Class Paths Have Strange Behavior on 8.3.4

[ilan-gold]

Hello! We are having an issue in our repo where 8.3.4 pytest (but not 8.3.3) caused our classes to have strange behavior at runtime with errors like:

        if (dest_type, src_type, modifiers) not in self.write:
>           raise IORegistryError._from_write_parts(dest_type, src_type, modifiers)
E           anndata._io.specs.registry.IORegistryError: No method registered for writing <class 'opt.hostedtoolcache.Python.3_12_7.x64.lib.python3_12.site-packages.anndata._core.views.AwkwardArrayView'> into <class 'h5py._hl.group.Group'>
E           Error raised while writing key 'awk_2d_ragged' of <class 'h5py._hl.group.Group'> to /obsm

when in reality, the AwkwardArrayView is in the self.write dictionary, but as
<class 'anndata._core.views.AwkwardArrayView'>:

(This screenshot is from my attempt to fix the issue where I printed out the contents of self.write)

Also weirdly I cannot reproduce this locally. I have tried creating a local reproducer with an src/package/__init__.py file for a dummy package like:

d = {}

class Foo():
    pass

d[Foo] = True

def check(typ: type) -> bool:
    return d[typ]

__all__ = ["Foo", "check"]

with a single pytest

from package import Foo, check

def test_check():
    f = Foo()
    assert check(type(f))

But this seems to work when I run pytest instead of throwing a KeyError. I will continue to try to dig but I wanted to report.

See also my various attempted to fix the issue

Python Env:

anndata 0.11.1.dev18+g6594d93 anyio 4.6.2.post1 array-api-compat 1.9.1 asciitree 0.3.3 scikit-learn 1.5.2 scipy 1.14.1 seaborn 0.13.2 session-info 1.0.0 setuptools 75.6.0 setuptools-scm 8.1.0 six 1.16.0 sniffio 1.3.1 sortedcontainers 2.4.0 statsmodels 0.14.4 stdlib-list 0.11.0 tblib 3.0.0 textual 0.88.1 threadpoolctl 3.5.0 toolz 1.0.0 tornado 6.4.2 towncrier 24.8.0 tqdm 4.67.1 typing-extensions 4.12.2 tzdata 2024.2 uc-micro-py 1.0.3 umap-learn 0.5.7 urllib3 2.2.3 uv 0.5.5 zarr 2.18.3 zict 3.0.0

Compute Env:

Starting: Initialize job Agent name: 'Azure Pipelines 5' Agent machine name: 'fv-az360-809' Current agent version: '4.248.0' Downloading task: PublishPipelineMetadata (0.248.1) Checking job knob settings. Knob: DockerActionRetries = true Source: $(VSTSAGENT_DOCKER_ACTION_RETRIES) Knob: AgentToolsDirectory = /opt/hostedtoolcache Source: ${AGENT_TOOLSDIRECTORY} Knob: UseGitLongPaths = true Source: $(USE_GIT_LONG_PATHS) Knob: AgentPerflog = /home/vsts/perflog Source: ${VSTS_AGENT_PERFLOG} Knob: EnableIssueSourceValidation = true Source: $(ENABLE_ISSUE_SOURCE_VALIDATION) Knob: AgentEnablePipelineArtifactLargeChunkSize = true Source: $(AGENT_ENABLE_PIPELINEARTIFACT_LARGE_CHUNK_SIZE) Knob: ContinueAfterCancelProcessTreeKillAttempt = true Source: $(VSTSAGENT_CONTINUE_AFTER_CANCEL_PROCESSTREEKILL_ATTEMPT) Knob: ProcessHandlerSecureArguments = false Source: $(AZP_75787_ENABLE_NEW_LOGIC) Knob: ProcessHandlerSecureArguments = false Source: $(AZP_75787_ENABLE_NEW_LOGIC_LOG) Knob: ProcessHandlerTelemetry = true Source: $(AZP_75787_ENABLE_COLLECT) Knob: UseNewNodeHandlerTelemetry = True Source: $(DistributedTask.Agent.USENEWNODEHANDLERTELEMETRY) Knob: ProcessHandlerEnableNewLogic = true Source: $(AZP_75787_ENABLE_NEW_PH_LOGIC) Knob: EnableResourceMonitorDebugOutput = true Source: $(AZP_ENABLE_RESOURCE_MONITOR_DEBUG_OUTPUT) Knob: EnableResourceUtilizationWarnings = true Source: $(AZP_ENABLE_RESOURCE_UTILIZATION_WARNINGS) Knob: IgnoreVSTSTaskLib = true Source: $(AZP_AGENT_IGNORE_VSTSTASKLIB) Knob: FailJobWhenAgentDies = true Source: $(FAIL_JOB_WHEN_AGENT_DIES) Knob: CheckForTaskDeprecation = true Source: $(AZP_AGENT_CHECK_FOR_TASK_DEPRECATION) Knob: CheckIfTaskNodeRunnerIsDeprecated246 = True Source: $(DistributedTask.Agent.CheckIfTaskNodeRunnerIsDeprecated246) Knob: UseNode20ToStartContainer = True Source: $(DistributedTask.Agent.UseNode20ToStartContainer) Knob: LogTaskNameInUserAgent = true Source: $(AZP_AGENT_LOG_TASKNAME_IN_USERAGENT) Knob: UseFetchFilterInCheckoutTask = true Source: $(AGENT_USE_FETCH_FILTER_IN_CHECKOUT_TASK) Knob: Rosetta2Warning = true Source: $(ROSETTA2_WARNING) Knob: AddForceCredentialsToGitCheckout = True Source: $(DistributedTask.Agent.AddForceCredentialsToGitCheckout) Finished checking job knob settings. Start tracking orphan processes. Finishing: Initialize job

• a detailed description of the bug or problem you are having
• output of pip list from the virtual environment you are using
• pytest and operating system versions
• minimal example if possible (attempted)

[dongfangtianyu]

Transferred Traceback over:

__ test_backed_raw_subset[sparray_bool_subset-sparray_bool_subset-csr_matrix] __

tmp_path = PosixPath('/tmp/pytest-of-vsts/pytest-0/test_backed_raw_subset_sparray28')
array_type = <class 'scipy.sparse._csr.csr_matrix'>
subset_func = <function sparray_bool_subset at 0x7febceb65c60>
subset_func2 = <function sparray_bool_subset at 0x7febceb65c60>

    @pytest.mark.parametrize(
        "array_type",
        [
            pytest.param(asarray, id="dense_array"),
            pytest.param(sparse.csr_matrix, id="csr_matrix"),
            pytest.param(sparse.csr_array, id="csr_array"),
        ],
    )
    def test_backed_raw_subset(tmp_path, array_type, subset_func, subset_func2):
        backed_pth = tmp_path / "backed.h5ad"
        final_pth = tmp_path / "final.h5ad"
        mem_adata = gen_adata((10, 10), X_type=array_type)
        mem_adata.raw = mem_adata
        obs_idx = subset_func(mem_adata.obs_names)
        var_idx = subset_func2(mem_adata.var_names)
       if (
            array_type is asarray
            and isinstance(obs_idx, list | np.ndarray | sparse.spmatrix | SpArray)
            and isinstance(var_idx, list | np.ndarray | sparse.spmatrix | SpArray)
        ):
            pytest.xfail(
                "Fancy indexing does not work with multiple arrays on a h5py.Dataset"
            )
        mem_adata.write(backed_pth)
    
        ### Backed view has same values as in memory view ###
        backed_adata = ad.read_h5ad(backed_pth, backed="r")
        backed_v = backed_adata[obs_idx, var_idx]
        assert backed_v.is_view
        mem_v = mem_adata[obs_idx, var_idx]
    
        # Value equivalent
        assert_equal(mem_v, backed_v)
        # Type and value equivalent
        assert_equal(mem_v.copy(), backed_v.to_memory(copy=True), exact=True)
        assert backed_v.is_view
        assert backed_v.isbacked
    
        ### Write from backed view ###
>       backed_v.write_h5ad(final_pth)

tests/test_backed_hdf5.py:225: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
/opt/hostedtoolcache/Python/3.12.7/x64/lib/python3.12/site-packages/anndata/_core/anndata.py:1865: in write_h5ad
    write_h5ad(
/opt/hostedtoolcache/Python/3.12.7/x64/lib/python3.12/site-packages/anndata/_io/h5ad.py:105: in write_h5ad
    write_elem(f, "obsm", dict(adata.obsm), dataset_kwargs=dataset_kwargs)
/opt/hostedtoolcache/Python/3.12.7/x64/lib/python3.12/site-packages/anndata/_io/specs/registry.py:488: in write_elem
    Writer(_REGISTRY).write_elem(store, k, elem, dataset_kwargs=dataset_kwargs)
/opt/hostedtoolcache/Python/3.12.7/x64/lib/python3.12/site-packages/anndata/_io/utils.py:248: in func_wrapper
    return func(*args, **kwargs)
/opt/hostedtoolcache/Python/3.12.7/x64/lib/python3.12/site-packages/anndata/_io/specs/registry.py:355: in write_elem
    return write_func(store, k, elem, dataset_kwargs=dataset_kwargs)
/opt/hostedtoolcache/Python/3.12.7/x64/lib/python3.12/site-packages/anndata/_io/specs/registry.py:71: in wrapper
    result = func(g, k, *args, **kwargs)
/opt/hostedtoolcache/Python/3.12.7/x64/lib/python3.12/site-packages/anndata/_io/specs/methods.py:353: in write_mapping
    _writer.write_elem(g, sub_k, sub_v, dataset_kwargs=dataset_kwargs)
/opt/hostedtoolcache/Python/3.12.7/x64/lib/python3.12/site-packages/anndata/_io/utils.py:248: in func_wrapper
    return func(*args, **kwargs)
/opt/hostedtoolcache/Python/3.12.7/x64/lib/python3.12/site-packages/anndata/_io/specs/registry.py:352: in write_elem
    write_func = self.find_write_func(dest_type, elem, modifiers)
/opt/hostedtoolcache/Python/3.12.7/x64/lib/python3.12/site-packages/anndata/_io/specs/registry.py:319: in find_write_func
    return self.registry.get_write(dest_type, type(elem), modifiers, writer=self)
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
self = <anndata._io.specs.registry.IORegistry object at 0x7febceaf3260>
dest_type = <class 'h5py._hl.group.Group'>
src_type = <class 'opt.hostedtoolcache.Python.3_12_7.x64.lib.python3_12.site-packages.anndata._core.views.AwkwardArrayView'>
modifiers = frozenset()

    def get_write(
        self,
        dest_type: type,
        src_type: type | tuple[type, str],
        modifiers: frozenset[str] = frozenset(),
        *,
        writer: Writer,
    ) -> Write:
        import h5py
    
        if dest_type is h5py.File:
            dest_type = h5py.Group
    
        if (dest_type, src_type, modifiers) not in self.write:
>           raise IORegistryError._from_write_parts(dest_type, src_type, modifiers)
E           anndata._io.specs.registry.IORegistryError: No method registered for writing <class 'opt.hostedtoolcache.Python.3_12_7.x64.lib.python3_12.site-packages.anndata._core.views.AwkwardArrayView'> into <class 'h5py._hl.group.Group'>
E           Error raised while writing key 'awk_2d_ragged' of <class 'h5py._hl.group.Group'> to /obsm

/opt/hostedtoolcache/Python/3.12.7/x64/lib/python3.12/site-packages/anndata/_io/specs/registry.py:135: IORegistryError

At first glance, the anomaly comes from within anndata. It requires people who are familiar with anndata to dig further.

---

self.write is dict:

{
...
(<class 'zarr.hierarchy.Group'>, <class 'anndata._core.views.AwkwardArrayView'>, frozenset()): <function write_awkward at 0x7fbae30dc680>,  
(<class 'h5py._hl.group.Group'>, <class 'anndata._core.views.AwkwardArrayView'>, frozenset()): <function write_awkward at 0x7fbae30dc5e0>,  
...}

and

dest_type = <class 'h5py._hl.group.Group'>
src_type = <class 'opt.hostedtoolcache.Python.3_12_7.x64.lib.python3_12.site-packages.anndata._core.views.AwkwardArrayView'>
modifiers = frozenset()

---

result:

 (dest_type, src_type, modifiers) in self.write is False

diff:

 <class 'anndata._core.views.AwkwardArrayView'>,

vs

<class 'opt.hostedtoolcache.Python.3_12_7.x64.lib.python3_12.site-packages.anndata._core.views.AwkwardArrayView'>

[dongfangtianyu]

I guess something happened during the import process that resulted in a class receiving two types of import results: relative path and absolute path.
Boolean judgment is used to classify it as having different contents.

From the changelog, it seems to be related to #12592