anndata 0.12.7__tar.gz → 0.12.9__tar.gz

{anndata-0.12.7 → anndata-0.12.9}/.github/workflows/test-cpu.yml

@@ -44,6 +44,8 @@ jobs:
       matrix:
         env: ${{ fromJSON(needs.get-environments.outputs.envs) }}
         io_mark: ["zarr_io", "not zarr_io", "dask_distributed"] # dask_distributed should not be run with -n auto as it uses a client with processes
+    permissions:
+      id-token: write  # for codecov OIDC
     env:  # environment variables for use in codecov’s env_vars tagging
       ENV_NAME: ${{ matrix.env.name }}
       IO_MARK: ${{ matrix.io_mark }}
@@ -72,23 +74,25 @@ jobs:
           hatch run ${{ matrix.env.name }}:run-cov -v --color=yes ${{ matrix.io_mark != 'dask_distributed' && '-n auto' || '' }} --junitxml=test-data/test-results.xml -m "${{ matrix.io_mark }}" ${{ matrix.env.args }}
           hatch run ${{ matrix.env.name }}:cov-combine
           hatch run ${{ matrix.env.name }}:coverage xml
+          hatch run ${{ matrix.env.name }}:cov-report
 
       - name: Upload test results
         if: ${{ !cancelled() }}
-        uses: codecov/test-results-action@v1
+        uses: codecov/codecov-action@v5
         with:
-          token: ${{ secrets.CODECOV_TOKEN }}
+          report_type: test_results
           env_vars: ENV_NAME,IO_MARK
+          files: test-data/test-results.xml
+          use_oidc: true
           fail_ci_if_error: true
-          file: test-data/test-results.xml
 
       - name: Upload coverage data
         uses: codecov/codecov-action@v5
         with:
-          token: ${{ secrets.CODECOV_TOKEN }}
           env_vars: ENV_NAME,IO_MARK
-          fail_ci_if_error: true
           files: test-data/coverage.xml
+          use_oidc: true
+          fail_ci_if_error: true
 
   build:
     runs-on: ubuntu-24.04

{anndata-0.12.7 → anndata-0.12.9}/.github/workflows/test-gpu.yml

@@ -52,9 +52,10 @@ jobs:
       - name: Nvidia SMI sanity check
         run: nvidia-smi
 
-      - name: Install yq
+      - name: Install yq # https://cirun.slack.com/archives/C09SNDRB3A8/p1766512487317849?thread_ts=1766512112.938459&cid=C09SNDRB3A8
         run: |
-          sudo snap install yq
+          sudo wget -qO /usr/local/bin/yq https://github.com/mikefarah/yq/releases/latest/download/yq_linux_amd64
+          sudo chmod +x /usr/local/bin/yq
 
       - name: Extract max Python version from classifiers
         run: |

{anndata-0.12.7 → anndata-0.12.9}/.readthedocs.yml

@@ -2,7 +2,7 @@ version: 2
 build:
   os: ubuntu-24.04
   tools:
-    python: "3.13"
+    python: "3.14"
   jobs:
     post_checkout:
       # unshallow so version can be derived from tag

{anndata-0.12.7 → anndata-0.12.9}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: anndata
-Version: 0.12.7
+Version: 0.12.9
 Summary: Annotated data.
 Project-URL: Documentation, https://anndata.readthedocs.io/
 Project-URL: Source, https://github.com/scverse/anndata
@@ -32,7 +32,7 @@ Requires-Dist: natsort
 Requires-Dist: numpy>=1.26
 Requires-Dist: packaging>=24.2
 Requires-Dist: pandas!=2.1.2,<3,>=2.1.0
-Requires-Dist: scipy>=1.12
+Requires-Dist: scipy!=1.17.0,>=1.12
 Requires-Dist: zarr!=3.0.*,>=2.18.7
 Provides-Extra: cu11
 Requires-Dist: cupy-cuda11x; extra == 'cu11'
@@ -74,11 +74,11 @@ Requires-Dist: boltons; extra == 'test'
 Requires-Dist: dask[array]!=2024.8.*,!=2024.9.*,!=2025.2.*,!=2025.3.*,!=2025.4.*,!=2025.5.*,!=2025.6.*,!=2025.7.*,!=2025.8.*,>=2023.5.1; extra == 'test'
 Requires-Dist: dask[distributed]; extra == 'test'
 Requires-Dist: filelock; extra == 'test'
-Requires-Dist: httpx<1.0; extra == 'test'
 Requires-Dist: joblib; extra == 'test'
 Requires-Dist: loompy>=3.0.5; extra == 'test'
 Requires-Dist: matplotlib; extra == 'test'
 Requires-Dist: openpyxl; extra == 'test'
+Requires-Dist: pooch; extra == 'test'
 Requires-Dist: pyarrow; extra == 'test'
 Requires-Dist: pytest; extra == 'test'
 Requires-Dist: pytest-cov; extra == 'test'
@@ -96,11 +96,11 @@ Requires-Dist: boltons; extra == 'test-min'
 Requires-Dist: dask[array]!=2024.8.*,!=2024.9.*,!=2025.2.*,!=2025.3.*,!=2025.4.*,!=2025.5.*,!=2025.6.*,!=2025.7.*,!=2025.8.*,>=2023.5.1; extra == 'test-min'
 Requires-Dist: dask[distributed]; extra == 'test-min'
 Requires-Dist: filelock; extra == 'test-min'
-Requires-Dist: httpx<1.0; extra == 'test-min'
 Requires-Dist: joblib; extra == 'test-min'
 Requires-Dist: loompy>=3.0.5; extra == 'test-min'
 Requires-Dist: matplotlib; extra == 'test-min'
 Requires-Dist: openpyxl; extra == 'test-min'
+Requires-Dist: pooch; extra == 'test-min'
 Requires-Dist: pyarrow; extra == 'test-min'
 Requires-Dist: pytest; extra == 'test-min'
 Requires-Dist: pytest-cov; extra == 'test-min'

{anndata-0.12.7 → anndata-0.12.9}/benchmarks/benchmarks/dataset2d.py

@@ -18,11 +18,11 @@ class Dataset2D:
     params = (
         ("zarr", "h5ad"),
         ((-1,), None),
-        ("cat", "numeric", "string-array", "nullable-string-array"),
+        ("cat", "numeric", "string-array", "nullable-string-array", "all"),
     )
 
     def setup_cache(self):
-        n_obs = 10000
+        n_obs = 100_000
         array_types = {
             "numeric": np.arange(n_obs),
             "string-array": np.array(["a"] * n_obs),
@@ -40,17 +40,30 @@ class Dataset2D:
                 if writing_string_array_on_disk := (
                     isinstance(v, np.ndarray) and df["a"].dtype == "string"
                 ):
-                    df["a"] = df["a"].to_numpy()
+                    with pd.option_context("future.infer_string", False):  # noqa: FBT003
+                        df["a"] = df["a"].to_numpy()
                 with ad.settings.override(allow_write_nullable_strings=True):
                     ad.io.write_elem(store, "df", df)
                 if writing_string_array_on_disk:
                     assert store["df"]["a"].attrs["encoding-type"] == "string-array"
+        for store in [
+            h5py.File("data_all.h5ad", mode="w"),
+            zarr.open("data_all.zarr", mode="w", zarr_version=2),
+        ]:
+            df = pd.DataFrame(array_types, index=[f"cell{i}" for i in range(n_obs)])
+            # write a string array by triggering:
+            # https://github.com/scverse/anndata/blob/71966500949adcac4e49d2233f06e9f11f438e19/src/anndata/_io/specs/methods.py#L557-L559
+            df["string-array"] = df["string-array"].to_numpy().astype(object)
+            with ad.settings.override(allow_write_nullable_strings=True):
+                ad.io.write_elem(store, "df", df)
 
     def setup(
         self,
         store_type: Literal["zarr", "h5ad"],
         chunks: None | tuple[int],
-        array_type: Literal["cat", "numeric", "string-array", "nullable-string-array"],
+        array_type: Literal[
+            "cat", "numeric", "string-array", "nullable-string-array", "all"
+        ],
     ):
         self.store = (
             h5py.File(f"data_{array_type}.h5ad", mode="r")

{anndata-0.12.7 → anndata-0.12.9}/docs/conf.py

@@ -85,6 +85,7 @@ napoleon_use_rtype = True  # having a separate entry generally helps readability
 napoleon_use_param = True
 napoleon_custom_sections = [("Params", "Parameters")]
 typehints_defaults = "braces"
+always_use_bars_union = True  # use `|`, not `Union` in types even when on Python ≤3.14
 todo_include_todos = False
 nitpicky = True  # Report broken links
 nitpick_ignore = [  # APIs without an intersphinx entry
@@ -133,10 +134,8 @@ intersphinx_mapping = dict(
     loompy=("https://linnarssonlab.org/loompy", None),
     numpy=("https://numpy.org/doc/stable", None),
     obstore=("https://developmentseed.org/obstore/latest/", None),
-    pandas=("https://pandas.pydata.org/pandas-docs/stable", None),
-    # TODO: switch to `/3` once docs are built with Python 3.14
-    # https://github.com/readthedocs/readthedocs.org/issues/12523
-    python=("https://docs.python.org/3.13", None),
+    pandas=("https://pandas.pydata.org/pandas-docs/version/2.3", None),
+    python=("https://docs.python.org/3", None),
     scipy=("https://docs.scipy.org/doc/scipy", None),
     sklearn=("https://scikit-learn.org/stable", None),
     xarray=("https://docs.xarray.dev/en/stable", None),
@@ -174,6 +173,7 @@ qualname_overrides = {
     "numpy.dtypes.StringDType": ("py:attr", "numpy.dtypes.StringDType"),
     "pandas.DataFrame.iloc": ("py:attr", "pandas.DataFrame.iloc"),
     "pandas.DataFrame.loc": ("py:attr", "pandas.DataFrame.loc"),
+    "pandas.core.dtypes.dtypes.BaseMaskedDtype": "pandas.api.extensions.ExtensionDtype",
     # should be fixed soon: https://github.com/tox-dev/sphinx-autodoc-typehints/pull/516
     "types.EllipsisType": ("py:data", "types.EllipsisType"),
     "pathlib._local.Path": "pathlib.Path",

anndata-0.12.9/docs/release-notes/0.12.8.md

@@ -0,0 +1,10 @@
+(v0.12.8)=
+### 0.12.8 {small}`2026-01-27`
+
+#### Bug fixes
+
+- Actually copy single zarr store input in {func}`anndata.experimental.concat_on_disk` {user}`ilan-gold` ({pr}`2267`)
+- Fix `compressor` kwarg handling when writing to zarr v3 {user}`ilan-gold` ({pr}`2270`)
+- Only open HDF5 file once in backed mode {user}`flying-sheep` ({pr}`2274`)
+- Fix {obj}`numpy.uint` support in {func}`anndata.experimental.read_lazy` and {func}`anndata.experimental.read_elem_lazy` {user}`flying-sheep` ({pr}`2287`)
+- `var` index is loaded into memory if it wasn't previously for {func}`anndata.concat` when {attr}`anndata.AnnData.var` is a {class}`anndata.experimental.backed.Dataset2D` {user}`ilan-gold` ({pr}`2299`)

anndata-0.12.9/docs/release-notes/0.12.9.md

@@ -0,0 +1,6 @@
+(v0.12.9)=
+### 0.12.9 {small}`2026-01-29`
+
+#### Performance
+
+- Add a `write_csr_csc_indices_with_min_possible_dtype` option to {attr}`anndata.settings` to enable downcasting of the `indices` of csr and csc matrices to a smaller dtype when writing. For example, if your csr matrix only has 30000 columns, then you can write out the `indices` of that matrix as `uint16` instead of `int64`. {user}`ilan-gold` ({pr}`2159`)

{anndata-0.12.7 → anndata-0.12.9}/hatch.toml

@@ -14,6 +14,7 @@ scripts.build = "python3 ci/scripts/towncrier_automation.py {args}"
 scripts.clean = "git restore --source=HEAD --staged --worktree -- docs/release-notes"
 
 [envs.hatch-test]
+python = "3.14"
 default-args = [  ]
 features = [ "dev", "test-min" ]
 extra-dependencies = [ "ipykernel" ]
@@ -36,10 +37,6 @@ overrides.matrix.deps.pre-install-commands = [
 ]
 overrides.matrix.deps.python = [
     { if = [ "min" ], value = "3.11" },
-    # transitive test dep numba doesn’t support 3.14 in a stable release yet:
-    # https://github.com/numba/numba/issues/9957
-    { if = [ "stable" ], value = "3.13" },
-    { if = [ "pre" ], value = "3.14" },
 ]
 overrides.matrix.deps.features = [
     { if = [ "stable", "pre" ], value = "test" },

{anndata-0.12.7 → anndata-0.12.9}/pyproject.toml

@@ -40,7 +40,7 @@ dependencies = [
     "pandas >=2.1.0, !=2.1.2, <3",
     "numpy>=1.26",
     # https://github.com/scverse/anndata/issues/1434
-    "scipy >=1.12",
+    "scipy >=1.12,!=1.17.0",
     "h5py>=3.8",
     "natsort",
     "packaging>=24.2",
@@ -93,11 +93,10 @@ test-min = [
     "joblib",
     "boltons",
     "scanpy>=1.10",
-    # TODO: Is 1.0dev1 a real pre-release? https://pypi.org/project/httpx/#history
-    "httpx<1.0",         # For data downloading
     "dask[distributed]",
     "awkward>=2.3.2",
     "pyarrow",
+    "pooch",
     "anndata[dask]",
 ]
 test = [ "anndata[test-min,lazy]" ]
@@ -159,12 +158,11 @@ filterwarnings_when_strict = [
     "default::scipy.sparse.SparseEfficiencyWarning",
     "default::dask.array.core.PerformanceWarning",
     "default:anndata will no longer support zarr v2:DeprecationWarning",
-    "default:The codec `vlen-utf8:UserWarning",
-    "default:The dtype `StringDType():UserWarning",
     "default:Consolidated metadata is:UserWarning",
     "default:.*Structured:zarr.core.dtype.common.UnstableSpecificationWarning",
     "default:.*FixedLengthUTF32:zarr.core.dtype.common.UnstableSpecificationWarning",
     "default:Automatic shard shape inference is experimental",
+    "default:Writing zarr v2:UserWarning",
 ]
 python_files = [ "test_*.py" ]
 testpaths = [

{anndata-0.12.7 → anndata-0.12.9}/src/anndata/_core/anndata.py

@@ -362,7 +362,12 @@ class AnnData(metaclass=utils.DeprecationMixinMeta):  # noqa: PLW1641
 
         # init from file
         if filename is not None:
-            self.file = AnnDataFileManager(self, filename, filemode)
+            fileobj, filename = (
+                (filename, None)
+                if isinstance(filename, h5py.File)
+                else (None, filename)
+            )
+            self.file = AnnDataFileManager(self, filename, filemode, fileobj)
         else:
             self.file = AnnDataFileManager(self, None)
 

{anndata-0.12.7 → anndata-0.12.9}/src/anndata/_core/index.py

@@ -25,12 +25,6 @@ def _normalize_indices(
     # deal with tuples of length 1
     if isinstance(index, tuple) and len(index) == 1:
         index = index[0]
-    # deal with pd.Series
-    if isinstance(index, pd.Series):
-        index = index.values
-    if isinstance(index, tuple):
-        # TODO: The series should probably be aligned first
-        index = tuple(i.values if isinstance(i, pd.Series) else i for i in index)
     ax0, ax1 = unpack_index(index)
     ax0 = _normalize_index(ax0, names0)
     ax1 = _normalize_index(ax1, names1)
@@ -45,6 +39,9 @@ def _normalize_index(  # noqa: PLR0911, PLR0912
         msg = f"Don’t call _normalize_index with non-categorical/string names and non-range index {index}"
         raise TypeError(msg)
 
+    if isinstance(indexer, pd.Index | pd.Series):
+        indexer = indexer.array
+
     # the following is insanely slow for sequences,
     # we replaced it using pandas below
     def name_idx(i):
@@ -65,16 +62,21 @@ def _normalize_index(  # noqa: PLR0911, PLR0912
     elif isinstance(indexer, str):
         return index.get_loc(indexer)  # int
     elif isinstance(
-        indexer, Sequence | np.ndarray | pd.Index | CSMatrix | np.matrix | CSArray
+        indexer,
+        Sequence
+        | np.ndarray
+        | pd.api.extensions.ExtensionArray
+        | CSMatrix
+        | np.matrix
+        | CSArray,
     ):
-        if hasattr(indexer, "shape") and (
-            (indexer.shape == (index.shape[0], 1))
-            or (indexer.shape == (1, index.shape[0]))
+        if (shape := getattr(indexer, "shape", None)) is not None and (
+            shape == (index.shape[0], 1) or shape == (1, index.shape[0])
         ):
             if isinstance(indexer, CSMatrix | CSArray):
                 indexer = indexer.toarray()
             indexer = np.ravel(indexer)
-        if not isinstance(indexer, np.ndarray | pd.Index):
+        if not isinstance(indexer, np.ndarray):
             indexer = np.array(indexer)
             if len(indexer) == 0:
                 indexer = indexer.astype(int)
@@ -111,7 +113,7 @@ def _normalize_index(  # noqa: PLR0911, PLR0912
             return indexer.data.compute()
         return indexer.data
     msg = f"Unknown indexer {indexer!r} of type {type(indexer)}"
-    raise IndexError()
+    raise IndexError(msg)
 
 
 def _fix_slice_bounds(s: slice, length: int) -> slice:

{anndata-0.12.7 → anndata-0.12.9}/src/anndata/_core/merge.py

@@ -10,7 +10,7 @@ from collections.abc import Callable, Mapping, MutableSet
 from functools import partial, reduce, singledispatch
 from itertools import repeat
 from operator import and_, or_, sub
-from typing import TYPE_CHECKING, Literal, TypeVar
+from typing import TYPE_CHECKING, Literal, TypeVar, cast
 from warnings import warn
 
 import numpy as np
@@ -44,7 +44,7 @@ if TYPE_CHECKING:
 
     from anndata._types import Join_T
 
-    from ..compat import XDataArray, XDataset
+    from ..compat import XDataArray
 
 T = TypeVar("T")
 
@@ -244,110 +244,89 @@ def as_cp_sparse(x) -> CupySparseMatrix:
 def unify_dtypes(
     dfs: Iterable[pd.DataFrame | Dataset2D],
 ) -> list[pd.DataFrame | Dataset2D]:
-    """
-    Attempts to unify datatypes from multiple dataframes.
+    """Attempt to unify datatypes from multiple dataframes.
 
     For catching cases where pandas would convert to object dtype.
     """
     dfs = list(dfs)
     # Get shared categorical columns
-    df_dtypes = [dict(df.dtypes) for df in dfs]
+    df_dtypes = [
+        cast("pd.Series[ExtensionDtype]", df.dtypes).to_dict()
+        if isinstance(df, pd.DataFrame)
+        else df.dtypes
+        for df in dfs
+    ]
     columns = reduce(lambda x, y: x.union(y), [df.columns for df in dfs])
-
-    dtypes: dict[str, list[np.dtype | ExtensionDtype]] = {col: [] for col in columns}
-    for col in columns:
-        for df in df_dtypes:
-            dtypes[col].append(df.get(col, None))
-
+    dtypes = {
+        col: (
+            [df[col] for df in df_dtypes if col in df],
+            any(col not in df for df in df_dtypes),
+        )
+        for col in columns
+    }
     if len(dtypes) == 0:
         return dfs
-    else:
-        dfs = [df.copy(deep=False) for df in dfs]
 
     new_dtypes = {
         col: target_dtype
-        for col, dtype in dtypes.items()
-        if (target_dtype := try_unifying_dtype(dtype)) is not None
+        for col, (dts, has_missing) in dtypes.items()
+        if (target_dtype := try_unifying_dtype(dts, has_missing=has_missing))
+        is not None
     }
 
+    dfs = [df.copy(deep=False) for df in dfs]
     for df in dfs:
         for col, dtype in new_dtypes.items():
             if col in df:
                 df[col] = df[col].astype(dtype)
-
     return dfs
 
 
-def try_unifying_dtype(  # noqa PLR0911, PLR0912
-    col: Sequence[np.dtype | ExtensionDtype],
-) -> pd.core.dtypes.base.ExtensionDtype | None:
-    """
-    If dtypes can be unified, returns the dtype they would be unified to.
+def try_unifying_dtype(
+    dtypes: Sequence[np.dtype | ExtensionDtype], *, has_missing: bool
+) -> ExtensionDtype | type[object] | None:
+    """Determine unified dtype if possible.
 
-    Returns None if they can't be unified, or if we can expect pandas to unify them for
-    us.
+    Returns None if they can’t be unified, or if we can expect pandas to unify them for us.
 
     Params
     ------
-    col:
-        A list of dtypes to unify. Can be numpy/ pandas dtypes, or None (which denotes
-        a missing value)
+    dtypes
+        A list of dtypes to unify. Can be numpy or pandas dtypes
+    has_missing
+        Whether the result needs to accommodate missing values
     """
-    dtypes: set[pd.CategoricalDtype] = set()
     # Categorical
-    if any(isinstance(dtype, pd.CategoricalDtype) for dtype in col):
-        ordered = False
-        for dtype in col:
-            if isinstance(dtype, pd.CategoricalDtype):
-                dtypes.add(dtype)
-                ordered = ordered | dtype.ordered
-            elif not pd.isnull(dtype):
-                return None
-        if len(dtypes) > 0:
-            categories = reduce(
-                lambda x, y: x.union(y),
-                (dtype.categories for dtype in dtypes if not pd.isnull(dtype)),
-            )
+    if any(isinstance(dtype, pd.CategoricalDtype) for dtype in dtypes):
+        if not all(isinstance(dtype, pd.CategoricalDtype) for dtype in dtypes):
+            return None
+        if TYPE_CHECKING:
+            dtypes = cast("Sequence[pd.CategoricalDtype]", dtypes)
+
+        all_categories = reduce(
+            lambda x, y: x.union(y), (dtype.categories for dtype in dtypes)
+        )
+        if not any(dtype.ordered for dtype in dtypes):
+            return pd.CategoricalDtype(natsorted(all_categories), ordered=False)
+
+        dtypes_with_categories = [
+            dtype for dtype in dtypes if len(dtype.categories) > 0
+        ]
+        if dtypes_with_categories and all(
+            len(dtype.categories) == len(all_categories)
+            and dtype.ordered
+            and np.all(all_categories == dtype.categories)
+            for dtype in dtypes_with_categories
+        ):
+            return dtypes_with_categories[0]
+
+        return object
 
-            if not ordered:
-                return pd.CategoricalDtype(natsorted(categories), ordered=False)
-            else:  # for xarray Datasets, see https://github.com/pydata/xarray/issues/10247
-                categories_intersection = reduce(
-                    lambda x, y: x.intersection(y),
-                    (
-                        dtype.categories
-                        for dtype in dtypes
-                        if not pd.isnull(dtype) and len(dtype.categories) > 0
-                    ),
-                )
-                if len(categories_intersection) < len(categories):
-                    return object
-                else:
-                    same_orders = all(
-                        dtype.ordered
-                        for dtype in dtypes
-                        if not pd.isnull(dtype) and len(dtype.categories) > 0
-                    )
-                    same_orders &= all(
-                        np.all(categories == dtype.categories)
-                        for dtype in dtypes
-                        if not pd.isnull(dtype) and len(dtype.categories) > 0
-                    )
-                    if same_orders:
-                        return next(
-                            dtype
-                            for dtype in dtypes
-                            if not pd.isnull(dtype) and len(dtype.categories) > 0
-                        )
-                    return object
     # Boolean
-    elif all(pd.api.types.is_bool_dtype(dtype) or dtype is None for dtype in col):
-        if any(dtype is None for dtype in col):
-            return pd.BooleanDtype()
-        else:
-            return None
-    else:
-        return None
+    if all(pd.api.types.is_bool_dtype(dtype) for dtype in dtypes) and has_missing:
+        return pd.BooleanDtype()
+
+    return None
 
 
 def check_combinable_cols(cols: list[pd.Index], join: Join_T):
@@ -1207,15 +1186,13 @@ def make_dask_col_from_extension_dtype(
     A :class:`dask.Array`: representation of the column.
     """
     import dask.array as da
-    import xarray as xr
-    from xarray.core.indexing import LazilyIndexedArray
 
     from anndata._io.specs.lazy_methods import (
         compute_chunk_layout_for_axis_size,
         get_chunksize,
         maybe_open_h5,
     )
-    from anndata.compat import XDataArray
+    from anndata.compat import xarray as xr
     from anndata.experimental import read_elem_lazy
 
     base_path_or_zarr_group = col.attrs.get("base_path_or_zarr_group")
@@ -1224,7 +1201,6 @@ def make_dask_col_from_extension_dtype(
         base_path_or_zarr_group is not None and elem_name is not None
     ):  # lazy, backed by store
         dims = col.dims
-        coords = col.coords.copy()
         with maybe_open_h5(base_path_or_zarr_group, elem_name) as f:
             maybe_chunk_size = get_chunksize(read_elem_lazy(f))
             chunk_size = (
@@ -1238,17 +1214,14 @@ def make_dask_col_from_extension_dtype(
             # reopening is important to get around h5py's unserializable lock in processes
             with maybe_open_h5(base_path_or_zarr_group, elem_name) as f:
                 v = read_elem_lazy(f)
-                variable = xr.Variable(data=LazilyIndexedArray(v), dims=dims)
-                data_array = XDataArray(
-                    variable,
-                    coords=coords,
-                    dims=dims,
+                variable = xr.Variable(
+                    data=xr.core.indexing.LazilyIndexedArray(v), dims=dims
                 )
                 idx = tuple(
                     slice(start, stop)
                     for start, stop in block_info[None]["array-location"]
                 )
-                chunk = np.array(data_array.data[idx])
+                chunk = np.array(variable.data[idx])
             return chunk
 
         if col.dtype == "category" or col.dtype == "string" or use_only_object_dtype:  # noqa PLR1714
@@ -1268,7 +1241,7 @@ def make_dask_col_from_extension_dtype(
 
 def make_xarray_extension_dtypes_dask(
     annotations: Iterable[Dataset2D], *, use_only_object_dtype: bool = False
-) -> Generator[XDataset, None, None]:
+) -> Generator[Dataset2D, None, None]:
     """
     Creates a generator of Dataset2D objects with dask arrays in place of :class:`pandas.api.extensions.ExtensionArray` dtype columns.
 
@@ -1710,6 +1683,9 @@ def concat(  # noqa: PLR0912, PLR0913, PLR0915
                 alt_annotations, use_only_object_dtype=True
             )
         )
+        for a in annotations_with_only_dask:
+            if a.true_index_dim != a.index_dim:
+                a.index = a.true_index
         annotations_with_only_dask = [
             a.ds.rename({a.true_index_dim: "merge_index"})
             for a in annotations_with_only_dask
@@ -1717,7 +1693,6 @@ def concat(  # noqa: PLR0912, PLR0913, PLR0915
         alt_annot = Dataset2D(
             xr.merge(annotations_with_only_dask, join=join, compat="override")
         )
-        alt_annot.true_index_dim = "merge_index"
 
     X = concat_Xs(adatas, reindexers, axis=axis, fill_value=fill_value)