fix: write iospec directly with sparse dataset

src/anndata/_io/specs/methods.py

@@ -780,10 +780,10 @@ def write_sparse_compressed(
     _REGISTRY.register_write(store_type, cls, spec)(func)
 
 
-@_REGISTRY.register_write(H5Group, _CSRDataset, IOSpec("", "0.1.0"))
-@_REGISTRY.register_write(H5Group, _CSCDataset, IOSpec("", "0.1.0"))
-@_REGISTRY.register_write(ZarrGroup, _CSRDataset, IOSpec("", "0.1.0"))
-@_REGISTRY.register_write(ZarrGroup, _CSCDataset, IOSpec("", "0.1.0"))
+@_REGISTRY.register_write(H5Group, _CSRDataset, IOSpec("csr_matrix", "0.1.0"))
+@_REGISTRY.register_write(H5Group, _CSCDataset, IOSpec("csc_matrix", "0.1.0"))
+@_REGISTRY.register_write(ZarrGroup, _CSRDataset, IOSpec("csr_matrix", "0.1.0"))
+@_REGISTRY.register_write(ZarrGroup, _CSCDataset, IOSpec("csc_matrix", "0.1.0"))
 def write_sparse_dataset(
     f: GroupStorageType,
     k: str,
@@ -800,9 +800,6 @@ def write_sparse_dataset(
         fmt=elem.format,
         dataset_kwargs=dataset_kwargs,
     )
-    # TODO: Cleaner way to do this
-    f[k].attrs["encoding-type"] = f"{elem.format}_matrix"
-    f[k].attrs["encoding-version"] = "0.1.0"
 
 
 def write_cupy_dask(f, k, elem, _writer, dataset_kwargs=MappingProxyType({})):

tests/test_backed_sparse.py

@@ -623,6 +623,22 @@ def test_anndata_sparse_compat(tmp_path: Path, diskfmt: Literal["h5ad", "zarr"])
     assert_equal(adata.X, base)
 
 
+def test_write(tmp_path: Path, diskfmt: Literal["h5ad", "zarr"]):
+    base = sparse.random(10, 10, format="csr")
+
+    f = (
+        open_write_group(tmp_path / f"parent_store.{diskfmt}", mode="a")
+        if diskfmt == "zarr"
+        else h5py.File(tmp_path / f"parent_store.{diskfmt}", "a")
+    )
+
+    ad.io.write_elem(f, "a_sparse_matrix", base)
+    adata = ad.AnnData(sparse_dataset(f["a_sparse_matrix"]))
+    ad.io.write_elem(f, "adata", adata)
+    adata_roundtripped = ad.io.read_elem(f["adata"])
+    assert_equal(adata_roundtripped.X, base)
+
+
 def test_backed_sizeof(
     ondisk_equivalent_adata: tuple[AnnData, AnnData, AnnData, AnnData],
 ):

tests/test_io_dispatched.py

@@ -5,6 +5,7 @@
 
 import h5py
 import pytest
+import scipy.sparse as sp
 import zarr
 
 import anndata as ad
@@ -16,6 +17,7 @@
 if TYPE_CHECKING:
     from collections.abc import Callable
     from pathlib import Path
+    from typing import Literal
 
 
 @pytest.mark.zarr_io
@@ -93,6 +95,24 @@ def test_read_dispatched_null_case(tmp_path: Path):
     assert_equal(expected, actual)
 
 
+@pytest.mark.zarr_io
+@pytest.mark.parametrize("sparse_format", ["csr", "csc"])
+def test_write_dispatched_csr_dataset(
+    tmp_path: Path, sparse_format: Literal["csr", "csc"]
+):
+    ad.io.write_elem(
+        open_write_group(tmp_path / "arr.zarr"),
+        "/",
+        sp.random(10, 10, format=sparse_format),
+    )
+    X = ad.io.sparse_dataset(zarr.open(tmp_path / "arr.zarr"))
+
+    def zarr_writer(func, store, elem_name: str, elem, iospec, dataset_kwargs):
+        assert iospec.encoding_type == f"{sparse_format}_matrix"
+
+    write_dispatched(zarr.open(tmp_path / "check.zarr", mode="w"), "/X", X, zarr_writer)
+
+
 @pytest.mark.zarr_io
 def test_write_dispatched_chunks(tmp_path: Path):
     from itertools import chain, repeat