Does dpdata not support custom data type?

[link89]

Recently I have a case that need to create fparam.npy. So I follow the example in this test case https://github.com/deepmodeling/dpdata/blob/master/tests/test_custom_data_type.py, but it turns out nothing is created.

[njzjz]

You also need to follow tests/plugin.

[link89]

I follow the code in tests/plugin and it doesn't work

register_data_type(
    DataType("fparam", np.ndarray, (Axis.NFRAMES, -1), required=False), labeled=True,
)

However, it works correctly with the following code

dpdata.LabeledSystem.register_data_type(
    DataType("fparam", np.ndarray, (Axis.NFRAMES, -1), required=False)
)

My version is dpdata==0.2.17.

[njzjz]

register_data_type must be called before importing LabeledSystem. dpdata.LabeledSystem.register_data_type is used when LabeledSystem has been imported.

[link89]

I see, thank you. IMO it would be hard for a large project to guarantee the import order, perhaps dpdata.LabeledSystem.register_data_type should be the recommended way.