self, tmp_path, non_coord_dim, local_registry):
             with xr.open_dataset(kerchunk_file, engine="kerchunk") as roundtrip:
                 xrt.assert_equal(ds, roundtrip)
 
+    def test_singleton_dim_roundtrip(
+        self, tmp_path, singleton_padded_dimension_hdf5_file, local_registry
+    ):
+        import numpy as np
+
+        parser = HDFParser(squeeze=True)
+        filepath, *_ = singleton_padded_dimension_hdf5_file
+        with (
+            xr.open_dataset(
+                filepath, engine="h5netcdf", backend_kwargs={"phony_dims": "sort"}
+            ).squeeze(drop=True) as ds,
+            open_virtual_dataset(
+                url=filepath,
+                registry=local_registry,
+                parser=parser,
+            ) as vds,
+        ):
+            kerchunk_file = str(tmp_path / "kerchunk.json")
+            vds.vz.to_kerchunk(kerchunk_file, format="json")
+            with xr.open_dataset(kerchunk_file, engine="kerchunk") as roundtrip:
+                for var_name in ds.data_vars:
+                    assert ds[var_name].shape == roundtrip[var_name].shape
+                    np.testing.assert_array_equal(
+                        ds[var_name].values, roundtrip[var_name].values
+                    )
+
     @requires_icechunk
     def test_cf_fill_value_roundtrip(
         self, tmp_path, cf_fill_value_hdf5_file, local_registry

From 51dd9111d46dd2b6811d195a8e1b3214bc80f76e Mon Sep 17 00:00:00 2001
From: Aimee Barciauskas <aimee@developmentseed.org>
Date: Fri, 24 Apr 2026 14:11:23 -0700
Subject: [PATCH 7/8] Modify comment

---
 virtualizarr/parsers/hdf/hdf.py | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/virtualizarr/parsers/hdf/hdf.py b/virtualizarr/parsers/hdf/hdf.py
index a0e6ecb8..3e75c72d 100644
--- a/virtualizarr/parsers/hdf/hdf.py
+++ b/virtualizarr/parsers/hdf/hdf.py
@@ -68,10 +68,8 @@ def _construct_manifest_array(
     # this uses dataset.chunks if present or dataset.shape if chunks are not present
     # if using dataset.shape, it enforces that each dimension is at least size one
     chunks = dataset.chunks or tuple(max(s, 1) for s in dataset.shape)
-    # this says we only want to keep indices of dimensions where the size is greater than 1,
-    # which only somewhat overrides the logic above since we will remove any dimensions that are of size 1
-    # taken together it means we remove any dimension that is of size 1 or less
-    # and any chunk dimension that is of size 1 or less
+    # keep only dimensions where chunk size != 1; after clamping above, no dim can be < 1,
+    # so this effectively drops any dim that was zero-length in the original dataset.
     keep_indices = _squeeze_indices(chunks) if squeeze else list(range(len(chunks)))
     keep_chunks = tuple(chunks[i] for i in keep_indices)
     keep_shape = tuple(dataset.shape[i] for i in keep_indices)

From 6dc9337d38ea4456740c951ce493eec1bddd536e Mon Sep 17 00:00:00 2001
From: Aimee Barciauskas <aimee@developmentseed.org>
Date: Fri, 24 Apr 2026 14:40:00 -0700
Subject: [PATCH 8/8] Modify comments

---
 virtualizarr/parsers/hdf/hdf.py | 22 ++++++++++------------
 1 file changed, 10 insertions(+), 12 deletions(-)

diff --git a/virtualizarr/parsers/hdf/hdf.py b/virtualizarr/parsers/hdf/hdf.py
index 3e75c72d..874bd153 100644
--- a/virtualizarr/parsers/hdf/hdf.py
+++ b/virtualizarr/parsers/hdf/hdf.py
@@ -35,9 +35,9 @@
     from h5py import Group as H5Group
 
 
-def _squeeze_indices(chunks: tuple) -> list[int]:
-    """Return indices of dimensions where chunk size is not 1."""
-    return [i for i, s in enumerate(chunks) if s != 1]
+def _squeeze_indices(shape: tuple) -> list[int]:
+    """Return indices of dimensions where shape is greater than 1."""
+    return [i for i, s in enumerate(shape) if s > 1]
 
 
 def _construct_manifest_array(
@@ -65,12 +65,13 @@ def _construct_manifest_array(
     # Clamp each dim to >= 1: zarr v3 allows shape=(0,) but forbids zero-length
     # chunk dimensions (enforced by zarr-python >= 3.2.0). See
     # https://github.com/zarr-developers/zarr-python/issues/3711.
-    # this uses dataset.chunks if present or dataset.shape if chunks are not present
-    # if using dataset.shape, it enforces that each dimension is at least size one
     chunks = dataset.chunks or tuple(max(s, 1) for s in dataset.shape)
-    # keep only dimensions where chunk size != 1; after clamping above, no dim can be < 1,
-    # so this effectively drops any dim that was zero-length in the original dataset.
-    keep_indices = _squeeze_indices(chunks) if squeeze else list(range(len(chunks)))
+    # When squeeze=True, we keep only dimensions of size > 1.
+    # So squeeze=True on its own drops any dim that was length 0 (or less) in the original dataset and the clamp is technically redundant.
+    # But when squeeze=False, the clamp is necessary to prevent having zero-length chunk dimensions.
+    keep_indices = (
+        _squeeze_indices(dataset.shape) if squeeze else list(range(len(dataset.shape)))
+    )
     keep_chunks = tuple(chunks[i] for i in keep_indices)
     keep_shape = tuple(dataset.shape[i] for i in keep_indices)
 
@@ -246,11 +247,8 @@ def _dataset_chunk_manifest(
         A Virtualizarr ChunkManifest
     """
     dsid = dataset.id
-    chunks_or_shape = dataset.chunks or dataset.shape
     keep_indices = (
-        _squeeze_indices(chunks_or_shape)
-        if squeeze
-        else list(range(len(chunks_or_shape)))
+        _squeeze_indices(dataset.shape) if squeeze else list(range(len(dataset.shape)))
     )
     if dataset.chunks is None:
         if dsid.get_offset() is None: