Add deprecation warning for the functional API.

[ejguan]

As a follow-up task, add deprecation warning for the functional API.

Originally posted by @ejguan in #479 (comment)

[ejguan]

Add a flag to indicate deprecation and add a wrapper function to the DataPipe cls constructor in https://github.com/pytorch/pytorch/blob/235f8db3d896e1c4810a2adc3c8a0996292793db/torch/utils/data/datapipes/_decorator.py#L14-L21

[NivekT]

I am thinking we can add a class attribute _deprecated_functions: Set[str] and a check here to see if the method name is deprecated, and raise a warning if it is.

https://github.com/pytorch/pytorch/blob/fe67dff82abf883dfb3623757dda6dab4dd848b7/torch/utils/data/datapipes/datapipe.py#L111-L112

[ejguan]

I am thinking we can add a class attribute _deprecated_functions: Set[str] and a check here to see if the method name is deprecated, and raise a warning if it is.

https://github.com/pytorch/pytorch/blob/fe67dff82abf883dfb3623757dda6dab4dd848b7/torch/utils/data/datapipes/datapipe.py#L111-L112

Doable, we still need to let functional_datapipe know this API is deprecating. Or, we can add a deprecating_functional_datapipe to achieve the similar functionality.

[linminhtoo]

just to clarify, does this mean we should use the DataPipe class directly? the documentations seemed to suggest that functional form is recommended

i.e.

from my_custom_datapipes import MyDataPipe
dp = FileLister(...).filter(...)
dp = MyDataPipe(dp, **my_args)
# instead of
dp = dp.call_my_datapipe(**my_args)

[ejguan]

just to clarify, does this mean we should use the DataPipe class directly? the documentations seemed to suggest that functional form is recommended

Nope. We still recommend to use functional api.

Currently, FSSpecFileOpenerIterDataPipe support two different functional apis: open_file_by_fsspec and open_files_by_fsspec. And, we are deprecating one of these apis open_file_by_fsspec

[ejguan]

Closing this issue as pytorch/pytorch#78970 has landed