continuous integration stuff

[gregjohnso]

No description provided.

[heeler]

TeamCity or Travis or ...? Rory is suggesting Travis barring objects I'll start looking into that.

[heeler]

I'm trying to make travis work and I have the integration mostly configured but when it runs the pytest block it runs into problems with torch

ImportError while importing test module '/home/travis/build/AllenCellModeling/geneselection/geneselection/tests/dataset_test.py'.
Hint: make sure your test modules/packages have valid Python names.
Traceback:
geneselection/tests/dataset_test.py:3: in <module>
    import geneselection.datasets.dataset as gds
geneselection/datasets/__init__.py:1: in <module>
    from .dataset import *
geneselection/datasets/dataset.py:4: in <module>
    from ..utils.dataloader import default_collate
geneselection/utils/__init__.py:1: in <module>
    from .utils import *  # noqa
geneselection/utils/utils.py:2: in <module>
    import torch
../../../miniconda/envs/gsel/lib/python3.6/site-packages/torch/__init__.py:84: in <module>
    from torch._C import *
E   ImportError: dlopen: cannot load any more object with static TLS```

[donovanr]

this issue looks relevant if unhelpful pytorch/pytorch#2083

[donovanr]

could it be an issue with the import * in our __init__.py file? that might be importing not just the functions we defined ourselves but e.g. all the numpy and pandas functions that are exposed via our own imports.

[donovanr]

using an __all__ = ["foo", "bar"] statement as in https://github.com/AllenCellModeling/geneselection/blob/master/geneselection/__init__.py is (I think?) usually considered the "right" way to do it so that you're only exposing an explicit list of the functions you actually want users to see.

[heeler]

I tried the above changes modifying the init.py to use the all = [] mechanism but it still couldn't load and run pytorch.
I eventually dug into the default build environment and realized the default build is Ubuntu 14.04 (trusty). That worried me so I specified the more recent Ubuntu 16.04 (xenial). All of a sudden the tests successfully built and ran all the tests.