input is too complicated to be hashable

[yibeichan]

when debugging first-level glm for pydra tutorial we (@yibeichan @djarecka) encountered an error related to hash for a specific node.

Node:

# get glm report
@pydra.mark.task
@pydra.mark.annotate(
    {'model': ty.Any, 'contrasts': str, 'return': {'output_file': str}}
)
def glm_report(model, contrasts):
    output_file = os.path.join(workflow_out_dir, 'glm_report.html')
    report = make_glm_report(model, contrasts)
    report.save_as_html(output_file)
    return output_file

Error:

Task exception was never retrieved
future: <Task finished name='Task-7' coro=<ConcurrentFuturesWorker.exec_as_coro() done, defined at /Users/yibeichen/miniconda3/envs/pydra-tutorial/lib/python3.11/site-packages/pydra/engine/workers.py:173> exception=UnhashableError('Cannot hash object {\'model\': FirstLevelModel(smoothing_fwhm=5.0, subject_label=\'10159\', t_r=2), \'contrasts\': \'StopSuccess - Go\', \'_func\': b\'\\x80\\x05\\x95\\x04\\x04\\x00\\x00\\x00\\x00\\x00\\x00\\x8c\\x17cloudpickle.cloudpickle\\x94\\x8c\\x0e_make_function\\x94\\x93\\x94(h\\x00\\x8c\\r_builtin_type\\x94\\x93\\x94\\x8c\\x08CodeType\\x94\\x85\\x94R\\x94(K\\x02K\\x00K\\x00K\\x04K\\x04K\\x03C\\x9a\\x97\\x00t\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00j\\x01\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\xa0\\x02\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00t\\x06\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00d\\x01\\xa6\\x02\\x00\\x00\\xab\\x02\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00}\\x02t\\t\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00|\\x00|\\x01\\xa6\\x02\\x00\\x00\\xab\\x02\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00}\\x03|\\x03\\xa0\\x05\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00|\\x02\\xa6\\x01\\x00\\x00\\xab\\x01\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x00\\x01\\x00|\\x02S\\x00\\x94N\\x8c\\x0fglm_report.html\\x94\\x86\\x94(\\x8c\\x02os\\x94\\x8c\\x04path\\x94\\x8c\\x04join\\x94\\x8c\\x10workflow_out_dir\\x94\\x8c\\x0fmake_glm_report\\x94\\x8c\\x0csave_as_html\\x94t\\x94(\\x8c\\x05model\\x94\\x8c\\tcontrasts\\x94\\x8c\\x0boutput_file\\x94\\x8c\\x06report\\x94t\\x94\\x8cN/var/folders/wr/x5xt_yqs2cvc_gb3sf147lvc0000gn/T/ipykernel_55279/3311347549.py\\x94\\x8c\\nglm_report\\x94h\\x18K\\x0eCD\\x80\\x00\\xf5\\n\\x00\\x13\\x15\\x94\\\'\\x97,\\x92,\\xd5\\x1f/\\xd01B\\xd1\\x12C\\xd4\\x12C\\x80K\\xdd\\r\\x1c\\x98U\\xa0I\\xd1\\r.\\xd4\\r.\\x80F\\xd8\\x04\\n\\xd7\\x04\\x17\\xd2\\x04\\x17\\x98\\x0b\\xd1\\x04$\\xd4\\x04$\\xd0\\x04$\\xd8\\x0b\\x16\\xd0\\x04\\x16\\x94C\\x00\\x94))t\\x94R\\x94}\\x94(\\x8c\\x0b__package__\\x94N\\x8c\\x08__name__\\x94\\x8c\\x08__main__\\x94uNNNt\\x94R\\x94\\x8c\\x1ccloudpickle.cloudpickle_fast\\x94\\x8c\\x12_function_setstate\\x94\\x93\\x94h"}\\x94}\\x94(h\\x1fh\\x18\\x8c\\x0c__qualname__\\x94h\\x18\\x8c\\x0f__annotations__\\x94}\\x94(h\\x12\\x8c\\x06typing\\x94\\x8c\\x03Any\\x94\\x93\\x94h\\x13\\x8c\\x08builtins\\x94\\x8c\\x03str\\x94\\x93\\x94\\x8c\\x06return\\x94}\\x94h\\x14h0su\\x8c\\x0e__kwdefaults__\\x94N\\x8c\\x0c__defaults__\\x94N\\x8c\\n__module__\\x94h \\x8c\\x07__doc__\\x94N\\x8c\\x0b__closure__\\x94N\\x8c\\x17_cloudpickle_submodules\\x94]\\x94\\x8c\\x0b__globals__\\x94}\\x94(h\\x0bh\\x00\\x8c\\tsubimport\\x94\\x93\\x94h\\x0b\\x85\\x94R\\x94h\\x0e\\x8c\\x07pathlib\\x94\\x8c\\tPosixPath\\x94\\x93\\x94(\\x8c\\x01/\\x94\\x8c\\x03tmp\\x94\\x8c\\x07outputs\\x94\\x8c\\x056_glm\\x94t\\x94R\\x94h\\x0f\\x8c\\x1enilearn.reporting.glm_reporter\\x94h\\x0f\\x93\\x94uu\\x86\\x94\\x86R0.\'}')>
concurrent.futures.process._RemoteTraceback: 
"""
Traceback (most recent call last):
  File "/Users/yibeichen/miniconda3/envs/pydra-tutorial/lib/python3.11/site-packages/pydra/utils/hash.py", line 106, in bytes_repr
    dct = obj.__dict__
          ^^^^^^^^^^^^
AttributeError: 'weakref.ReferenceType' object has no attribute '__dict__'

The above exception was the direct cause of the following exception:

Traceback (most recent call last):
  File "/Users/yibeichen/miniconda3/envs/pydra-tutorial/lib/python3.11/site-packages/pydra/utils/hash.py", line 73, in hash_object
    return hash_single(obj, Cache({}))
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/yibeichen/miniconda3/envs/pydra-tutorial/lib/python3.11/site-packages/pydra/utils/hash.py", line 89, in hash_single
    for chunk in bytes_repr(obj, cache):
  File "/Users/yibeichen/miniconda3/envs/pydra-tutorial/lib/python3.11/site-packages/pydra/utils/hash.py", line 223, in bytes_repr_dict
    yield from bytes_repr_mapping_contents(obj, cache)
  File "/Users/yibeichen/miniconda3/envs/pydra-tutorial/lib/python3.11/site-packages/pydra/utils/hash.py", line 264, in bytes_repr_mapping_contents
    yield bytes(hash_single(mapping[key], cache))
                ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/yibeichen/miniconda3/envs/pydra-tutorial/lib/python3.11/site-packages/pydra/utils/hash.py", line 89, in hash_single
    for chunk in bytes_repr(obj, cache):
  File "/Users/yibeichen/miniconda3/envs/pydra-tutorial/lib/python3.11/site-packages/pydra/utils/hash.py", line 114, in bytes_repr
    yield from bytes_repr_mapping_contents(dct, cache)
  File "/Users/yibeichen/miniconda3/envs/pydra-tutorial/lib/python3.11/site-packages/pydra/utils/hash.py", line 264, in bytes_repr_mapping_contents
    yield bytes(hash_single(mapping[key], cache))

We figured that the error is due to the input model being too complicated; model is the first-level glm model after model.fit, which probably contains too much information.

We solved this problem by integrating glm_report into the model_fitting (l1estimation) node but we want to note this issue here for further reference. We'll probably encounter similar issue in the future.

[tclose]

Another use case for #776