Bug: Pipeline.from_descriptor does not work for all steps

[wkoch-cfs]

Hi, I believe i discovered a bug in table-pivot.

To reproduce, building off of table_pivot test:

 from frictionless import Pipeline, Resource, steps

    data = [
        {"region": "east", "gender": "boy", "style": "tee", "units": 12},
        {"region": "east", "gender": "boy", "style": "golf", "units": 14},
        {"region": "east", "gender": "boy", "style": "fancy", "units": 7},
        {"region": "east", "gender": "girl", "style": "tee", "units": 3},
        {"region": "east", "gender": "girl", "style": "golf", "units": 8},
        {"region": "east", "gender": "girl", "style": "fancy", "units": 18},
        {"region": "west", "gender": "boy", "style": "tee", "units": 12},
        {"region": "west", "gender": "boy", "style": "golf", "units": 15},
    ]
    source = Resource(data=data)

    pipeline = Pipeline(
        steps=[
            steps.table_normalize(),
            steps.table_pivot(f1="region", f2="gender", f3="units", aggfun=sum),
        ]
    )
    # This is OK
    target = source.transform(pipeline)
    print(target.to_view())

    json_descriptor = """
    { "steps": 
        [
            {"type": "table-normalize"},
            {
                "type": "table-pivot",
                "f1": "region",
                "f2": "gender",
                "f3": "units",
                "aggfun": "sum"
            }
        ]
    }"""
    source2 = Resource(data=data)
    pipeline2 = Pipeline.from_descriptor(json.loads(json_descriptor))
    # This fails
    target2 = source2.transform(pipeline2)

    # This also fails
    pipeline2 = Pipeline.from_descriptor(json.loads(pipeline.to_json()))

This fails with,

frictionless.exception.FrictionlessException: [step-error] Step is not valid: "table_pivot" raises "[source-error] The data source has not supported or has inconsistent contents: 'str' object is not callable "

In table_pivot aggfun is passed directly to petl, there's no check and conversion for it the parameter is a string. I would have also thought Pipeline.from_descriptor would contain the logic but it fails to export to json as well.

[wkoch-cfs]

I'm a bit confused as to what from_descriptor supports. The documentation implies that you can define declaratively the same steps you can in python. This was the reason I select Frictionless but its turning out that a number of the steps are not cast properly by from_descriptor.

I also found that the resource-transform also doesn't work. Its expecting a list of Steps but the dict is not converted to a step so it gives a the error,

frictionless.exception.FrictionlessException: [step-error] Step is not valid: "resource_transform" raises "[step-error] Step is not valid: "dict" raises "'dict' object has no attribute 'transform_resource'" "

To reproduce the second bug,

from frictionless.steps.field.field_update import field_update

transform_spec = """
        [
            {
                "type": "resource-transform",
                "name": "name_change", 
                "steps": [
                    {
                        "type": "field-update", 
                        "descriptor": {"name": "new_name"}
                    }
                ]
            }
        ]
    """

    transforms = json.loads(transform_spec)
    pipeline = Pipeline.from_descriptor({"steps": transforms})
    assert isinstance(pipeline.steps[0].steps[0], field_update) # fails
   
    assert not isinstance(pipeline.steps[0].steps[0], dict) # true

Any help would be must appreciated. Trying to understand if these are indeed bugs and there is just a lack of test coverage and support. A warning in the docs indicating this feature is in development and may not be fully supported would be really helpful to clarify this all up for new people.