[caffe2] Caffe2 relies on futures

[louisabraham]

🐛 Bug

"caffe2/python/core.py", line 9, in <module>
    from past.builtins import basestring
ModuleNotFoundError: No module named 'past'

With latest version pytorch-nightly=1.0.0.dev20181227=py3.7_cuda9.0.176_cudnn7.4.1_0

futures should be added to the dependencies

[soumith]

we dont add all caffe2 dependencies to the wheel dependencies (only the pytorch ones). you can install the caffe2 deps withpip install -r https://raw.githubusercontent.com/pytorch/pytorch/master/requirements.txt

[willprice]

This seems to be an issue with the new tensorboard support

  File "train_cifar.py", line 11, in <module>
    from torch.utils.tensorboard import SummaryWriter
  File "/home/will/.miniconda3/envs/env/lib/python3.7/site-packages/torch/utils/tensorboard/__init__.py", line 6, in <module>
    from .writer import FileWriter, SummaryWriter  # noqa F401
  File "/home/will/.miniconda3/envs/env/lib/python3.7/site-packages/torch/utils/tensorboard/writer.py", line 18, in <module>
    from ._convert_np import make_np
  File "/home/will/.miniconda3/envs/env/lib/python3.7/site-packages/torch/utils/tensorboard/_convert_np.py", line 12, in <module>
    from caffe2.python import workspace
  File "/home/will/.miniconda3/envs/env/lib/python3.7/site-packages/caffe2/python/workspace.py", line 15, in <module>
    from past.builtins import basestring

I'd argue that pytorch should depend on past if its now a transitive dependency.

[soumith]

this seems like a huge oversight. cc: @lanpa @orionr why exactly is from ._convert_np import make_np importing caffe2.python?

[orionr]

The caffe2 import should have been done internal to one of the functions in that file. I'm going to put up a PR to fix this, but it implies that our CI tests don't cover the scenario where people don't have future installed.

[soumith]

@kostmo can you work with @orionr to open up one of the cases to cover this setting (where requirements.txt deps are not installed in the test environment)

[orionr]

cc @sanekmelnikov to work with @kostmo as well.

[geetickachauhan]

@orionr Just following up on this: It does not seem like conda soumith channel has removed the hard dependency (for tensor board) on caffe2 yet because I still get the above issue referenced in this issue. Tensorboard v 1.14 is being installed for me.

[orionr]

@geetickachauhan is this the equivalent of PyTorch nightly? Not sure how you'd still be getting this... can you post a stack trace for the error? Thanks.